proceedings of the2nd international conference on bio signals, images and instrumentation, (icbsii -...
TRANSCRIPT
Proceedings of the
2Nd International conference on
Bio signals, Images and Instrumentation
Centre for Healthcare Technologies
Department of Biomedical Engineering
SSN College of Engineering
19th
- 21st March 2015
Editorial Board
Chief Editor : Dr. A. Kavitha
Co-editors : D. Kanchana
R. Anandha Sree
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
ii
Contents
From the President's desk
Mrs. Kala Vijayakumar
vi
From the Principal's desk
Dr. S. Salivahanan
vii
Convener's Message
Dr. A. Kavitha
viii
Message from the Conference Secretary
Dr. L. Suganthi
ix
Conference Organizing Committee
x
Organizing Team
xi
Technical Advisory Committee
xi
Student Organizing Team
xii
Keynote Speakers
Mr. Kannan Thiruvengadam
II
Dr. Sudhir Ganeshan
Biomechanics of Spine and Spinal Instrumentation
III
Dr. Syrpailyne Wankhar
Physiological Measurement and issues of Signal Processing
V
Dr. Vijayalakshmi
Advances in imaging methods applied for genetic diagnosis
VII
Dr. Prashant Kumar Srivastava
Mathematical Modeling of HIV Infection: in vivo
IX
Dr. Brajesh K Kunwar
Advances in cardiac interventional Techniques in management of structural heart
disease
XI
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
iii
Research Papers
Sl.
No
Paper ID
Title Page No.
1 ICBSII-BMI013
Characterization of Electrohysterograms using Hurst
Exponents Tharini M, Kamalanand K
1
2 ICBSII-SP001
Finger Movement Recognition using Neural Network Angana Saikia, Nitin Sahai
6
3 ICBSII-BMI011
Development of A System for Analysis of Remote
Auscultation Signal for Management of Chronic Asthma
Cases Bony George, Ajith K.G., George Varkey, Rajesh M
9
4 ICBSII-IP017
FPGA Implementation of Distributed Canny Edge
Detector Algorithm Dhinakaran M, C. M. Sujatha
15
5 ICBSII-IP019
VLSI Implementation for Imaging Based Classification
of Neurodegenerative Diseases- A proposal Neha Gopal. N, T. Christy Bobby
21
6 ICBSII-BMI003
Real Time pH Measurement for Monitoring Effluent
Dialysate in an Artificial Kidney Hemalatha R. J, Shaliya B, Saranya B.
23
7 ICBSII-IP001
Spatial Domain Filters in Preprocessing of Optical
Coherence Tomography Arati Sinha, Jintu Das, K.Venkataraman
27
8 ICBSII-IP004
Impulse Noise Reduction from Mammogram Images
using Novel KT Filters R. Subash Chandra Boss, K. Thangavel
34
9 ICBSII-IP002
Detection and Extraction of Blood Vessel of Retinal
Images in Diabetic Retinopathy using Filters Subhra Pattnaik, Asit Subudhi, Sunita Sarangi, Sukanta Sabut
40
10 ICBSII-IP010
A Novel Earlier Stage Detection Scheme for Glaucoma
Disease using Automated Classification Method M.Kirthikavathi , M.Mahendraselvi, S.Veluchamy
45
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
iv
11 ICBSII-IP016
Feature Extraction using SIFT for Magnetic Resonance
Image Sathiyaseelan. J, Mr. C. John Moses
51
12 ICBSII-BMI001
Simulation of Critical Home Health Monitoring & Drug
Delivery System N. Abhinaya, G. Hari Krishnan, R. J. Hemalatha, G. Umashankar
55
13 ICBSII-BMI012
Design And Development of Diagnostic Tool for Strain
Injuries of Finger Joints using Flex Sensor Balakumaran.V, Bhuvaneshwari.S, Nithyaa.A.N, Premkumar.R
59
14 ICBSII-GE001
Cluster Analysis of Gene Expression Data using
Optimization Techniques: A Survey B. Nithya, R. Rathipriya
63
15 ICBSII-BMI004
Automated Ambu Bag System used in Emergency
Situation Archana J, Mary Helta Diasy
68
16 ICBSII-SP003
Discrimination of SEMG Signals Based on Temporal and
Spectral Approach for Frailty Analysis Vidya K V, E. Priya
72
17 ICBSII-BMI015
Model Identification and Controller Design of a
Perfusion System During an ECMO Support M.Dhinakaran , Dr S.Abraham Lincon, P.Praveen Kumar
80
18 ICBSII-IP009
A Technique to Implement Bimodal Biometrics, for
Preventing Infant Mix-Ups using Raspberry PI S.Sivaranjani, Dr. S.Sumathi
86
19 ICBSII-SP006
Feature Extraction and Comparison of Motor Activities
for Cursor Control Sasweta Pattnaik, Manasa Dash, Sukant Sabut
93
20 ICBSII-BMI014
Non-Invasive Device for Measurement of Glucose and
Haemoglobin in Blood K.M.Nivetha, K.Pavithra, N.Indhujha, D.Arulkumar
98
21 ICBSII-BMI008
Physiotherapy for Rehabilitation using Motion Sensor Sheeba Abraham, Vamsi Krishna
102
22 ICBSII-IP008
Palm Print Based Security System for Lockers Angelena Mathias, S. Caroline
107
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
v
Dedicated to
all Staff and Students of
the Department of
Biomedical Engineering
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
vi
From the President's desk
The numerous achievements of SSN College of Engineering are a direct result of the institution’s strong
backbone of faculty and students. Academic excellence apart, the institution also strives to instill in its students, traits such as critical thinking, work ethic, and the sense of responsibility towards society. Every
aspect of SSN, from the campus to the libraries, is aimed at ensuring the holistic development of its
students. The Research Centre at SSN and its success is a testimony to the emphasis laid on research and development.
Biomedical Engineering (BME) as a discipline is truly multi-disciplinary and offers varied research
avenues. We also observe that in this day and age, it has become imperative to stay abreast with all the latest developments in technology. The importance is two-fold when the technology is linked with
healthcare.
The department collaborates with Centre for Healthcare Technology (CHT), a multi-disciplinary R&D
centre, which is an initiative for bringing together technologists, engineers, doctors, and healthcare
professionals, industry, and government to develop healthcare technologies for the country.
The mission of the BME Department is to pursue excellence in biomedical engineering education,
research, and innovation; creating and imparting knowledge for improving society, human health, and
healthcare. Taking cognizance of this fact, the BME department’s effort to host an International Conference certainly deserves appreciation.
On behalf of the SSN CE Management, I congratulate the BME department and also wish this initiative all success.
Mrs. Kala Vijayakumar
Chief Patron, ICBSII – 2015.
Mrs. Kala Vijayakumar
President, SSN Institutions
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
vii
From the Principal's desk
Dr. S. Salivahanan
Principal, SSN College of Engineering
The SSN Group of Institutions has emerged as one of the best private educational institutions in
the country in terms of academics, infrastructure, and extra-curricular achievements. The main
aim of SSN is to make a positive difference to the society through education.
Biomedical engineering is indeed a holistic discipline and successfully integrates engineering
technology with healthcare. It serves as a home to award-winning faculty, exceptional students,
numerous research centres, and laboratories engaged in an array of interdisciplinary biomedical
activities. The department’s inclination towards research has to be appreciated and the fact that
the enthusiasm is prevalent in students and teachers alike is commendable. The department has
also signed various memorandums with many companies and medical colleges to facilitate
students in the path of their research.
The department collaborates with Centre for Healthcare Technology (CHT), a multi-disciplinary
R&D centre, which is an initiative in bringing together the experts of various fields, to develop
healthcare technologies for mankind.
The Biomedical Engineering Department continues its trail of success by organizing the
International Conference on Bio Signals, Images and Instrumentation-ICBSII in a manner
befitting the stream. I congratulate the entire team at BME for planning this to perfection and
wish them all the very best!
Dr. S. Salivahanan Patron, ICBSII – 2015.
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
viii
Convener's Message
It gives me great pleasure to convene the second International Conference on Bio Signals,
Images, and Instrumentation-ICBSII, 2015. The department is effectively affirming its part in
imparting knowledge to young researchers with great thirst for knowledge and scientific
curiosity by organizing this conference.
The Biomedical Engineering fraternity is one which is oblivious to the boundaries of institutions,
states, countries, etc. This is understandable, given that the field of study is intertwined with
medicine and healthcare. The outcomes of research assume a whole new level of importance and
significance. The discipline also sees growth on a regular basis, an aspect which is exciting while
being challenging. We have strengths in numerous research areas, boast a wealth of research
resources and facilities. Our BME department is known for its highly quantitative approach to
biomedical science, with liberal application of engineering principles and physical sciences.
The department joins hand with Centre for Healthcare Technologies (CHT), a multi-disciplinary
R&D centre of SSN, which focuses on developing healthcare technologies. CHT collaborates
with leading medical institutions and healthcare industries in developing R&D solutions, joint
development of technology products, technology assessment, and evaluation by standardizing
diagnostic procedures, building rural clinics, and developing streamlined health IT systems. CHT
also works to develop human resources in healthcare technology in the country through
internships and projects.
The International Conference on Bio Signals, Images, and Instrumentation (ICBSII), organized
by SSN College of Engineering, is a core biomedical academic event aimed at discussing latest
trends in bio signal and image processing research and bridging the industry institution gap
through debates on bioinstrumentation. The conference provides a platform for academicians,
students, clinicians and researchers to observe, discuss and showcase advancements in
biomedical research. They say teamwork divides the tasks and multiplies the success. I hope that
the success of the conference sets a new benchmark in academia.
Dr. A. Kavitha
Chair Person, ICBSII – 2015.
Dr. A. Kavitha
Head, Department of Bio-Medical Engineering,
SSN College of Engineering
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
ix
Message from the Conference Secretary
The international conference on Bio Signals Images and Instrumentation (ICBSII 2015) is being
organized to share and enhance the knowledge of young biomedical engineers, researchers and
faculty of various institutions in the field of biomedical signal, image and instrumentation. This
conference aims to stimulate the students for transforming their theoretical knowledge into
biomedical products which will help the society. This conference creates a bridge between the
doctors and engineers. Since doctors are handling the biomedical instruments, signal and image
processing systems which is proposed and validated by engineers, this conference brings out real
time research problems addressed by the doctors in their fields of specialization and young
engineers can work towards those research problem. The conference also creates an environment
for the scientists, researchers and students from all over the world to share their ideas,
experiences and results of their scientific research work and shows the way for research
collaboration.
I take immense pleasure to express my sincere and deep sense of gratitude to the Management of
SSN College of Engineering, Ms. Kala Vijayakumar, President, SSN Institutions and Dr.
Salivahanan, Principal, SSN College of Engineering for granting a wonderful opportunity to
organize this conference and for helping us to promote our research. I would like to give my
special thanks to delegates Mr. Kannan Thiruvengadam, Director, Platinum,Aurea Inc., Austin,
TX, USA, for inaugurating this conference, Dr. Sudhir ganeshan, Gangaram Hospital, Delhi, Dr.
Syrpailyne Wankhar, CMC, Vellore, Dr. Brajesh Kumar Kunwar, care hospital, Surat,
Dr.Prashant Srivastava, IIT Patna, Dr. Vijayalakshmi, Sri Ramachandra University,Chennai, for
their valuable speech to motivate us towards the quality research. I would like to thank all the
faculty, staff and students of Biomedical department for making this function grand success.
Dr.L.Suganthi
Secretary, ICBSII – 2015.
Dr. L. Suganthi,
Associate Professor, Department of Bio-Medical Engineering,
SSN College of Engineering
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
x
Conference Organizing Committee
Chief Patron:
Mrs. Kala Vijayakumar,
President
SSN Institutions
Patron:
Dr. S. Salivahanan,
Principal,
SSN College of Engineering.
Chairperson:
Dr. A. Kavitha,
Head, Department of Biomedical Engineering,
SSN College of Engineering, Chennai.
Secretary:
Dr. L. Suganthi
Ms. D. Kanchana
Treasurers:
Dr. R. Subashini
Ms. Delpha J
Mr. R.Yuvaraj
Organizing Team:
Dr. S. Pravin Kumar
Dr. V. Mahesh
Dr. Guruprakash subbiahdoss
Dr. Mallika Jainu
Mr. R. Sivaramakrishnan
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
xi
Mrs. B. Geethanjali
Dr. Sachin Gaurishankar Sarate
Mrs. M. Dhanalakshmi
Ms. R. Nithya
Technical Advisory Committee
Dr. Neeradha Chandramohan, NIEPMD
Dr. Sudhir Ganesh, Ganga Ram Hospital, Delhi
Dr. S. Ramakirishna, IITM
Dr. S. Rajendran, Sri Ramachandra University
Dr. Ganesh Venkatraman, Sri Ramachandra University
Student Organizing Team
Food Committee
A. Siva., Final Year
Nagasai., Final Year
Vijaya Balaji S., Final Year
Hall Arrangement Committee
Muthu Vijay., Third Year
Deepak., Third Year
Muthumeenakshi ., Third Year
Transportation Committee
Hemanath N., Final Year
Prasanth P., Final Year
Cultural Committee
Hemavardhini S., Final Year
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
xii
Aishwariya., Final Year
Photography Committee
Raj Kumar., Final Year
Saravana Prakash., Third Year
Registration Committee
Siva A., Final Year
Kulddep Surana., Final Year
Vardhini., Final Year
Guest House/Accommodation Committee
R. Naren., Final Year
V. Naren., Final Year
Conference Proceedings Committee
Angel Jenifer.P, M.E– First Year
Bhuvaneshwari.B, M.E– First Year
Deepthi.K.S, M.E– First Year
Diwakar.M, M.E– First Year
Guhan Seshadri.N.P, M.E– First Year
Meenachi.P, M.E– First Year
Murali.S, M.E– First Year
Sriranjani.S, M.E– First Year
Vaishali.R, Final Year
Yuvadharshini.I, Final Year
Sarah Rajitha Thilagam.V, Final Year
Prasanna Bharati.R, Final Year
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
I
2nd
International Conference on Bio Signals,
Images and Instrumentation
ICBSII - 2015
Keynote Talks
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
II
Chief Guest
Mr. Kannan Thiruvengadam,
Director, Platinum
Aurea Inc., Austin, TX, USA
Kannan has been a technology person right from school. He could speak the language of the
machine, and took great joy in programming in both high-level and low-level languages. He studied
algorithms and data structures, and enjoyed choosing the right method and structure depending on
the nature and size of the problem. He’s won several awards in college. He went to NIT Trichy.
Alongside, he has gone out and taught village school children languages he knew, through the
Computer Society of India and for raising funds for college events.
He rejected his campus interview offers and stayed back in his college, working for the computer
department developing learning utilities for his juniors. At that time, even a network file transfer tool
was a novelty. After a year of such service, from the hot and chaotic climate of Trichy, he went to
the cold and calm of Western Canada, close to the Rockies. He completed his Masters in Computing
Science at the University of Alberta, and moved to the United States for research and work in
multimedia.
He has held a variety of industrial positions since then, covering a myriad of topics, and in various
industries, including health-care, travel and entertainment, telecommunications, retail, insurance
(Swiss Reinsurance) and finance (Bank of New York, Bank of America). His work took him to
many parts of the world, including Europe and East Asia, which means he has worked with people
of various cultural and linguistic backgrounds, and on problems that are unique to their contexts. His
technical background includes business process management, middleware, communication security
and operation of Production systems. Lately, Kannan has taken an interest in the application of
technology in what he considers the most critical issue of our times: climate change and possible
applications of biotechnology there.
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
III
Speaker Profile:
Dr Sudhir Ganeshan is a Consultant Orthopaedic and Spine Surgeon registered with
Indian Medical Council and Tamil Nadu Medical Council. His academic accolades include a
FNB (Fellowship in National Board) in Spine Surgery from Sir Ganga Ram Hospital, New Delhi
The only certified course for Spine Surgery in the country under National Board of
Examinations, Ministry of Health and Family Welfare, MNAMS (Membership of National
Academy of Medical Sciences) in 2012, DNB in Orthopedics 2008 to 2011 - Ganga Hospital,
Coimbatore under National Board of Examinations and he is currently pursuing a 2 year
Fellowship in EULAR (European League against Rheumatism). He was also awarded with the
Best Outgoing Spine Fellow of Ortho Spine Department, Sir Gangaram Hospital. He is also an
integral member of the Delhi Spine Society and Coimbatore Orthopedic Society and has
performed and assisted almost all types Spine surgeries for trauma, deformity correction,
degenerative disorders, inflammatory disorders, tumours, infections, including minimally
invasive surgeries and Endoscopic spine surgeries.
Dr. Sudhir Ganeshan
Consultant Orthopaedic and Spine
Surgeon
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
IV
Biomechanics of Spine and Spinal Instrumentation
-Dr. Sudhir Ganeshan.
Abstract:
The spine is a complex mechanical structure complete with levers (vertebrae), pivots (facets
and discs), passive restraints (ligaments), and actuators (muscles). Instrumentation of spine for
various pathologies including deformity corrections requires a thorough understanding of the
biomechanics of Spine. Moreover, designing implants and instrumentation for spine fixations
warrants an adequate insight into the biomechanics of normal and abnormal states. The field
of modern biomechanics has deep historical roots from the ancient Egyptians, who
documented the earliest accounts of spinal injury. Building on this foundation, the modern
explosion of spinal instrumentation introduced the concept of internal fixation for spinal
stabilization, further advancing the understanding of the mechanics of musculoskeletal
motion. The spinal instrumentation has revolutionized since the advent of pedicle screw based
system. Various techniques and instrumentation systems like fenestrated screw, minimally
invasive instrumentation are on the rise with improvisation of the biomechanical principles of
implants. In spite of such advances in this field, all the techniques have some cons which
eventually results in poor outcome for the patients. This may be due to poor understanding of
the normal biomechanics of the spine or of the implants or of both. Clinical biomechanics
requires the assessment of 3 key questions: 1) how do the components of the implant connect
together, 2) how does the implant connect to the spine, and 3) how does the construct function
biomechanically. The ability to apply these basic biomechanical principles in the clinical
arena is not only important for the Spine surgeons for decision making in treating the spinal
disorders but also for the Biomedical engineers for development of new instrumentation
systems which will overcome the existing disadvantages ultimately resulting in a better
clinical outcome for the patients.
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
V
Speaker Profile:
Dr. Syrpailyne Wankhar has received the Ph.D. degree in Bioengineering from CMC Vellore,
M.Tech degree in Biomedical Engineering from IIT, Bombay and B.E. degree in Electrical
Engineering from MMMEC, Gorakhpur. Her professional experience include working for a year
in the department of Physiology, AIIMS, New Delhi, and research assistant and teaching
assistant at CMC, Vellore. She has been one the faculty of CMC, Vellore since 2012, currently
as a Lecturer in the Department of Bioengineering. In the last 5 years she has collaborated with
the departments of Orthopedics, Surgery, Physiology, Psychiatry in CMC Vellore for research
projects like Assessment of knee function and proprioception following anterior cruciate
ligament repair, Development of an in-house NIR imaging for lymph node detection in breast
cancer, Reflex modulation before and after exercise. In addition, she holds a patent for a
magnetic stimulator design for high frequency stimulation. Some of her works have been
published in the Journal of Bone and Joint Surgery, Journal of Medical Devices.
Dr. Syrpailyne Wankhar,
Department of Bioengineering,
CMC Vellore.
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
VI
Recording and Analysis of Biopotential Signals
-Dr. Syrpailyne Wankhar.
Abstract:
All measurement system requires prior information about the quantity being measured in order to
be able to distinguish signal of interest from any unwanted signal. It is also important that the
measuring system does not alter the system being measured. Similarly, for physiological
measurement we need to understand the physiological system under investigation so that the
signal of interest can be recorded accurately and any interfering signal can be detected and
removed. The purpose of a physiological measurement is to understand the underlying system.
In this lecture, I will discuss some aspects of the physiological measurement and issues of signal
processing. As an engineer more than a mere theoretical explanation, a demonstration of the
physiological measurement will be more meaningful in understanding the underlying system. I
will present such a demonstration in this lecture. Discussion of systems characteristics, dynamic
response, physiological interpretation will be based on the demonstration. For example, the
temporal and spatial characteristics of ECG can provide useful insight about the location or
origin of a problem in the cardiovascular system. This aspects will also be brought out in the
demonstration and subsequent discussion during the lecture.
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
VII
Speaker Profile:
Dr.Vijayalakshmi has completed Ph.D at Department of Human Genetics, Sri Ramachandra
University. She has participated in over 20 workshops including International Conference on
Radiation Biology, Nanotechnology, Imaging and Stem Cell Research in Radiation Oncology
(ICRB-NISRRO), Prenatal & Postnatal Diagnosis of Genetic Disorders using Molecular
methods(conducted by AIIMS, New Delhi).She has published articles in International Journal of
Human Genetics.In addition,she has worked as an investigator in projects funded by DRDO-
INMAS.She has received Dr Todla Ekambaram Prize for General Proficiency in Botany and
award for having rendered Ten years of Continuous Service at the Sri Ramachandra University.
Dr. Vijayalakshmi J,
Assistant Professor,
Department of Human Genetics,
Sri Ramachandra University,
Chennai
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
VIII
Advances in imaging methods applied for genetic diagnosis
-Dr. Vijayalakshmi J
Abstract:
Cytogenetic analysis is the study involving structure and function of chromosomes which is an
important tool employed in the diagnosis of genetic disorders. The chromosomes are prepared
either from chorionic villi, amniotic fluid, cord blood, bone marrow or peripheral blood
lymphocytes in patients presenting with abnormal clinical findings. The preparation of
chromosomes involves cell culture, metaphase arrest, hypotonic treatment, cell fixation, slide
preparation and use of banding methods. The analysis is done by taking photomicrographs of the
metaphase spreads, from which each chromosome is cut, paired and the arrangement of the
chromosomes (karyotype) is prepared manually. This whole procedure is time consuming and
laborious. The increased need of the cytogenetic laboratories and availability of improved
computer capabilities for image processing, the Image Analyzing System, with appropriate
software, are being used for cytogenetic analysis. The system includes a high resolution research
microscope with or without an automatic metaphase scanning stage, charged couple device
(CCD) camera, computer with special software for image capturing, chromosome counting, and
automatic karyotyping. At its present stage of development, the system will perform a complete
analysis of well spread, homogeneously stained metaphase chromosomes. The system
incorporates a metaphase finder to detect cells of good quality. Chromosomes in a selected cell
are segmented by their grey level using a locally determined threshold. An axis and a
centromere are automatically assigned to each chromosome. Software is used to classify the
chromosomes into seven groups based on number, size and position of the centromere. A grey-
level karyogram can be displayed for interpretation with an analysis time is about two minutes
per cell. This system has made storing and retrieval of data very easy. Therefore, the output
exquisite analytical report is conducted with the soft ware may increase the rapidity in reporting
genetic test.
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
IX
Speaker Profile:
Dr.Prashant Kumar Srivastava has worked as Assiatant Professor at BITS, Pilani and Post
Doctoral Fellow at IIT Kanpur. He has presented papers in Mathematical Modeling in Ecology
and Epidemiology, Applications of Differential Equations in Biology in various journals. He has
received fellowships such as the JRF,SRF and NBHM Post Doctoral Fellowship that is given by
Department of Atomic Energy. He has also received awards such as IMS Prize for Best Paper in
Biomathematics by Indian Mathematical Society , Best Tutor award from Director IIT Kanpur
and Best Teacher Awards of IIT Patna .In addition, he is a member of societies such as Society
for Mathematical Biology (SMB), Indian Mathematical Society (IMS) and Indian Academy for
Mathematical Modelling and Simulation (IAMMS).
Dr. Prashant Kumar Srivastava
Asst. Professor,
IIT-Kanpur
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
X
Mathematical Modeling of HIV Infection: in vivo
-Dr. Prashant Kumar Srivastava
Abstract:
AIDS, which begins with deterioration of the immune system, is caused by infection with
Human Immunodeficiency Virus (HIV). It is one of the most fatal diseases and is causing huge
loss to social and economical development of many countries, especially in developing countries.
HIV primarily attacks CD4+ T cells that cause the destruction and depletion of these cells
leading to compromised immune system and inability to fight with opportunistic infections. In
this talk we shall give an overview of mathematical modeling of the HIV and immune system
interaction. We introduced a simple mathematical model to understand the dynamics and
development of HIV in vivo during primary phase of infection. Further we develop and improve
this model to include various biological phenomena. We shall also look for models with therapy
and multiple viral strains. With help of tools of differential equation we shall analyse the models
to get meaningful information.
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
XI
Speaker Profile:
Dr.Brajesh Kumar Kunwar has completed his MBBS from Government College, Thrissur, his
MD from MLN Medical College, Allahabad and his DM in Cardiology from CMC, Vellore. He
has received numerous awards like SN Gupta Award, BN Braun Award, Young Investigator
Award at the 6th International Congress Of Cardiovascular Disease. He has presented 7 papers in
International journals and 1 paper in a national journal. He holds the distinction of being the first
doctor to do EVAR (Endovascular Aortic Repair) and single or dual chamber AICD in coastal
Andhra. In addition he has performed more than 7000 angiograms, greater than 1000 PCI,
greater than 50 renal PTA, greater than 20 ASD device closures, greater than 50 PDA device
closure, greater than 10000 TTE, greater than 350 TEE, greater than 10 AICD.
Dr. Brajesh Kumar Kunwar:
Consultant in Cardiology,
Care hospital, Surat
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
XII
Advances in cardiac interventional Techniques in
management of structural heart disease
-Dr. Brajesh Kumar Kunwar
Abstract:
The presentation briefly explains about CAD and PAVM. Coronary artery disease (CAD) causes
major death around the world. The symptoms of CAD are chest pain due to narrowing of arteries
in the heart which leads to limitation of blood flow. The corrective measure for the same is
explained with case studies. Pulmonary arteriovenous malformation (PAVM) is the abnormal
communication between pulmonary arties and veins which leads to dyspnea. Closure of the
PAVM with an Amplatzer-type duct occluder was hampered by inability to advance the device
delivery sheath into the PAVM due to vessel tortuosity and inadequate guide wire support. Atrial
septal puncture was performed and a femoral arteriovenous guide wire loop through the right
pulmonary artery, PAVM, and left atrium was created. Traction on both ends of the guide wire
loop allowed advancement of the device delivery sheath into the PAVM and successful
completion of the procedure. Transseptal guide wire stabilization can be a valuable option during
device closure of large PAVMs when advancement, stability, or kinking of the device delivery
sheath is an issue.
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
XIII
2nd
International Conference on Bio Signals,
Images and Instrumentation
ICBSII - 2015
Research Papers
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
1
CHARACTERIZATION OF ELECTROHYSTEROGRAMS USING HURST
EXPONENTS Tharini M*, Kamalanand K
Department of Instrumentation Engineering, Anna University, MIT Campus, Chennai, India
ABSTRACT Preterm delivery or birth is described as the delivery of babies who are born alive before 37 weeks of gestation. Term delivery is
the delivery after 37 weeks and before 42 weeks of gestation. Electrohysterograms are the signals representing the uterine muscle activity
that are used to predict the preterm delivery. In this work, (N=80) electrohysterograms consisting of term and preterm signals have been acquired and the Hurst exponent of each signal has been estimated using four different approaches namely Absolute moment method, Aggregate variance method, Higuchi method and R/S method. Results demonstrate that the estimated Hurst exponents of preterm signals are more persistent than term signals. In this paper, objectives of the study, methodology and significant observations are presented.
Keywords: Electrohysterograms, Hurst exponents, Chaos, Term/Preterm delivery
I. INTRODUCTION
Preterm birth, which is also known as premature birth or delivery, is described by the World Health Organisation (WHO) as the delivery of babies who are born, alive, before 37 weeks of gestation. In contrast, term births are the live births after 37 weeks, and before 42 weeks of gestation [1]. Electrohysterography (EHG) is a noninvasive technique for monitoring uterine electrical activity. The Electrohysterogram (EHG) is the recording of uterine electrical activity from the abdominal surface [2].
Each measurement is composed of three channels, recorded from 4 electrodes. The first electrode (E1) is placed 3.5 cm to the left and 3.5 cm above the navel. The second electrode (E2) is placed 3.5 cm to the right and 3.5 cm above the navel. The third electrode (E3) is placed 3.5 cm to the right and 3.5 cm below the navel. The fourth electrode (E4) is placed 3.5 cm to the left and 3.5 cm below the navel. The differences in the electrical potentials of the electrodes are recorded, producing three channels: S1=E2–E1 (first channel); S2=E2–E3 (second channel); S3=E4–E3 (third channel) [3]. The electrical activity during pre-term labor is significantly different from the activity of term labor [4].
Preterm birth has adverse effects on the new born, such as increased risk of death and health defects. These include impairments to hearing, vision, the lungs, the cardiovascular system and non-communicable diseases. Also up to 40% of survivors of extreme preterm delivery can develop chronic lung disease. In other cases, survivors suffer with neuro-developmental or behavioural defects, including cerebral palsy, motor, learning and cognitive impairments [1].
Nonlinear dynamics, study of systems governed by equations in which a small change in one variable can induce a large systematic change; the discipline is more popularly known as chaos. Unlike a linear system, in which a small change in one variable produces a small and easily quantifiable systematic change, a nonlinear system exhibits a sensitive dependence on initial conditions: small or virtually unmeasurable differences in initial conditions can
lead to wildly differing outcomes. Hurst exponent is used as a measure of the long-term memory of a time series, a statistical methodology for distinguishing random from non-random systems and to identify the persistence of trends [5]. The Hurst Exponent is a dimensionless estimator for the self-similarity of a time series. Initially it was defined by Harold Edwin Hurst to develop a law for regularities of the Nile water level. Now it has applications in medicine and finance. Meaningful values are in the range of 0 to 1 [6].
The objective of this work is to analyze the persistent behaviour of term and preterm electrohysterograms using Hurst exponent.
II. METHODOLOGY
Prerecorded term and preterm EHG signals (N=80) were
obtained from the Physiobank ATM
(http://www.physionet.org/cgi-bin/atm/ATM). Forty two
term and thirty eight preterm EHG signals were used for this study. The first sixty seconds of the adopted signals
was utilized and the Hurst exponents were calculated.
A.Hurst Exponent
The Hurst exponent (H) is used as a measure of the long-term memory of a time series, a statistical methodology for distinguishing random from non-random systems and to identify the persistence of trends. A Hurst exponent value between 0 and 0.5 is indicative of anti-persistent behavior and the closer the value is to 0, the stronger is the tendency for the time series to revert to its long-term means value. A Hurst exponent value between 0.5 and 1.0 indicates persistent behavior; the larger the H value the stronger is the trend. A Hurst exponent close to 0.5 indicates a Brownian time series. In a Brownian time series, (also known as a random walk) there is no correlation between the previous or present observations and a future observation; being higher or lower than the current observation is equally likely [5]. The Hurst Exponent H has the relationship with the fractal dimension D as D=2-H. Initially defined by Harold Edwin Hurst to develop a law for regularities of the Nile water level, it now
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
2
has applications in medicine and finance. Meaningful values are in the range of 0 to 1 [6].
Definition of the Hurst exponent developed by Harold Hurst is as follows:
R(n) is defined on a time series Xi, i=1, 2, . . . , n as
follows:
R(n) = max (Xi, i = 1, 2, . . . , n) – min (Xi, i = 1, 2, . . . ,
n) . S(n) is the standard deviation and C an arbitrary
constant [6].
B.Adopted methods to estimate Hurst Exponent
In this work, four methods are used to estimate the Hurst exponents which are explained as follows:
i. Aggregated Variance Method:
The original time series is divided into blocks of size m. Then the sample variance within each block is computed. The slope beta=2*H-2 from the least square fit of the logarithm of the sample variances versus the logarithm of the block sizes provides an estimate for the Hurst exponent H [7].
ii. Absolute Value/Moment Method:
The Hurst exponent is estimated from the moments moment=M of absolute values of an aggregated time series process. The first moment M=1 coincides with the absolute value method, and the second moment M=2 with the aggregated variance method. Again, the slope beta=M*(H-1) of the regression line of the logarithm of the statistic versus the logarithm of the block sizes provides an estimate for the Hurst exponent H [7].
iii. Higuchi Method:
This method is very similar to the absolute value method. Instead of blocks a sliding window is used to compute the aggregated series. The function involves the calculation the calculation of the length of a path and, in principle, finding its fractal Dimension D. The slope D=2-H from the least square fit of the logarithm of the expected path lengths versus the logarithm of the block (window) sizes provides an estimate for the Hurst exponent H [7].
iv. Rescaled Range(R/S) Method:
The rescaled range is calculated from dividing the range of the values exhibited in a portion of the time series by the standard deviation of the values over the same portion of the time series. The increase of the rescaled range can be characterized by making a plot of the logarithm of R/S versus the logarithm of n. The slope of this line gives the Hurst exponent, H. If the time series is generated by a random walk (or a Brownian motion process) it has the value of H =1/2 [8].
III. RESULTS AND DISCUSSION
(1)
0 0.5 1 1.5 2 2.5-0.5
0
0.5
1
1.5
2
2.5
Log of Aggregate Level
Log o
f A
bsolu
te M
om
ent
Absolute Moment Method
(a)
0 0.5 1 1.5 2 2.52
2.5
3
3.5
4
4.5
log10(Aggregate Level)
log1
0(V
aria
nce)
Time Variance Method
(b)
0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 5.53
4
5
6
7
8
9
10
11
12
Log10(Aggregate Level)
Log10(C
urv
e L
ength
)
Higuchi Method
(c)
0 0.5 1 1.5 2 2.5 3 3.5 40
0.5
1
1.5
2
2.5
3
log10(blocks of size m)
log10(R
/S)
R/S Method
slope 1/2
slope 1
(d)
Fig. 1. Typical plots in the estimation of the Hurst exponent of
term EHGs using: (a) Absolute Moment method (b)
Aggregate/Time Variance method, (c) Higuchi method and (d)
R/S method
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
3
The Hurst exponents of term and preterm electrohysterograms were estimated using four methods in this work.
Fig. 1 and 2 shows typical plots obtained in estimating the Hurst exponents of term and preterm electrohysterograms respectively. Fig. 1(a) and 2(a) shows the typical plot of the variation of Absolute moment as a function of aggregate level, using Absolute Moment method. Fig. 1(b) and 2(b) shows the typical plot showing the variation of variance as a function of aggregate level, using Aggregate Variance method.
Fig. 1(c) and 2(c) shows the typical plot showing the variation of log of curve length as a function of log of aggregate level using Higuchi method. Fig. 1(d) and 2(d) shows the typical plot showing the variation of log of Rescaled range(R/S) as a function of log of block size using R/S method.
Fig. 3 (a), (b), (c) and (d) show the variation of Mean values of Hurst exponents of term and preterm Electrohysterograms obtained using four different methods such as Absolute value method, Aggregate variance method, Higuchi method and R/S method respectively. From Fig. 3, it has been observed that, in all the four methods of determining Hurst exponent, their mean values remained to be close to 0.5 or greater than 0.5 indicating the persistent nature of the Electrohysterograms.
From Fig. 3(a) and Fig. 3(b), it is seen that the Mean values of Hurst exponents of preterm Electrohysterograms are more persistent than that of term Electrohysterograms obtained from all the three channels. From Fig. 3(c) and Fig. 3(d), it is observed that there is no notable variation in the mean values of the Hurst exponents between term and preterm Electrohysterograms indicating equally persistent nature.
Fig. 4. (a), (b) and (c) shows the variation of Hurst exponent as a function of Mean Muscle Force of term and preterm electrohysterograms from channel-1, channel-2 and channel-3 respectively.
From Table 1, it was found that, in all the channels, a positive correlation exists between the Hurst exponent and the mean muscle force for term EHG signals whereas it is negative for preterm EHG signals. However, the correlation between the Hurst exponent and the mean muscle force is found to be very poor (< 0.5).
TABLE. 1. CORRELATION BETWEEN HURST EXPONENT
AND MEAN MUSCLE FORCE
Electrohysterogram
(EHG)
Correlation value
Channel-1 Channel-2 Channel-3
Term EHG 0.2162 0.3226 0.3464
Preterm EHG -0.1031 -0.1977 -0.1749
0 0.5 1 1.5 2 2.50.5
1
1.5
2
2.5
3
Log of Aggregate Level
Log
of A
bsol
ute
Mom
ent
Absolute Moment Method
(a)
(a)
0 0.5 1 1.5 2 2.54
4.5
5
5.5
6
6.5
log10(Aggregate Level)
log1
0(V
aria
nce)
Time Variance Method
(b)
0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 5.55
6
7
8
9
10
11
12
13
Log10(Aggregate Level)
Log10(C
urv
e L
ength
)
Higuchi Method
(c)
0 0.5 1 1.5 2 2.5 3 3.5 40
0.5
1
1.5
2
2.5
3
log10(blocks of size m)
log10(R
/S)
R/S Method
slope 1/2
slope 1
(d)
(d) Fig. 2. Typical plots in the estimation of the Hurst exponent of preterm
EHGs using: (a) Absolute Moment method (b) Aggregate/Time Variance
method, (c) Higuchi method and (d) R/S method
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
4
1 2 30
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
CHANNEL NUMBER
ME
AN
OF
HU
RS
T E
XP
ON
EN
T
(AB
SO
LU
TE
VA
LU
E M
ET
HO
D)
TERM
PRETERM
(a)
1 2 30
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
CHANNEL NUMBER
ME
AN
OF
HU
RS
T E
XP
ON
EN
T
(AG
GR
EG
AT
E V
AR
IAN
CE
ME
TH
OD
)
TERM
PRETERM
(b)
1 2 30
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
CHANNEL NUMBER
ME
AN
OF
HU
RS
T E
XP
ON
EN
T
(HIG
UC
HI M
ET
HO
D)
TERM
PRETERM
(c)
1 2 30
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
CHANNEL NUMBER
ME
AN
OF
HU
RS
T E
XP
ON
EN
T
(R/S
ME
TH
OD
)
TERM
PRETERM
(d)
Fig. 3. Mean values of Hurst exponents obtained by (a) Absolute Value method (b) Aggregate Variance method (c) Higuchi
method (d) R/S method
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
5
CONCLUSION
Electrohysterograms are the signals representing the uterine muscle activity and are useful in predicting the delivery as Term or Preterm delivery [2]. Preterm delivery is the case where the babies are born alive before 37 weeks of gestation while Term delivery is the delivery after 37 weeks and before 42 weeks of gestation. The Preterm births suffer from health issues such as neuro-developmental or behavioural defects. Therefore, it is necessary to predict the delivery so that, preventive measures in terms of medication and also treatment can be done at an earlier stage for the health issues, in case of predicted Preterm delivery [1].
In this work, the Hurst exponent have been analysed to characterize the persistent nature of term / preterm electrohysterograms. The results demonstrate that the Hurst exponents of the term and preterm electrohysterograms lie in the range of 0.5 to 1. This shows that the electrohysterograms are persistent in nature. Further, it is found that the preterm electrohysterograms are more persistent than the term electrohysterograms. This work is clinically relevant since characterization of the persistent nature of term and preterm electrohysterograms is useful for acquiring diagnostic information.
References
[1]. G. Fergus P, Cheung P, Hussain A, Al-Jumeily D, Dobbins C, et al., “Prediction of Preterm Deliveries from EHG Signals
Using Machine Learning”, (2013), PLoSONE 8(10): e77154. doi:10.1371/journal.pone.0077154
[2]. Yiyao Ye-Lin, Javier Garcia-Casado, Gema Prats-Boluda, José
Alberola-Rubio, and Alfredo Perales, “Automatic Identification of Motion Artifacts in EHG Recording for Robust Analysis of
Uterine Contractions”, Computational and Mathematical Methods in Medicine Volume, Article ID 470786, 11 pages,
(2014)
[3]. http://www.physionet.org/pn6/tpehgdb/
[4]. Sindhiya Arora and Girisha Garg, “A Novel Scheme to Classify EHG Signal for Term and Pre-Term Pregnancy Analysis”,
(2012), International Journal of Computer Applications (0975 – 8887) Volume 51– No.18
[5]. Subir Mansukhani, “Predictability of Time Series A statistical measure used to classify time series and infer the level of
difficulty in predicting and choosing an appropriate model for the series at hand”, (2012), The Hurst Exponent: Predictability
of Time Series
[6]. Roman Racine, “Estimating the Hurst Exponent”, April 14, 2011, MOSAIC Group, Prof. Ivo F. Sbalzarini, ETH Zurich
Bachelor Thesis
[7]. http://help.rmetrics.org/fArma/LrdModelling.html
[8]. Hurst, H. E., "Long term storage capacity of reservoirs", (1951), Trans. Am. Soc. Eng., 116: 770–799.
-150 -100 -50 0 50 100 150 200 250 300 3500.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Mean Muscle Force(Channel-1)
Hu
rst E
xp
on
en
t(C
ha
nn
el-
1)
TERM
PRETERM
(a)
-30 -20 -10 0 10 20 30 40 50 60-0.1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
Mean Muscle Force (Channel-2)
Hu
rst E
xp
on
en
t (C
ha
nn
el-
2 )
TERM
PRETERM
(b)
-30 -20 -10 0 10 20 30 40 50 60 70-0.1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
Mean Muscle Force (Channel-3)
Hu
rst E
xpo
ne
nt (
Ch
an
ne
l-3
)
TERM
PRETERM
(c)
Fig. 4. Hurst exponent as a function of Mean muscle force from
(a)Channel-1, (b)Channel-2 and (c)Channel-3
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
6
FINGER MOVEMENT RECOGNITION USING
NEURAL NETWORK
Angana Saikia1, Nitin Sahai
1
1 Department of Biomedical Engineering, North East Hill University, Shillong-793022, India
ABSTRACT Biosignals based natural gesture recognition is one of the most active trust in the area of rehabilitation robotics. The work reported
in this paper presents the recognition of index finger movements based on forearm EMG signals. The focus is on derivation of a feature vector for higher recognition rate based on lower number of EMG channels. Feature vector obtained through combination of three time and frequency domain features(modified mean frequency, waveform length and RMS value) resulted into highest recognition rate of
94%.The recognition is through a back propagation neural network trained using L-M and C-G training algorithm. The L-M training algorithm gave the highest recognition rate. The achieved recognition rate is comparable to those reported in literature and better in terms of lower number of channel.
Keywords: EMG, Neural Network, Feature Extraction, Classification, Feature Selection, Feature Combination.
I. INTRODUCTION
The surface electromyogram (EMG) provides a non
invasive method of measuring muscle activity, and has been
extensively investigated as a means of controlling prosthetic
devices. Fig 1 shows the raw EMG signal. Human hand is
an in-dispensable organ of human structure, function, and expression. It is capable of producing complex and
expressive articulations. Physiological variations in the
membranes of muscle fibre cause myoelectric signals. A
technique used for reading and analyzing the myoelectric
signal is called Electromyography (EMG). The EMG
signals are acquired through dual channel Ag/AgCl surface
electrodes. The choice of muscles is based on the type of
movement required. The EMG signals were collected from
ten healthy subjects. EMG pattern recognition is an
advanced, intelligent signal processing technology and has
been proposed as a potential method for reliable user intent
classification [1] [2]. Beyond signal magnitude, a typical pattern recognition algorithm extracts a set of features that
characterize the acquired EMG signals and then classifies
the user's intended movement for external device control.
The benefit of pattern recognition algorithms are that they
can increase the neural information extracted from EMG
signals using a small number of monitored muscles and
allow intuitive control of external devices. Previous studies
have evaluated the ability of various EMG features and
classifiers to recognize user intent [1] [3] [4] [5]. These
studies were mainly done on able-bodied subjects or on
subjects with transradial amputations. The results demonstrated over 90% classification accuracy for either
online or online testing. The comparison of classification
accuracies resulting from utilization of different types of
classifiers and EMG features demonstrated that the type of
classifier used does not significantly affect the classification
performance, while the choice of features has a significant
impact on classification performance [4] [6].
II. MATERIALS AND METHODS
A. EMG signal acquisition and preprocessing
The EMG signals were collected from ten healthy
subjects of age group between 21 to 30. Their right arm is
been cleaned with a cleanser so that the dust particles are
removed. The subject is then asked to sit in a comfortable
state. The EMG signals are acquired through dual channel Ag/AgCl surface electrodes. We have used AD Instruments
PowerLab 4/25T for signal acquisition. Table (1) shows the
specification of the settings of the EMG unit during
acquisition phase. And Table (2) shows the Muscle
selection and electrode placements. In the pre-processing
unit the EMG signal undergo windowing. The raw EMG
signal is first pre-processed using a Rectangular window.
This window is moved along the entire length of the signal,
partitioning the whole signal into smaller signals of equal
length (equal to the length of the window) and individual
partitions are then used for further normalized for
processing. The required figure of the raw EMG signal is in the fig1(a) .
B. Feature Extraction and Classification
To make a system based on EMG first we need to
extract the features of the acquired EMG signal based on
which it can be further classified for various movements.
Here we focus on few of the time domain features and
frequency domain features. Features in time domain are
usually quick and easy to implement as they don’t require
any transformation and hence have lower computational
complexity.
C. Features obtained
Some of the time domain features are: Mean, Root
Mean Square, Variance, Waveform length, Standard
deviation.
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
7
And some of the frequency domain features are: Modified
mean frequency, Modified median frequency, Mean
frequency, Median frequency, Power. Classification was done using neural networks. Neural
networks are composed of simple elements operating in
parallel. As in nature, the network function is determined
largely by the connections between elements. Normally neural networks are adjusted, or trained, so that a particular
input leads to a specific target output. There are mainly
three activation functions used in neural network. They are :
Hardlim, Purelin, Logsig. Here in this paper we are using
Back propagation algorithm. The input vectors and
corresponding target vectors are used to train the neural
network until it can approximate a function or associate
input vectors. All the time domain and frequency domain
features were used as the input for the neural network. The
designed network consists of three-layers: input layer, tan-
sigmoid hidden layer and linear output layer. Each layer except input layer has a weight matrix W, a bias vector b
and an output vector a. The weight matrices connected to
inputs called input weights (IW) and weight matrices
coming from hidden layer outputs called layer weights
(LW). Fig 2 shows the Neural Network Architecture.
Levenberg-Marquardt (trainlm) algorithm was utilized for
BP training. It is the fastest method for training of
moderate- sized feed forward neural networks and based on
numerical optimization techniques. This was done by
dividing the training input data: 70% for training, 15% for
validation and 15% for testing. Furthermore, the number of data points in training set was more than sufficient to
estimate the total number of parameters in the network. The
weights and bias of input layer and hidden layer were saved
after each training session. When the simulation results are
not satisfactory, the network trained again with the last
saved weight and bias values. This was done to improve the
network performance and to reduce the number of time for
training. Another type of BP algorithm named scaled
conjugate gradient (trainscg) is also used.
D.Feature Selection and Combination
Feature selection is an optimization problem. Its work is
to search the space of possible feature subsets and pick the
subset that is optimal or near-optimal with respect to a
certain criterion Assuming m features, an exhaustive search
would require, examining all possible subsets of size d and
selecting the subset that performs the best according to the
criterion function. There are two types of evaluation strategies: Filters method, Wrappers methods. Here in this
paper we have used filters method. In filters method
evaluation is independent of the classification algorithm.
The objective function evaluates feature subsets by their
information content, typically interclass distance, statistical
dependence or information theoretic measures.
In order to design an effective object classification system,
feature combination is usually adopted in an attempt to
combine the strengths of multiple complementary features
and produce better performance than any individual feature.
Feature combination is a popular method for improving
object classification performances. There are five types of
feature combination methods [7]: Product, Averaging,
Multiple kernel Learning, LP-B, LP-Beta. Here we have
proposed to use a selection based average combination
algorithm to obtain the best classification performance from
average combination. In average combination we have
taken the average of all the 10 features and have organized
them in descending order. The best three features having
the highest recognition rate has been combined. As we have three features so each features contributes around 33%.
III. RESULTS AND DISCUSSION
The summary of the classification performance using
back-propagation neural network is shown in tables below
(table 3 and table 4). The network was trained by 10 sets of
data for different movements. Each set consists of input
feature vector obtained from specific type of hand
movement and corresponding output vector. Different
number of hidden neurons selected for both type of BP
training and their classification efficiency are reported.
Feature vector obtained through combination of three time
and frequency domain features (modified mean frequency, waveform length and RMS value) resulted into highest
recognition rate of 94% compared to combination of all 10
features.
IV. CONCLUSION
The objectives of this study is to derive a feature vector
for higher recognition rate based on lower number of EMG
channels. Four index finger movements (flexion, extension,
aduction, abduction) were classified using back propagation
neural network using L-M and C-G training algorithm. Pre-
processed EMG segments of 1sec duration have been used
as test patterns. Ten different features comprising five time domain and five frequency domain features are used in
evaluating the performance of the neural networks. Two
types of training schemes have been employed for training
the neural network. The experimental results show that the
movement detection can be carried out with an accuracy
rate as high as 94% with combination of best three input
features. The levenberg-Marquardt training algorithm
showed the highest recognition rate compared to Scaled-
Conjugate training algorithm
A. Figures and Tables
Fig1:Raw EMG Signal
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
8
Fig 2: Neural network architecture
TABLE 1. SPECIFICATIONS OF THE SETTINGS OF
THE EMG UNIT DURING ACQUISITION
TABLE 2. MUSCLE SELECTION AND ELECTRODE
PLACEMENT
Electrodes Muscles Function Channel1-lead1 Extensor Digitorum
Muscle Extension of
finger Channel1-lead2 Flexor Digitorum
Superficialis muscle Flexion of
finger Channel2-lead1 Dorsal Interossei Muscle Adduction of
finger Channel2-lead2 Plamar Interossei Muscle Abduction of
finger Ground Styloid Process
TABLE 3. COMPARISON OF RECOGNITION RATES USING
DIFFERENT COMBINATION
Features Flexion Extension Adduction Abduction
Combination
of 10
features
50% 50.5% 54% 54.6%
Combination
of best 3
features
94% 95% 94% 96%
TABLE 4. RECOGNITION RATES OF 10 DIFFERENT FEATURES
Features Flexion Extension Adduction Abduction Modified
Mean
Frequency
75%
80% 85% 90%
Modified
Median
Frequency
50% 45% 70% 65%
Mean
Frequency 40% 25% 10% 25%
Median
Frequency 35% 55% 20% 45%
Power 40% 25% 25% 35% Waveform
Length 75% 70% 70% 85%
RMS 60% 70% 75% 70% Variance 55% 50% 55% 58% Standard
Deviation 20% 50% 40% 40%
Mean 60% 35% 90% 85%
Acknowledgment We acknowledge the department of Electronics and
Communication Engineering, Tezpur University for
providing us proper laboratory facilities to accomplish this
project.
References [1] Choi et al. " Development and quantitative performance
evaluation of a non-invasive EMG computer interface",
IEEE Trans Biomed Eng, 2009, 56:188-191.
[2] Englehart et al."A robust, real-time control scheme for multifunction myoelectric control", IEEE Trans Biomed
Eng, 2003, 50:848-854.
[3] Englehart et al. " Classification of the myoelectric signal
using time-frequency based representations", Med Eng
Phys, 1999, 21:431-438.
[4] Finley et al. "Myocoder studies of multiple myocoder
response", Archives of Physical Medicine and
Rehabilitation, 1967, 48(supple3): 598.
[5] Chan et al. "Fuzzy EMG classification for prosthesis
control", IEEE Transactions on Rehabilitation
Engineering,2000, 8(supple3):305-311.
[6] Farry et al. "Myoelectric teleoperation of a complex robotic hand", IEEE Transactions on Robotics and
Automation, 1996,12(supple5):775-788.
[7] Gehler et al. "On Feature Combination for Multiclass
Object Classification", Max Planck Institute for Biological
Cybernetics Spemannstr. Tubingen, Germany, 38, 72076.
Parameters Values
High cut off frequency 400 Hz
Low cut off frequency 40 Hz
Notch cut off frequency 50 Hz
Amplification rate 5
Common mode rejection ratio 110db
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
9
DEVELOPMENT OF A SYSTEM FOR ANALYSIS OF REMOTE
AUSCULTATION SIGNAL FOR MANAGEMENT OF CHRONIC
ASTHMA CASES
Bony George1, Ajith K.G
2., George Varkey
3, Rajesh M
4
1,2,3 NIELIT, Calicut, Kerala
4 Mobilexion Technologies
ABSTRACT
Asthma is a chronic pulmonary disease affecting millions around the globe. World Health Organization estimates that roughly the equivalent of the population of the Russian Federation suffers from asthma and that this number is rising. World-
wide, deaths from this condition have reached over 180,000 annually. Asthmatic condition reduces quality of life. Pulmonary auscultation using a stethoscope is the common method of diagnosis. However, there are two major limitations. (a) The transient nature of the attack makes it difficult to observe the actual condition by the doctor, unless the patient is admitted (b) The process is subjective (depends on the individual's own hearing, experience and ability to differentiate between different sound patterns). We have developed a low cost embedded system for acquisition, storage, analysis and real-time transmission of auscultation signals that can help in solving both these problems. A custom made stethoscope is used to acquire and store the auscultation signal into the receiving segment of this system. The stored packets are routed to a smart phone that performs local analysis at patient end and real-time transmission to a cloud server and a receiving smart phone at the doctor’s end. The setup allows the
doctor at a remote location to listen to the pulmonary auscultation in real-time so as to decide on the optimum care regime. The analysis and visualizations provided by the software brings in more objectivity for this decision making, by its provisions for comparing current signal and its parameters with previously stored ones.
Keywords: Asthma, Auscultation, Stethoscope, Acoustic Analysis, Spectrogram
I. INTRODUCTION
Asthma is a disease affecting the airways
that carry air to and from the lungs. People who
suffer from this chronic condition (long-lasting or recurrent) are said to be asthmatic. The inside walls
of an asthmatic's airways are swollen or inflamed.
This swelling or inflammation makes the airways
extremely sensitive to irritations and increases
susceptibility to an allergic reaction. The
inflammation causes the airways to become narrower.
Symptoms of the narrowing include wheezing (a
hissing sound while breathing), chest tightness,
breathing problems, and coughing. Asthmatics
usually experience these symptoms most frequently
during the night and the early morning.
Part A of Fig 1 shows the location of the
lungs and airways in the body. Part B shows a cross-
section of a normal airway. Part C shows a cross-
section of an airway during asthma symptoms.
Fig.1:
Physiology of Cause of Asthma
When the airways react, the muscles around
them tighten. This narrows the airways, causing less
air to flow into the lungs. The swelling also can
worsen, making the airways even narrower. Cells in
the airways might make more mucus than usual that
further narrows the airways.
Sometimes asthma symptoms are mild and
go away on their own or after minimal treatment.
Other times, symptoms continue to get worse.
Asthma attacks also are called flare-ups or
exacerbations. Severe asthma attacks can be fatal and
may require emergency care. *[email protected],[email protected]
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
10
According to WHO estimates, 255,000
people died of Asthma in 2005. Over 80% of this is
from low and lower-middle income countries [3].
Asthma creates a substantial burden on individuals
and families as it is more often under-diagnosed and
under-treated. In India, an estimated 57,000 deaths were attributed to Asthma in 2004 and it was seen as
one of the leading cause of morbidity and mortality in
rural India.
Some of the facts about asthma brought out
by WHO illustrate the severity of the problem [2].
Around 8% of the Swiss population suffers from
asthma as against only 2% some 25-30 years ago. In
Western Europe as a whole, asthma has doubled in ten years. In the United States, the number of
asthmatics has leapt by over 60% since the early
1980s and deaths have doubled to 5,000 a year. There
are about 3 million asthmatics in Japan of whom 7%
have severe asthma. In Australia, one child in six
under the age of 16 is affected. India has an estimated
15-20 million asthmatics. Prevalence is between 10%
and 15% in 5-11 year old children.
Being a chronic condition, continuous
medical care is required for asthma. The disorder can
be adequately controlled with drugs. Under diagnosis
and inappropriate therapy remain the major cause of
Asthma morbidity and mortality.
Asthma diagnosis is done by listening to the
internal sounds of the lungs using a stethoscope, which is known in medical terms as auscultation. The
clinician listens to the different sounds in the lungs
corresponding to the inhalation and exhalation and
based upon that decides the intensity of asthma.
One problem with the above is that most of
the time patients do not have obvious asthma
symptoms when they arrive at the doctor’s office. For
instance, you may have coughed and wheezed for a week, and by the time you see your doctor, you have
no symptoms at all. Then suddenly, when you least
expect it, you might have asthma attack
symptoms such as shortness of breath, coughing, and
wheezing. Sometimes allergies to seasonal pollen or
weather changes can trigger asthma attack symptoms.
Other times, a viral infection such as cold or flu can
trigger it. Even exercise or sudden stress or allergies
to aspirin or other medications can cause asthma
attack symptoms.
The above transient nature of the attack is a
major problem for the management of asthma cases.
The attack may occur in the middle of the night far
away from a hospital. The patient is unable to
communicate coherently and the by-stander is unable
to determine the severity of the problem. The only
solution is to depend on the experience of the
clinician at the other end of the telephone line and
many times there are serious gaps in the
communication leading to non-optimum decisions.
Development of appropriate low cost
clinical instruments to be used in the home
environment will be a great help in the above
scenario [8][10].We are in the process of developing
a low cost device for acquisition, storage, analysis
and real-time transmission of auscultation signals for
usage in the home. A custom made chest-piece is
used to acquire and store the auscultation signal as small packets of digital data. These are routed to a
smart phone that performs local analysis at patient
end [11][12]. In case GPRS connectivity is available
at home, the packets are transferred in real time to a
cloud server. A smart phone at the doctor’s end can
pick this up to allow him to listen to the pulmonary
auscultation in real-time so as to decide on the
optimum care regime.
This paper presents the system developed
along with analysis schemes possible with it. It is
organized as follows. Section II describes the
characteristics of the signal and the type of
information that we wish to collect from it. Section
III gives an overview of the system hardware. Section
IV gives an overview of analysis software. Section V
gives a few typical visualization results. The system
is currently under development and the analysis
segment is continuously under improvements as per
the requirements of the consulting clinicians.
II. IMPORTANCE OF LOW FREQUENCY
BIO ACOUSTIC SIGNAL IN CHARACTERIZATION OF
ASTHMA CASES
The diagnosis of chest diseases is facilitated by pulmonary auscultation using a stethoscope. This
device, invented in 1821 by the French Physician,
Laennec, is still the commonest diagnostic tool used
by doctors. Major lung sounds include:
A. Crackle sound
These are discontinuous lung sounds. It is
explosive and transient in nature. Its frequency range
is from 10 to 2000Hz and duration is typically 20 ms.
The number of occurrences and the number of the
crackles in each are indicators of the type of
disease.Crackles can be found in many diseases like
Heart congestion failure, Pneumonia, Bronchiectasis,
Pulmonary fibrosis etc. They are an early sign for
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
11
respiratory diseases, since fine crackles are originated
in small air paths. [2][3]
B. Wheeze
Wheezing results from a narrowing or
partial blockage (obstruction) somewhere in the
airways. The narrowing may be widespread (as in
case of asthma, chronic obstructive pulmonary
disease [COPD], and some severe allergic reactions)
or only in one area (as may result from a tumor or a
foreign object lodged in an airway). The high-pitched whistling sound while you exhale - or wheezing - is a
key sign of both an obstructed airway and asthma.
Wheeze can be heard at the chest and the
trachea. It is a high-pitched sound with dominant
frequency at 300 Hz or more with a duration > 250
ms[1].
It is clear from the above that the low
frequency segment of the auscultation signal is of
primary importance in the characterization of asthma
cases. A clinician uses his previous experience in
identifying these segments for characterizing the
patient condition. This method, even though time
tested and well accepted, has many limitations. It is a
subjective process that depends on the individual's
own hearing, experience and ability to differentiate between different sound patterns. It is not easy to
produce quantitative measurements or make a
permanent record of an examination in documentary
form. Long-term monitoring or correlation of
respiratory sound with other physiological signals is
also difficult.
Over the last 30 years, computerized
methods for the recording and analysis of respiratory sounds have overcome many limitations of simple
auscultation. Respiratory acoustic analysis can now
quantify changes in lung sound, make permanent
records of the measurements made and produce
graphical representations that help with diagnosis and
management of patients suffering from chest diseases
[3]. However, these require costly instrumentation
normally found only in specialist hospitals.
The advent of low cost electronic devices
and advances in mobile and cloud computing now
makes it possible to provide such methods even in the
home settings. The EpionexTM Tele-stethoscope
(ETS) is a system being developed by us in the above
lines. The next two sections provide the hardware
and software details of ETS.
III. ETS - SYSTEM HARDWARE
Fig. 2 gives the block diagram of ETS
system hardware. It begins with the stethoscope chest
piece which is used as the sensor.The chest-wall
movements are so weak that a free-field recording is
not possible. It is essential to couple the diaphragm
acoustically with the chest wall through a closed air
cavity. Vibrations of the chest wall are converted into
pressure variations of the air in the stethoscope. An
electret microphone connected to the end of the chest
piece. Changing the distance between the two plates of a charged capacitor inside the microphone induces
a voltage fluctuation. This low amplitude signal is
amplified with a low noise audio amplifier and
filtered using a bandpass filter so that signal of
interest is amplified and unwanted noise is removed.
Fig.2 Overall system architecture
The filtered and amplified signal is sampled
using an analog to digital coverter (ADC) in the
microcontroller with 16 bit resolution. The resulting
digital signal stream is bufferred, packetized and
transferred to the other end of the set-up in real time
over 3G/GPRS communication channels. This is
achieved by first uploading it to a cloud server from
where it is downloaded to the doctor’s end. Here the
packets are decoded, reassembled and streamed to a DAC connected to the headphone of the doctor, so
that he can listen to the chestpiece. Additionally, he
can comminicate with the patient side using VoIP
communication.
Figure 3 gives the preamplifier and filtering
circuit used in the system.Op-amp OPA2336 based
active filter and amplifier circuit is designed with a
passband frequency of 3 to 2000 Hz so that all the relevent frequency components are included[4].The
amplifier provides a gain of 56 dB. Op-amp OPA336
from Texas Instruments is used. It’s features like low
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
12
power, low noise and single supply makes it ideal for
our purpose.
Fig.3 Filter amplifier
IV. ETS – SOFTWARE
STRUCTURE
Fig 5 gives the segments of the ETS software. It
consists of home segment, cloud segment and
clinician segment. The home segment system consists
of indipendent applications for signal acquisition,
transportation, analysis and coordination. The signal
acquisition application acquires the signal and stores
it in local database. It additionally sends it to the transport application which sends it to the cloud
server and to the smart phone of the remote clinician,
both over the GPRS link. The analysis application
computes the signal parameters and provides reports
and charts for usage by the patient. The cordination
application manages the communication with the
remote clinician. It uses VoIP for voice
communication and data packets for commands.
Using the latter, the doctor is able to take control over
the software so that he can visually indicate the
location of placing the chest-piece on the body of the patient. It also provides chat mode to augment voice
communication.
The cloud segment manages multiple requests from
many clients. It decodes the request packets to
determine the type of request and in case of a request
for DBMS entry, decodes the same and send to the
requesting entity. In case of a signal packet, it
additionally computes standard parameters and store in the DBMS.
Fig.5(a) Software System - Home Segment
Fig.5(b) Software System - Cloud Segment
The clinician segment provides tele-medicine
services to the remote patient. It decodes the packets
and in case of a voice packet depending on the
current state directs it to the head-phone. The state
could be either VoIP or Auscultation. This
management ensures that these voice streams are not
mixed up. It also has an analysis segment to view the
results of signal analysis to compare with previous values.
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
13
Fig.5(c) Software System - Clinician Segment
A number of analysis techniques are
provided to display the time domain signal, convert it
to spectral domain, and display the important
parameters. A number of researchers have come out
with various algorithms for determination and characterization of the signal [6][7][8][9]. Majority of
them uses Short Time Fourier Transform (STFT) as
the primary step in the analysis[4]. It computes the
signal energy distribution in the joint timefrequency
domain.The time-dependent Fourier transform is the
discrete-time Fourier transform for a sequence,
computed using a sliding window. The sliding
window divides the signal into several blocks of
data.Then an N point fast Fourier transform is applied
to obtain the frequency contents of each block ofdata,
where N is frequency. It aligns the centre of the first
sliding window with the first sample of the signal X and extends the beginning of the signal by adding
zeros. The sliding window moves time steps samples
to the next block of data. If the window moves out of
X, the signal is padded with zeros.
We thus have the time domain as well as the
frequency domain representations of the input signal.
This is now used to compute various parameters like dominant frequency within a given band, signal
energy, etc. We can also use these for detection of
weezing instances, cracles within a segment, etc.
These too are presented to the clinician for his
decision making.
V. USE CASE SCENARIOS
Fig 6 gives two use case scenarios using the
spectrogram provided by ETS. The pseudo colouring
used is arbitrary and may be modified based on
recommendation from clinical community.
Fig 6(a): Spectrogram of Volunteer 1
Fig 6(b): Spectrogram of Volunteer 2
In order to remove the influence of heart
sounds and other unwanted lung noises, the
frequency below 100 Hz are masked from the
spectrogram. A distinguishable amount of wheezes
are detected from the second volunteer’s
Spectrogram. Detailed trial with multiple patients is
needed to validate the utility of these for actual
diagnosis.
VI. CONCLUSION
Developing the system for the analysis of
low frequency bio acoustic signals is very useful for
the computer enabled diagnosis, characterization and
management of chronic asthma cases. We have
developed a system for real-time remote acquisition,
storage, analysis and transport of this signal for tele-
medicine application. The work on the development
of analysis methods are continuing. We hope that the
system will help in popularizing these analysis
methods so that remote management of asthma
patients will become more optimal in near future.
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
14
Acknowledgement
This paper is the outcome of a collaborative effort
from two institutions -National Institute of
Electronics and Information technology (NIELIT)
Calicut, and Mobilexion Technologies Pvt. Ltd
(Mobilexion) Trivandrum. We are deeply indebted to
the management of these institutions for having
provided a congenial atmosphere for undertaking this
work.
References
[1]. Abhishek Jain, Jithendra Vepa “Lung sound analysis for wheeze
episode detection” 30th Annual International IEEE EMBS
Conference Vancouver, British Columbia, Canada, August 20-24,
2008
[2]. Bronchial asthma, World health organization FactSheetN206
http://www.who.int/mediacentre/factsheets/fs206/en/
[3]. Sovijarvi, A. R. A., Malmberg, L. P., Charbonneau,Vandershoot,
J., Dalmasso, F., Sacco, C., Rossi, M., Earis, J.E.“Characteristics
of breath sounds and adventitious respiratory sounds”, EurRespir
Rev 2000; 10: 77, 591-596
[4]. Moussavi, Zahra “Fundamentals of Respiratory Sounds and
Analysis” Morgan & Claypool, ISBN 1598290975, 2006
[5]. Jayant V. Mankar, Praveen Kumar Malviya “Analysis of Lung
Diseases and Detecting Deformities in Human Lung by Classifying
Lung Sounds”. International Conference on Communication and
Signal Processing, April 3-5, 2014, India
[6]. Victor Grinchenko,Alexander Artemiev,Oleg Gurenko,Rustam
Nabiev,Anna Glazova “Mobile End user solution for system of
monitoring of Respiratory and cardiac sounds” IEEE XXXIV
international scientific conference electronics and nanotechnology
2014
[7]. R.Jane, S.Cortes, J.A. Fiz, J. Morera “Analysis of Wheezes in
Asthmatic Patients During Spontaneous Respiration.” Proceedings
of the 26th Annual International Conference of the IEEE EMBS
San Francisco, CA, USA • September 1-5, 2004
[8]. Sibghatuallah I. Khan, Naresh P. Jawarkar “Cell phone based
Remote Early Detection of Respiratory Disorders for Rural
Children using Modified Stethoscope” International Conference on
Communication Systems and Network Technologies 2012.
[9]. Saba Emrani, and Hamid Krim. “Wheeze detection and location
using spectro-temporal analysis of lung sounds.”29th Southern
Biomedical Engineering Conference-2013.
[10]. Md. Sarwar Jahan and A. B. M. Aowlad Hossain.“A Low Cost
Stethoscopic System for Real Time Auscultation of Heart
Sound”3rd International Conference on Informatics, Electronics &
Vision, 2014.
[11]. Gary Comtois, John I. Salisbury, and Ying Sun. “A Smartphone-
Based Platform for Analyzing Physiological Audio Signals.” 38th
Annual Northeast Bioengineering Conference (NEBEC), 2012
[12]. Ji Yun Shin, SehiL'Yi, Dong Hyun Jo, Jin Ho Bae, and Tae Sao
Lee. “Development of Smartphone-based Stethoscope
System”.13th International Conference on Control, Automation
and Systems-2013
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
15
FPGA IMPLEMENTATION OF DISTRIBUTED CANNY EDGE DETECTOR
ALGORITHM
DHINAKARAN M1, C. M. SUJATHA
2
1,2College of Engineering, Anna University, Chennai
ABSTRACT In this work implementation of distributed Canny edge detector algorithm onto FPGA platform is attempted. This algorithm computes
thresholds at the block-level based on non-uniform gradient magnitude histogram. Latency is now a function of block size instead of frame size, and is significantly reduced in this algorithm. The proposed algorithm takes 0.729ms with 2538 slices to detect edges when clocked at
50KHz. Fast edge detection of images and videos with high resolutions including full high definition is supported by this algorithm. In addition, quantitative conformance evaluations and subjective tests show that the edge detection performance of proposed algorithm is better than the conventional frame-based algorithm especially when noise is present in images.
Keywords: Canny Edge Detector, FPGA, Histogram
I. INTRODUCTION
Edge detection is a very important first step in image
processing applications. Many types of edge detection
algorithms, such as Prewitt, Robert, Sobel, Gauss-Laplacian, Kirsch and Canny edge detectors [6] are
designed to detect the step response due to the edge
location in the discontinuous part of the image. However,
these operators are sensitive to noise and the edge detected
is discontinuous. Among these algorithms, Canny
algorithm has been widely used in many real-world
applications due to its ability to extract significant edges
with good detection and localization performance. This
algorithm has the advantages of good signal to noise ratio,
and fine edge thinning.
Canny algorithm detects edge using gradients and hysteresis estimation. The image gradients are estimated
using different gradient operators [11]. The first-order
derivative does not guarantee the detection the edges that
map with continuous edge contours and unwanted
branches. The second-order derivative involving zero
crossing suffer from generating erroneous edges due to its
high sensitivity to noise. Gradient magnitudes of four
nearest neighbors along the direction are necessary to
compute two intermediate gradients. The hysteresis
thresholding is obtained by computing the higher and
lower thresholds for edge detection based on the entire
image statistics [8].
Recently, Field Programmable Gate Array (FPGA)
technology has become a viable target for the
implementation of image processing algorithms. Image
processing is difficult to achieve on a serial processor. This
is due to the large data set required to represent the
image and the complex operations that need to be
performed on the image [7]. Many hardware
implementations of Canny edge detection algorithm have
been proposed. The correct setting of the threshold affects
the performance of this algorithm greatly and the
computational cost is very high for real time
implementation [17].
The implementation of Canny edge detection algorithm
onto FPGA-based platforms involves translation of software design directly into VHDL or Verilog using
system-level hardware design tools, which results in a
decreased timing performance. The parallel implementation
proposed by Qian Xu, operates on only 4 pixels in parallel,
resulting in an increase in the number of memory accesses
and in the processing time when compared to the
conventional algorithm. Furthermore, all of these
implementations compute the high and low thresholds off-
line and use the same fixed threshold values for all the
images in order to reduce the computational complexity.
However, this results in a decreased edge detection performance.
When compared to other edge detection algorithms,
Canny edge detection algorithm also involves many pre-
processing and post-processing steps. In order to reduce
memory requirements, latency and increase throughput, a
distributed canny edge detection algorithm is proposed
[10]. In this algorithm, the threshold selection algorithm is
based on the distribution of pixel gradients in a block of
pixels. The threshold is calculated using the data of the
histogram of gradient magnitude instead of being set
manually in a failure or try fashion, and can give quite good
edge detection results without the intervening of an operator.
As the hysteresis thresholds calculation is based on a
very finely and non-uniformly quantized 64-bit gradient
magnitude histogram, this calculation is computationally
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
16
expensive and thereby, hinders the real-time
implementation. To overcome this, an adaptive iterative
threshold selection algorithm proposed in [18] has been
adopted. This algorithm computes the high and low
threshold for each block based on the type of block and the
local distribution of pixel gradients in the block. This algorithm can be mapped onto reconfigurable hardware
architecture efficiently.
Each block can be processed simultaneously, thus
reducing the latency significantly. Furthermore, this allows
the block-based canny edge detector to be pipelined very
easily with existing block-based codec, thereby improving
the timing performance of image/video processing systems
[19].
II. METHODOLOGY
A. Distributed Canny Edge Detector Algorithm
Canny developed an approach to derive an optimal
edge detector based on the criteria related to the detection
performance.
Fig 1: Block Diagram of Proposed Canny Edge Detector Algorithm
The workflow consists of smoothing, finding gradients,
direction calculations, non-maximal suppression, double
thresholding and edge tracing by hysteresis.
B. Pixel Classification
The computing engine local memory 1 consists of m x
m overlapping block, which determines the block type.
There are two stages in block classification units
architecture 1) Pixel classification, 2) block classification.
The local variance of each pixel is utilized. It takes one
adder, two accumulators, two multipliers and one squarer to
perform this computation. Two counters are involved in
counting total number of pixels under each pixel type. The
counter 1 generates the number of uniform pixels C1 and
the counter 2 generates the number of edge pixels C2. Once C1 and C2 are obtained the second classification stage is
initialized. The outputs obtained are given as control signals
to mux1 and mux2 to identify the value of P1. Finally, in
order to produce the enable signal EN, signal P1 is
compared with 0. EN is activated if the P1 value is larger
than 0.
Fig2: Block Diagram of Block Classification Unit (Qian X et.al 2014)
Once EN is activated, the gradient calculation,
magnitude calculation, direction, non-maximal suppression
and low threshold calculation and thresholding with hysteresis units are enabled. If P1 value is less than 0, the
above units are disabled.
C. Gradient Calculation-Vertical and Horizontal
Gradient
After smoothing the image and eliminating the
noise, the next step is to find the edge strength by taking the gradient of the image. The Sobel operator
performs a 2-D spatial gradient measurement on an image.
The gradient magnitude is given by:
|G| = |Gx| + |Gy| (1)
The Sobel operator uses a pair of 3x3 convolution
masks, one estimating the gradient in the x-direction
(columns) and the other estimating the gradient in the
y-direction (rows).
This block calculates the vertical and horizontal
gradients using 3 x 3 convolution kernels. An 8-bit pixel in
row order of the image produced during every clock cycle
in the image smoothing stage is used as the input in this stage. Since 3 x 3 convolution masks are used to calculate
the gradients, neighboring eight pixels are required to
calculate the gradient of the center pixel.
The output pixel produced in previous stage is a pixel
in row order. In order to access 8 neighboring pixels in a
single clock cycle, two First In First out (FIFO) buffers are
employed to store the output pixels of the previous stage.
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
17
The gradient calculation introduces negative numbers.
Negative numbers can be handled easily by using signed
data types. Signed data means that a negative number is
interpreted as the 2’s complement of number. In the design,
an extra bit is used for signed numbers as compared to
unsigned 8 bit numbers i.e. 9 bits are used to represent a gradient output instead of 8. Two gradient values are
calculated for each pixel, one for vertical and other for
horizontal. The 9 bits of vertical gradient and the 9 bits of
the horizontal gradient are concatenated to produce 18 bits.
Since the whole design is pipelined, an 18-bit number is
generated during every clock cycle, which turns out to be
the input of the next stage.
D. Calculation of Gradient Angle
Finding the edge direction is trivial once the
gradient in the x and y directions are known, then
computation of gradient angle is calculated. The edge
direction will equal to 90 or 0 depends on Gx and Gy
value. 0 depend on Gy value and 90 depend on Gx value.
The formula for finding the edge direction is given below:
(2)
E. Tracing Edge Angle
Once the edge direction is known, the next step is
to relate the edge direction to a direction that can be
traced in an image. Any edge direction falling between
to & to is set to . Any edge direction
falling between to is set to . Any edge
direction falling between to is set to .
And finally, any edge direction falling between to
is set to .
Fig3: Block Diagram of Magnitude Angle Calculation (Qian X et.al 2014)
The bigger the matrix, the more number of angles
one could get, which implies the edge angle will be
more precise, i.e. it will follow the edge better. On the
down side, it will be a bigger computation task as now the
kernel/mask size is bigger.
F. Non-Maximal Suppression
After the edge directions are known, non-maximal
suppression is applied. Non-maximal suppression is used
to trace along the gradient in the edge direction and
compare the value perpendicular to the gradient. Two
perpendicular pixel values are compared with the value in
the edge direction. If their value is lower than the pixel on the edge then they are suppressed i.e. their pixel value is
changed to 0, else the higher pixel value is set as the edge
and the other two suppressed with a pixel value of 0. Then
the points along the curve where the magnitude is
biggest are marked, by looking for a maximum along a
slice normal to the curve (non-maximum suppression).
These points should form a curve.
Fig4: Concept of Non-Maximal suppression operation
The non-maxima elimination filter is used in order to
eliminate pixels that are not part of a continuous line. In
other the pixels which are having high gradient magnitude
but they are not a part of a continuous line can be easily
eliminated using this filter. If the current pixel and the
values of the derivatives at that pixel are Gx and Gy, the
direction of the gradient can be approximated to any one of
the sectors.
Once the direction of the gradient is known, the values of the pixels found in the neighborhood of the pixel under
analysis are interpolated. The pixel which has no local
maximum gradient magnitude is eliminated. The
comparison is made between the actual pixel and its
neighbors along the direction of the gradient.
Since the gradient calculation depends on the direct
neighborhood pixels, a 3 x 3 window operator is used. The
output produced along the vertical and horizontal stage is
an 18-bit number, first nine bits are horizontal gradient and
other nine bits are vertical gradient.
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
18
Fig 5: Block Diagram of Non-Maximal Suppression
In order to access all the pixels in the 3 x 3 window at
the same time, two 18-bit FIFO buffers each having width
of image minus three array size are employed. To calculate
the phase and magnitude at each and every pixel, the
horizontal and vertical gradient values derived from the
eighteen bit number are used. The output produced in this
stage is subjected to a threshold and stored in the memory.
Instead of implementing each stage separately, the
pipelined design utilizes the resources efficiently on the FPGA and also increases the execution speed of the
algorithms, because the output is produced each and every
clock cycle with an initial latency till the pipe is full.
G. Hysteresis threshold
Hysteresis is used as a means of eliminating streaking.
Streaking is the breaking up of an edge contour caused by the operator output fluctuating above and below the
threshold. If a single high threshold is applied to an
image and an edge has an average strength equal to
high threshold, then due to noise, there will be
instances where the edge dips below the threshold.
Fig6: Block diagram of Threshold Calculation
To avoid this, hysteresis uses 2 thresholds, a high and a
low. Any pixel in the image that has a value greater
than high threshold is presumed to be an edge pixel, and
is marked as such immediately. Then, any pixels that are
connected to this edge pixel and that have a value
greater than low threshold are also selected as edge
pixels. The Altera DE2 kit includes a VGA display port via
a DB15 connector.
III. RESULTS
The distributed Canny edge detection algorithm is
simulated on MATLAB and modules were written and
tested in Verilog HDL.
A. Simulation Results
The edge detection scheme developed in this work
is first implemented using MATLAB software tool and tested for many images.
(A) (B)
(C) (D)
(E) (F)
Fig9 (A): Input Lena image, Fig9 (B): Lena image with 90% noise, Fig9 (C): Conventional Output Image of Canny Edge
Detection without noise, Fig9 (D): Conventional Output Image of Edge Detection with noise, Fig9 (E): Conventional Output Image
of Canny Edge Detection without noise, Fig9 (F): Proposed Output Image of Edge Detection with noise
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
19
Canny edge detection algorithm is performed with help
of 480 × 640 Lena image is shown in Fig (9A) obtained
from USC SIPI database along with the image corrupted
with 90% noise as in fig (9B). The horizontal gradient Gx
and vertical gradient Gy of these images are calculated at
each pixel location by convolving the image with gradient masks. Gradient calculation, non-maximal suppression is
performed.
Then output images of Lena obtained from
conventional Canny edge detector algorithm using 480 × 640 are shown in fig (9C and 9D) without and with noise
respectively. Fig (9E and 9F) represents output images
obtained using distributed Canny edge detection algorithm
without and with noise respectively which results in a
significant speedup without sacrificing the edge detection
performance. The input image segmented into 64 × 64
blocks then Canny edge detection is performed. The result
shown this proposed algorithm could detect edges properly
even for noise corrupted images.
B. Synthesis Results
The project has been implemented on an Altera DE2
board. The inherent features of this board make it suitable
for our application. The DE2 board has the Cyclone II
processor. DE2 board provides users many features to
enable various multimedia project developments.
TABLE I
Device Utilization Summary
Logic
Element
Combination
Function
Detected
Logic Reg
Total
Memory (bits)
Number
of Slices
4580 3198 4314 336725 2538
TABLE I presents resource usage and performance for
distributed canny edge detection algorithm.
Fig10: Test Bench Waveform of Proposed Algorithm
Simulations in Xilinx using fixed point operations
generated results for the 480 × 640 image using the proposed distributed Canny edge detector with block size of
64×64 and a 3×3 gradient mask.
(A) (B)
Fig11 (A): Input Lena image via VGA port Fig11 (B): Output Image of Distributed Canny Edge Detection via
VGA port
Result is obtained by loading the text file containing
the output image pixels onto MATLAB, and displaying the
output image. . Horizontal and vertical gradient magnitudes
are fetched from local memory and used as input to non-
maximal suppression unit. It computes gradient direction at each pixel and suppress all the non-maximal values.
TABLE II
PSNR Table
Sample
Image
Time
elapsed
MSE PSNR
Lena 0.7295 3.7189 38.4267
Fruits 0.7389 3.6982 37.4509
Barbara 0.7539 3.7245 37.4201
Peppers 1.1278 5.7189 36.5577
Baboon 0.7574 3.3603 37.8670
IV. CONCLUSION
The implementation of distributed Canny edge detector
algorithm on FPGA is carried out using Altera DE2.
The steps that were implemented in hardware were
approximated and simplified appropriately for a cost and
time efficient implementation. The architecture can be used
for any size images by adjusting the width of generic
memory modules. This architecture also capable to operate
at higher frequencies. Based on the timing results, computation of magnitude and direction of gradient steps of
the distributed canny edge detector algorithm are faster in
hardware. The resulting optimized enhanced image fits on a
FPGA such as Altera DE2, with sufficient resources
available for to make use of the desired edge information.
The simulation and synthesis results were obtained and
designs are successfully validated using hardware
simulation feature of using Cyclone II family chip type on
development kit type DE2. In order to reduce the large
latency and meet real time requirements, the distributed
Canny edge detector algorithm presented, to compute edges
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
20
of multiple blocks at the same time. Most importantly,
conducted conformance evaluations and subjective tests
show that, compared with the frame based canny edge
detector, the proposed algorithm yields better edge
detection results for both clean and noisy images.
Distributed Canny yields advantages such as low latency, better performance, pipelining. The algorithm is mapped
into the Quartus Cyclone 2 FPGA platform and tested using
ModelSim. The synthesized results shows 4314 slice
utilization and 336725(bits) memory utilization. The
proposed architecture takes only 0.729ms to detect edges of
480 × 640 images in USC SIPI data base when clocked at
50MHz.
REFERENCES
[1] Alzahrani F.M, and Chen T, “A real-time edge
detector algorithm and VLSI architecture,” Real-
Time Imaging, vol. 3, no. 5, pp. 363 – 78, 1997.
[2] Benedetti, A., Perona, P.: “Real-time 2 -D Feature
Detection on a Reconfigurable Computer,”
Proceedings of the 1998 IEEE Conference on
Computer Vision and Pattern Recognition, 1998.
[3] Chou, C., Mohanakrishnan, S., Evans, J.: “FPGA
Implementation of Digital Filters,” Proc. ICSPAT,
1993.
[4] Gentsos, C., Calliope, L.S., Spiridon, N., and
Vassiliadis, N., “Real Time Canny Edge
Detection Parallel Implementation for FPGAs”.
17th International Conference on Electronics,
Circuits, and Systems (ICECS), Athens 2010 IEEE.
[5] Lorca, F.G., Kessal, L., and Demigny, D., “Efficient
ASIC and FPGA implementation of IIR filters for
real time edge detection,” in Proc. IEEE ICIP, vol. 2.
Oct. 1997, pp. 406–409.
[6] Lourenco, L.H.A., “Efficient implementation of
canny edge detection filter for ITK using CUDA,”
inProc. 13th Symp. Comput. Syst., 2012, pp. 33–40.
[7] Luo, Y., and Duraiswami, R., “Canny edge detection
on NVIDIA CUDA,” in Proc. IEEE CVPRW, Jun.
2008, pp. 1–8.
[8] Narvekar, N.D., and Karam, L.J., “A no-reference
image blur metric based on the cumulative
probability of blur detection (CPBD),”IEEE Trans.
Image Process. vol. 20, no. 9, pp. 2678–2683, Sep.
2011.
[9] Neoh H, and Hazanchuck A, “Adaptive edge
detection for real-time video processing using
FPGAs,” Altera Corp., San Jose, CA, USA,
Application Note, 2005.
[10] Park, I. K., Singhal, N., Lee, M.H., Cho, S., and
Kim, C.W., “Design and performance evaluation of
image processing algorithms on GPUs,” IEEE Trans.
Parallel Distrib. Syst., vol. 22, no. 1, pp. 91–104,
Jan. 2011.
[11] Peter Mc Curry, Fearghal Morgan, Kilmartin, L.,
“Xilinx FPGA implementation of a pixel processor
for object detection applications”. In the Proc. Irish
Signals and Systems Conference, Volume 3,
Page(s):346 – 349, Oct. 2001.
[12] Qian Xu., Srenivas, V., Chaitali, C., Fellow, IEEE,
and Lina, J. K., Fellow, IEEE, “A Distributed Canny
Edge Detector algorithm and FPGA
Implementation”, IEEE transactions.
[13] Qian Xu, Lina, Karam J, “A Distributed Canny Edge
Detector: Algorithm and FPGA Implementation”,
IEEE Transactions on Image Processing DOI
10.1109/TIP.2014.2311656.
[15] Rao D.V, and Venkatesan M, “An efficient
reconfigurable architecture and implementation of
edge detection algorithm using handle-C,” in Proc.
IEEE Conf. ITCC, vol. 2. Apr. 2004, pp. 843–847.
[16] Varadarajan, S., Chakrabarti, C., Karam, L.J., and
Bauza, J.M., “A distributed psycho-visually
motivated Canny edge detector,”IEEE ICASSP, pp.
822 –825, Mar. 2010.
[17] Wenhao, H., and Kui, Y., “An Improved Canny
Edge Detector and its Realization on FPGA”
IEEE Proceedings of the 7th World Congress on Intelligent Control and Automation, Chongqing,
China, June 25 - 27, 2008, pp. 6561-6564.
[18] ftp://ftp.altera.com/up/pub/Webdocs/DE2_UserMan
ual.pdf
[19] ftp://ftp.altera.com/up/pub/University_Program_IP_
Cores/VGA.pdf
[20] http://uwf.edu/bshaer/EEL4713L/tut_quartus_intro_
vhdl.pdf
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
21
VLSI IMPLEMENTATION FOR IMAGING BASED CLASSIFICATION OF
NEURODEGENERATIVE DISEASES- A proposal Neha Gopal. N
1, T. Christy Bobby
2
1,2East Point College of Engineering and Technology, VTU, Bangalore, India
ABSTRACT
Neurodegenerative diseases are mainly occurs due to the deterioration of myelin sheath of neurons, brain spinal cord, and peripheral nerves. The economic and social burden of the diseases is massive and rising too rapidly. There are several kinds of neurodegenerative diseases in that the present work is focused Alzheimer’s diseases. The aim of this work is to automatically classify patients as probable Alzheimer’s diseases (AD) subjects or normal subjects. In this work a fusion strategy is proposed to mix together bottom-up and top-down information flows. Bottom up stage includes a multiscale analysis of different image features, top-down flow includes learning and a fusion strategy which is formulated as max-margin multiple kernel optimization problems.
I. INTRODUCTION
Neurodegenerative diseases are a debilitating condition
where progressive degeneration or death of nerve cells
takes place which causes problems in mental functioning, or
with movement. It basically affects neurons of human body;
neurons are the building blocks of nervous system that includes brain, spinal cord, peripheral nerves. Neurons
neither reproduce nor replace themselves so once when they
are damaged or die they cannot be replaced by human body.
Neurodegenerative diseases comprise variety of mental
symptoms which cannot be evolved by the visual analysis
made by radiologists. World wide it is estimated that
approximately 20-30 million people suffer from
neurodegenerative diseases.Many researchers have
suggested that neuroimaging [1] may become one of the
valuable tool in the early detection and diagnosis of
neurodegenerative diseases. Biochemical, clinical,
neuropsychological analysis against neuroimaging remains to be demonstrated for large population, but still there exists
sufficient evidence of patients suffering with different states
of neurodegenerative diseases.
II. METHODOLOGY
In this project we take the set of normal and abnormal MR images of the patients and compare them. The dataset
used in this work is MIRIAD(Minimal Interval Resonance
Imaging in Alzheimer’s Diseases).The entire dataset is
trained in traning phase; this dataset will calculate the
saliency information from the patients MR images, the
selected features include edges, intensity, orientation.
Support Vector Machine is one of the most popular
technique that is used in this work which will classify
individuals with several neurological disorders. By using
neuroimaging data [7]( MR brain images) a complete
review and comparision of SVM based approaches for classifying neurological and psychiatric diseases can be
made. Features like binary tissue [5] segmentation, cortical
thickness estimations, intensity [2], textural information
[3],[4] is fed to SVM classifier. The computational time, the
presence of unwanted, irrelevant and noisy features is
reduced by using dimensionality reduction technique. The required information for classification is extracted either
from specific regions of interest (ROI) [6]or from whole
brain volume. Analysis which is performed on known
diseases locations leads to more significant and stronger
conclusions. In this work a fusion strategy is used that will
together bottom-up and top-down information flows.
Bottom up stage includes a multiscale analysis of different
image features, top-down flow includes learning and fusion
strategies which are formulated as max-margin multiple
kernel optimization problem.This proposed system is
implemented in MatLab and the module is then
implemented in VLSI domain. The project detailed information is provided in architecture of the proposed
system presented in fig.1
Fig.1 block diagram of the proposed approach
III. Result
Fig2. and fig3. Represents intermediate results of the
current work. The bright region obtained from using visual
saliency technique in the above fig 2 and fig 3 represents
the tumor region in Alzheimer’s diseases and also it gives
information about the size of the tumor and severity the
diseases condition.
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
22
Fig.2
Fig3. Original MR image and saliency of original image
Conclusion This work is intended to design an automated classification
system for the pathological diagnostic of the disease (AD)
in its infancy stage so that people could be treated before it
becomes to worse over time.
Acknowledgment We acknowledge VGST by DST for funding this project.
References [1] J. Beutel, H. Kundel, and R. Van Metter, Handbook of Medical
Imaging. Bellingham, WA:SPIE Press, 2000, vol.1, Psychophys.
[2] P. Padilla, M. lopez, J. Gorriz, J. Ramirez, D. Salas-Gonazalez, and I.
Alvarez, “NMF-SVM based cad tool applied to functional brain
images for the diagnosis of Alzheimer’s diseases," IEEE
Trans.Med.Imag., vol.31, no.2, pp.207-216, Feb.2012.
[3] M. Garcia-Sebastian, A.Savio, M.Grana, and J. Villanua, “On the use
of morphometry based features for Alzheimer’s diseases detection on
MRI, “in Bio-Inspired System:Computational and Ambeint
Intelligence ser.Lecture Notes In Computer Science. Berlin,
Germany:Springer, 2009, vol.5517, pp.957-964
[4] N. Doan, B.Van Lew, B.Lelieveldt, M. Van Buchem, J. Reiber, and
J. Milles, “Deformation texture-based features for classification in
Alzheimer’s diseases,”SPIE Med.Imag.,2013
[5] M. Liu, D. Zhang, P. Yap, and D. Shen, “Hierarchical ensemble of
multi-level classifier for diagnosis of Alzheimer’s diseases.”in
Machine Learning in Medical Imaging, ser. Lecture Notes in
Computer Sscience. Berlin, Germany:Springer, 2012, vol.7588,
pp.27-35.
[6] B. Magnin et al., “Support Vector machine-based classification of
Alzheimer’s diseases from whole-brain anatomical
MRI,”Neuroradiology, vol.51, pp.73-83, Feb.2009.
[7] G.Orru, W. petterson-Yeo, A. Marquand, G. Sartori, and A. mechelli,
“Using Support Vector Machine To Identify Imaging Biomakers Of
Neurological And Psychiatric Diseases: Article review,
“Neurosic.Biobehav.Rev.,vol.36, no.4, pp.1140-1152, Apr.2012
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
23
REAL TIME pH MEASUREMENT FOR MONITORING EFFLUENT
DIALYSATE IN AN ARTIFICIAL KIDNEY
Hemalatha R.J1, Shaliya.B
2, Saranya.B
3
Assistant Professor1, Students
2,3, Sathyabama University, Chennai
ABSTRACT
Our paper envisages building an online pH measuring system by which the nephrologist can look at the pH values continuously. The safety with the system ensures that pH is maintained almost neutral and does not create
acidosis to the patient. There is no known work or study related to effluent dialysate. We believe that the effluent
dialysate can provide abundant data which can be analyzed under such circumstances of patient not feeling well
post dialysis.
Keywords: pH, Effluent Dialysate
I. INTRODUCTION There are around 150 countries in the
world reported to provide dialysis care with renal
failure. In the developed countries like Japan, USA
and European countries the rate of kidney failure
and dialysis undergoing patients increases gradually
from 1 to 4 % [1]. Even though dialysis is mainly
done for kidney failure patients but all the patients
are not treated with the dialysis. This is done in
patients with end stage kidney failure i.e, with a
loss of about 90% and GFR < 15. Hence dialysis helps in maintaining the body by removing the
waste from blood[2].
Hence under circumstances the
hemodialysis as treatment for end stage renal
failure patients implementing six times weekly
would harm the patient. Along with this ,there are
only few organ donors available for transplantation,
this gave way for developing the alternative
treatment and devices for kidney. Various factors
plays a vital role in the dialysate namely temperature, pH, volume etc. [3]. The temperature
should be set as 37 to 38 degrees,if it is too low it
leads to reduce d diffusion, if it is too high it leads
to hemolysis. subsequently the pH should be 7.0
and the pH less than 7.0 leads to acidicity and the
pH greater than 7.0 leads to alkalinity. Normal pH
of the body is about 7.35 – 7.45. During
Hemodialysis treatments are made based on the
survey of subjects weight and blood pressure.As a
result treatments are made based on symptoms and
side effects.
This concept of real time monitoring is
based on real time and consequent measurement of
the signals from the subject. Parameters available
for online monitoring are: thermal energy balance., dialysate conductivity, urea kinetics and blood
volume (BV) changes .Our paper envisages on monitoring pH in spent or effluent dialysate during hemodialysis for theoretical evaluation of post
dialysis prognosis of the patient.
II. METHODOLOGY
The scope of our paper lies on defining
that effluent dialysate has data and our paper is
designed to monitor one such data ie pH. Presently
the pH and conductivity are measured either in the
lab by sending the sample or by hand-held meter
that are injected by a syringe or pH strips. In this case the repeatability of the pH and conductivity
cannot be assured unless one does the sampling
frequently. Again, real time measurement is not
possible with pH strips .Our paper highlights on
building an online pH measuring system by which
the sister- in charge or the nephrologist can look at
the pH values continuously. The safety with the
system ensures that pH is maintained almost neutral
and can overcome human error or human
negligence of random sampling of the spent
dialysate. In case of End Stage Renal Failure
patient’s, usually prior to dialysis the patient’s
blood would have accumulated electrolytes, toxins
and excess water during the previous 48 hours
which indicates that the patient’s blood is more
acidic or more alkaline with increased toxins and
edema. The pH measurement after dialysis in the
spent dialysate indicates that the effluent is more
acidic or more alkaline, more toxic which is directly
proportional to the patient’s well being. Online
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
24
measurement of pH at regular intervals indicates the
changing values as per the patient’s using
electrolytes, excess water and other toxins, helps
sister-incharge or the nephrologists can look at the
pH values continuously and it ensures that pH is
maintained almost constant and does not create
acidosis or alkalosis to the patient
A. HYDRAULICS USED
The hydraulics is completely conceived
and fabricated in the inhouse lab. This being the
spinal cord of the entire project, care has been taken
to procedure high quality parts that are suitable to
the design. Keeping in mind that effluent dialysate
should have a free flow, the design of the hydraulics
of the project starts with ‘Y' connector [5]. The
output from the hemodialysis machine is from
12mm internal diameter connector. Usually a flexible pipe of 12.5mm internal diameter is
connected to this connector of dialysis machine.
The flexible pipe is taken to the drain inlet while the
connector on the dialysis machine is 6 inches from
the floor level. Our instrument is kept at a height of
3 feet from floor level to facilitate the flow gradient
from the output of dialysis machine to the project
device and then from project device to the actual
drain.
‘Y’ Connector : The ‘Y’-connector has same bore
diameter as that of the dialysis machine output. A
connector is threaded to the stem of the ‘Y’ which
will act as an inlet to the project device. While the
straight arm goes to a single short solenoid, the right
arm of the ‘Y’-connector is connected to a shunting
connector by an adaptor and a reducer.
Fig:2.1 Y Connector
Fig:2.2 T Joint
‘T’ Joint: 'T' joint is the backbone of the hydraulic
system. The reducer which is connected to the
adaptor in the Y connector is threaded to this ‘T’
joint. The output of the ‘T’-joint is mirror image to
its input, ending in a connector that takes the plastic
tube to the drain at the wall. The ‘T’ joint is placed
as inverted ‘T’ in the project device. The inner
diameter of the ‘T’ section on all its 3 opening is
uniformly 28mm. The stem of the ‘T’ has a
threaded nut into which the pH electrode is inserted. The bulb of the glass electrode is placed at 10mm
from the bottom of the inverted ‘T’ section.
B. SOLENOID VALVE
A solenoid valve has two main parts: the solenoid and the valve. The solenoid converts
electrical energy into mechanical energy which, in
turn, opens or closes the valve mechanically. The
design of a basic valve. At the top the valve in its
closed state. The function of the spring is irrelevant
for now as the valve would stay closed even without
it. The solenoid at the outlet port of the ‘T’-joint is
closed first. After the lapse of a preset time the
solenoid at the inlet port of the ‘T’-joint is closed.
Both the valves are in closed position hence holding
a prescribed volume of fluid in the ‘T’-joint. During this period, the dialysate is flowing freely from the
bypass and the shunt without creating any
backpressure to the effluent dialysis from the
dialysis machine.[6].
The pH electrode in the ‘T’-joint is activated
and a reading of the pH for that particular volume is
noted. The relay that has activated the pH meter
comes to ‘OFF’ state in conjunction with the other two relays that have activated the two solenoid
valves. The above cycle is repeated at preset
intervals for the entire duration of the dialysis
procedure.
The solenoids that store the effluent dialysate in
the pH measuring well are driven by two relays that
operate sequentially to the preprogrammed timing.
In other words, relay 1 which operates the solenoid valve 1 (at the exit of the measuring well) is turned
‘ON’ and relay 2 operates solenoid valve 2 (at the
entrance of the measuring well) is turned ‘ON’.
After a lapse of predetermined time, this
program ensures that the measuring well is filled
and stable during the measuring process.After SV1
and SV2 are closed, relay 3 takes the pH measurement from standby to ‘ON’ state. pH
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
25
measurement takes few seconds to indicate a
stabilized pH reading. The reading is noted down.
After completing the pH reading, all the three relays
return to their original state.
C. pH METER:
The pH electrode is fixed mechanically to a
threaded cap for facilitating to be immersed into the measuring well. A good feature is it is the addition of a temperature sensor to the pH electrode. By housing the temperature sensor in the same body as the reference elements of the electrode, temperature compensated readings can easily be made with a single pH probe. The meter circuit is no
more than a voltmeter that displays measurements
in pH units instead of volts.. After calibrating the
pH electrode, it is ready for measuring the pH of
other solutions. The pH probe measures pH as the
activity of hydrogen ions surrounding the glass bulb
at its tip. The probe produces a small voltage (about
0.06 volt per pH unit) that is measured and displayed as pH units by the meter.The manual
switch which brings the standby to pH measurement
for calibration purpose also comes with three
connections. In piano switch each of these
connections are taken to appropriate poles so that
the normal position pH measurement is controlled
by the relay [7].
TABLE 1: PH AND TEMPERATURE VALUES OF PATIENTS
The pH value given in the tabular column is
averaged value measured at 20 mins interval during
the dialysis procedure for 240 minutes.
III. RESULT
The pH electrode is calibrated in the
known solutions of 4 pH and 7 pH everyday before
connecting the device to the dialysis machine. The
device is switched ON during the priming of
dialysis machine with a dialysate. At the time of
priming the artificial kidney, the digital timer in the
device is programmed to measure pH of the effluent
dialysate at pre-fixed intervals such as 5 minutes, 3
minutes and so on. At the time of hooking the
patient to the artificial kidney the timer switch for
‘RUN’ is activated momentarily. The timer display
then will show as RUN. After ensuring that
everything is in place, the device is started by
activating ‘START’ button momentarily. The
AUTO/CAL button is returned to AUTO from its
CAL mode. The device is ready to take readings and the readings are noted down in the following
table.Consent of the patient is required only for
putting his name, age, sex and present health status.
IV. SUMMARY AND CONCLUSION
The entire paper represents the hypothesis of the
effluent dialysate that will have useful data.This
data represents invariable the post dialysis of the
patient or the deficiencies in the dialysis itself.The
Dialysis machine has two major subsystems such as
fresh prepared dialysate and artificial kidney. The
dialysate is prepared using pure water (reverse
osmosis water) with known amounts of salts
dissolved into premeasured pure water. The freshly
prepared dialysate is pumped into the artificial
kidney passing around the fibers. The patient’s blood (impure) is drawn into the fibers from one
end and it travels through the other end during
which the impure blood gets purified. During this
process the freshly prepared dialysate becomes
effluent dialysate.
We have divided the paper into three parts
such as hydraulics, a digital timer and the pH meter.
We have bought together all these parts as one integrated system. This system is easy to hook up to
the dialysis machine with simple tube connecting
output of dialysis machine to the inlet of the device.
The device is powered by 230V, 15 Hz mains
supply. The device can be utilised with any make of
hemodialysis machine.. The ease of connecting the
device to the dialysis machine and priming it has
been smooth and with out any drawbacks. It can be
said that the device is ready for further clinical
trails.
The study indicates possible variations and the mean of variations is given to the nephrologist
for verification of the patients’ well being post-
dialysis.It is suggested that the patients’ whole
blood pH may be recorded prior to dialysis and
post-dialysis for further study.If the mean pH of the
dialysate is found to be acidic / alkaline the
Patient Gender pH
value
Tempe
rature
Remarks
Patient
1
M 6.4 37 stable
Patient
2
M 6.5 36 stable
Patient
3
F 6.3 36 stable
Patient
4
M 6.62 37 stable
Patient 5
M 6.5 37 stable
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
26
patients’ post-dialysis blood may become necessary
for bringing it to normal value with appropriate
medications.There are instances when a patient
complains of ‘NOT FEELING WELL’, then the
nephrologist can take a note of the effluent dialysate
variation on the graph along with patients’ post-
dialysis pH value. This can prescribe appropriate
medication to switch the state of the patient to the
condition of ‘FEELING WELL’.The device can be
incorporated in the dialysis machine by its manufacturer for ease of correlating the
readings.Clinically the readings may help during
CRRT extensively if found suitable.
References [1] Kloppenburg WD, Stegeman CA, Hooyschuur
M, van der Ven J, de Jong PE, Huisman RM.
Assessing dialysis adequacy and dietary intake in
the individual hemodialysis patient. Kidney Int
1999; 55:1961-1969.
[2] Brimble KS, St Onge J, Treleaven DJ, Carlisle
EJ. Comparison of volume of blood processed on
haemodialysis adequacy measurement sessions vs
regular non-adequacy sessions. Nephrol Dial
Transplant 2002; 17:1470-1474.
[3] Eddy CV, Arnold MA.Near-infrared
spectroscopy for measuring urea in hemodialysis
fluids. Clin Chem 2001; 47:1279-1286.
[4] Eddy CV, Flanigan M, Arnold MA. Near-
infrared spectroscopic measurement of urea in
dialysate samples collected during hemodialysis
treatments. Appl Spectrosc 2003; 57:1230-1235. [5] Elaine N. Marieb,Human Anatomy and
Physiology,Pearson Education, Dorling
Kindersley,1989,997-1001
[6] Eli A. Friedman, Mary C. Mallappallil, Present
and Future Therapies for End-Stage Renal
Disease,World Scientific Publishing,2010,27-29.
[7] Angel L. M. DE FRANCISCO, Celestino
PIÑERA, Challenges and future of renal
replacement therapy, Christopher R. Blagg,
International Society for Hemodialysis, Blackwell
Publishing Ltd, PIÑERA, 2006.
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
27
SPATIAL DOMAIN FILTERS IN PREPROCESSING OF OPTICAL
COHERENCE TOMOGRAPHY
Arati Sinha1, Jintu Das
2,K.Venkataraman
3
1,2Department of Bio-Medical Engineering, Bharath University, Chennai. 3Assistant Professor,Department of Bio-Medical Engineering,Bharath University,Chennai.
ABSTRACT Imaging plays an important role in the field of diagnosis. Therefore it is necessary to have images with high resolution and reduced noise. Optical Coherence Tomography images are most commonly affected by speckle noise.There are various methodologies used for reducing this noise. In this paper filtering technique in spatial domain are evaluated to reduce speckle noise in Optical coherence tomography images which also effective for other imaging modalities like ultrasonography where speckle noises are prominent. The performance of these filters were evaluated visually and also based on three parameters namely mean square error (MSE), signal to noise ratio (SNR), peak signal to noise ratio (PSNR). Out of various filters used wiener filters were found comparatively effective. Most filters just average the images instead of removing noise.
Keywords: Speckle noise, Resolution, Mean Square Error, Signal to noise ratio, Peak Signal to noise ratio
1. INTRODUCTION
Optical coherence tomography (OCT) is evolving as a new and promising non-
invasive imaging modality. Ophthalmology
has most widely benefited from inventions and improvements made to OCT imaging. Retinal
images are taken and measurements are
performed on them for diagnosis of pathologies in retinal layers by physicians.
One example for such measurements is
determining the thickness of a single layer of
retina.Other than its primary application in OCT; it is also popular in other medical
imaging tools such as ultrasonography,
gastroenterology, dermatology, endoscopy, urology, brain morphology. The short
temporal coherence property of broadband
light is used by OCT imaging modality to extract structural information from
heterogeneous samples such as tissue. The
main advantage of OCT is that it is non-
invasive, contact-free and gives high resolution both in depth and transversally. But
it has limited penetration in scattering media.
The most severe problem faced in OCT imaging is that the images are corrupted by
speckle noise which complicates the task
ofinterpretation and diagnosis. The main
objective of image denoising is to remove this this type of noise such that the important
features of the signals are retained as much as
possible [1]-[5].
1.1 SPECKLE NOISE
Random variation of image intensity
appearing as grains called as noise. In other
words noise means pixels in the image shows different pixel values instead of original pixel
values. It is the undesirable effect in the image.
Several factors are responsible for introduction of noise in the image during image acquisition
or transmission. The noise can affect the image
to different extents depending on the type of disturbance such as interference in the
transmission channel, insufficient light
levels,sensor temperature etc. Image noise can
be classified as impulse noise (salt and pepper noise), amplifier noise (gaussian noise),
periodic noise, multiplicative noise (speckle
noise), quantization noise [6]. Generally our focus is to reduce speckle noise from OCT
images.
OCT images and other imaging
modalities that involve coherent light source
are affected by speckle noise. Constructive and
destructive interferences of the backscattered waves appear as grains in the image that
degrades the image quality. Physical
parameters of the imaging modality such as size and temporal coherence of light source
and the aperture of the detector also influence
the formation of speckle noise. Speckle is not
purely noise, it also transports image information. So there is difference between
speckle and speckle noise where the later is
pure unwanted component in OCT image signal. Speckle noise reduces contrast and
makes difficult for the highly scattering
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
28
structures in tissue difficult to resolve. Speckle
is image information and its pattern does not change until there is change in physical
parameters of the imaging system. Speckle and
thus noise are closely correlated. If the entire
speckle in an image of dense tissue is removed completely then no image would remain [6].
2. RELATED WORKS
The standard and most commonly
used filters for speckle noise reduction are adaptive filters which include Lee filter, Frost
filter and Kuan filter which uses multiplicative
noise model. For OCT images, the Rotating kernel transform filter is one of the best single-
resolution filters to reduce speckle noise. The
RKT technique filters an image with a set of kernels and keeps the largest output of the
filter at each pixel. RKT technique is efficient
but do not use any information of the speckle statistics. Noise reduction technique can be
applied before or after formation of image.
The techniques that are applied to the OCT interference signal that is after OCT image is
formed is called post processing techniques.
Other techniques are applied to the complex interference signal, called as complex domain
method. Techniques for despeckling can also be classified into single-resolution and multi-
resolution techniques, such as wavelet-based
techniques. Two single-resolution filters: Lee and RKT filters and four multi-resolution
filters: OCTWT (Wavelet thresholding for
OCT), FBT(Feature based Wavelet Thresholding), WGE (Wavelet shrinkage with
Gamma-Exponential method), BLS-GSM
(Bayesian Least Square estimator with Gaussian Scale Mixture prior) were used on
noise models and evaluated. The performance
of RKT filter was better than the Lee filter but both of the single-resolution filter’s efficiency
was inferior to the performance of
multiresolution images [1].
Reduction of speckle noise using a
partially spatially coherent source was performed in a Michelson interferometer in a
parallel detecting OCT system. A Gaussian-
Schell model for a partially spatially coherent source is used in the OCT stimulation. Such
source was generated by a spatially coherent
broadband light source and a multimedia fiber.
Large number of photons per coherence volume can be produced when multimedia
fiber is combined with broadband light source.
To illustrate speckle reduction with a partially spatially coherent source, low-coherence
interferograms of a scattering surface using
single-mode and multimode source fibers was recorded. There is no degradation in SNR and
no measurable mode cross-coupling is
observed in the multimedia fiber. A partially spatially coherent source is preferable not only
for reducing speckle noise but also for
eliminating airy rings in the image caused by the small field diameter of most single-mode
fibers. Superstition of coherent modes reduces
speckle noise by averaging because each mode forms a statistically independent speckle field.
The results of both simulations and
experiments indicatethat broadband light sources with reduced spatial coherence that
provide a large number of photons per
coherence volume may be effectively utilized to reduce speckle in OCT interferograms.
Image needs to be denoised before processing
them for further analysis [2]-[4].
The various type of noise removing
filters evaluated in [5] are mean filter, median filter, order statistic filter and adaptive filter
(BM3D). BM3D and median filters performed
well in removing speckle noise while averaging filters (mean and median filters) are
not efficient in removing speckle noise. BM3D
is the best filter for removing salt and pepper
noise. Wavelet denoising of multi-frame OCT uses the single capture as input instead of the
average of multiple frames as is common in
OCT processing today. The single frames are wavelet transformed and weights are
computed on transformed data. [6] proposed
two different weights. A significant weight
that determines if noise is locally present and a correlation weight thatdetermines if structure
is present within a local neighbourhood. A
combination of these weights is also possible. The wavelet detail coefficients are scaled with
the weights, averaged and transformed back.
This method removes noise more than the median filter, without degrading the structure
of the image [5]-[7].
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
29
[8],[10] proposed that filter
incorporates the advantages of median filter in structure retention and adaptability of Lee or
Frost filter to the local speckle and scene
statistics and is simple and effective like Lee sigma filter for reduction of speckle noise. It
computes simple statistics such as mean,
median, standard deviation. It then determines the speckle noise based on local mean and
standard deviation. Then central speckle noise
is then removed and valid pixels keep their original values. Local adaptive median filter
was compared with other adaptive filters and
the lee sigma filter and it was found to be more effective them in speckle noise reduction and
excellent capability of preserving edges and
details. Classical NL-means filter is most suitable for US imaging [8]-[10].
The method of partial differential equations (PDEs) in image significantly
reduce noise with good preservation of edges.
But PDEs such as SOPDE are good for reducing only Gaussian noise and not speckle
noise. For speckle noise reduction, this method
fails to preserve edges. Hence fourth-order PDEs are combined with various speckle
reduction filters such as SRAD filter, Kaun
filter and Lee filter. Gaussian filters can also reduce speckle noise but it do not preserve the
edges and details well. Neural network
perform non linear processing and construct a desirable input and output system by learning.
It is an important component of pattern
recognition and used in the same and in signal processing in various fields. This functions
recognise the noise and judges the area that a medical doctor dose at the diagnosis.[11]-[12].
Speckle denoising method called Volume-BM3D improves OCT image quality
with OCT volume data. This method improves
the SNR of the image of human fingertip efficiently and outperforms other denoising
methods. This method is a powerful tool in
image enhancement applications [13]. A despeckling technique based on multiple
image reconstruction and selective 3-
dimentional filtering was proposed in [14].
Another method of de-speckling of
SAR images in which at first, smoothing of the
coefficients of the highest wavelet sub-bands
is applied on decomposed wavelet coefficients. A Gaussian low pass filter is used for
decomposing the image. Then, the learning of
a Kohonen self-organising map (SOM) is performed directly on the de-noised image to
remove blur. This method is highly useful in
reducing speckle noise in SAR images. However the whole algorithm is
computationally expensive [15]-[16].
In transform domain techniques firstly
the image is transformed into frequency
domain than it will transform in to spatial domain and then domain specific properties
are exploited to process the image. Different
filtering methodologies like fourier transform,
wavelet transform, wavelet domain noise filtering and thresholding are used. All this
methods have some disadvantages. Wavelet
transform techniques are now used to overcome all this disadvantages. RPCA-based
technique was proposed for the first time in
[18] to reduce speckle noise from OCT images. The decomposition of the OCT image
matrix into a sparse matrix of speckle noise
and a low-rank matrix of de-noised image is a
unique feature of this method which can not only reduce speckle noise but also preserves
the structural information of imaged object.
This method is applied to the post-processing of OCT images to enhance the image quality
[17]-[18].
3. METHODOLOGY
The de-speckling filters are evaluated for efficiency in removing speckle noise from
noisy OCT images. The filters evaluated in
this paper are Mean filter [5], Median filter [5],[8], Frost filter [8], wiener filter[10]-[12],
Anisotropic diffusion filter [18]. These filters
are describe below:
3.1 Mean Filter
Mean filter is one of the widely used
low pass linear filter. It effectively smoothens
the image but is has the disadvantage of smearing the edges and fine features together
with the speckle noise, thus fails to preserve
the image structure. The filter computes the
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
30
average value of the noisy image in a moving
window. Then the center pixel intensity value is replaced by that average value. This process
id repeated for all pixels in the.
3.2 Median filter
The median filter removes pulse and
spike noise successfully while retaining ramp and step function. Therefore median filter is
better than mean filter in terms of preserving
edges between two different features but it doesn’t preserve single pixel-wide features,
which will be altered if speckle noise is
present.The 3*3 median filterpreserves the texture informationbut it doesn’t give the mean
value. The use of mean and filter is limited in
speckle noise reduction because of the
multiplicative nature of the speckle noise which means that amount of noise is
proportional to the intensity of the signal.
Mean and median filters are not adaptive filters. They do not consider any particular
speckle properties of the image.
3.3 Frost filter
The frost filter replaces the pixels of
interest by weighted sum of the values of all pixels within the moving window. This filter
was developed by Frost et al., in 1981-1982.
The weighting factors decreases with distance from the pixel of interest and increases for the
central pixels as variance within the window
increases. Multiplicative noise and stationary noise statistics are assumed by this type of
filter.
3.4 Anisotropic diffusion filter
Diffusion means balancing the
difference in concentration of an image.
Anisotropic diffusion is a generalization of this diffusion process. It performs two diffusion
processes: edge-enhancing anisotropic
diffusion and coherence-enhancing anisotropic diffusion. Anisotropic diffusion filter extracts
a family of multi-resulution derived images in
order to identify global objects through blurring. These filters are mostly used in
segmentation and edge detection, as they are
good in preserving edges while smoothing the
other regions of the image.
3.5 Wiener filter
There are two methods for
implementing wiener filter : Fourier-transform
method (frequency-domain) and mean-squared error (spatial-domain). The former method is
used for denoising and deblurring and the later
method is used only for denoising. In Fourier transform method of Wiener filtering,
normally it requires a priori knowledge of the
power spectra of noise and the original image. But in mean-squared method no such a priori
knowledge is required. Hence, it is easier to
use the mean-squared method for image
denoising. Wiener filter works on the principle of least-square, i.e., the filter minimizes the
MSE between the actual output and the desired
output. Wiener filter assumes both global and local statistics as image statistics vary largely
from one region to another within the same
image. Wiener filter estimates the original data with minimum MSE and hence the noise
power in the filtered output is least.
These filters are evaluated quantitatively in terms of Mean Squared Error (MSE), Signal
to noise ratio (SNR), Peak signal to noise ratio
(PSNR), correlation, homogeneity and Energy. In addition to these quantitative measures,
qualitative visual inspection is also used to
compare results of various filters mentioned above. Fig. 1 below shows an OCT image
affected with speckle noise and its
monochromatic image.
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
31
Fig 1.OCT image with speckle noise (colour and monochromatic)
4. RESULTS
The outputs of the various filters used on speckle affected OCT images are shown in figure 2 below:
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
32
Fig 2: Output of various speckle reduction filters. (a) Mean filter 3*3, (b) Mean filter 5*5, (c) Mean
filter 7*7, (d) Median filter 3*3, (e) Median filter 5*5, (f) Median filter 7*7, (g) Wiener filter 3*3, (h) Wiener filter 5*5, (i) Wiener filter 7*7(j) Anisotropic filter (k) Frost filter
After filtering the noisy images, quantitative statistics were calculated. The MSE, SNR and PSNR of the filters are shown in table 1.
Table 1. Performance factors of various filters
Filter MSE SNR PSNR Correlation Homogeneity Energy
Mean (3*3) 0.0038 76.6442 76.6765 0.9674 0.889 0.1879
Mean (5*5) 0.0076 69.9243 69.2546 0.983 0.9343 0.2188
Mean (7*7) 0.0100 68.7523 68.7645 0.9883 0.9543 0.2329
Median(3*3) 0.0101 65.7553 65.6474 0.9844 0.899 0.1993
Median (5*5) 0.0102 65.7776 65.8753 0.9848 0.9212 0.2044
Median (7*7) 0.0110 65.8753 65.9836 0.9901 0.9311 0.2143
Anisotropic 0.0075 69.8882 69.2785 0.924 0.914 0.203
Frost 0.0034 77.6671 76.64532 0.97 0.88 0.199
Wiener(5*5) 1.1*10-7
95 95 0.98 0.94 0.225
As seen in the table and plots, wiener
filter shows comparatively greater value of SNR and PSNR value than other filters. So
comparatively it is a better speckle removing
filter than other filters evaluated here. In addition to this quantitative measure,
qualitative visual inspection is also used to
compare results of the different filters evaluated. Mean and Median filters just
averaged the image. Frost filter removed the
speckle noise. However it was unsuccessful in preserving the edges and fine details.
Anisotropic diffusion filter de-speckled the
noisy image quite effectively, but poorer than wiener filter as it blurs the image.
Not restricting to Error and SNR, also taking correlation, homogeneity and energy
into consideration, it could be observed from
the table that Wiener filter seems to be optimistic by all means. It could be seen that
the wiener filter not only reduces the speckle
noise but also preserves required edge information thereby easing the process of
further analysis and classification.
5. CONCLUSION
Presence of speckle noise as grainy
salt-and-pepper patterns in OCT images
degrades the image quality and further
complicates the process of future tasks. In order to take full advantage of the OCT
imaging in the field of ophthalmology and
other imaging modalities for disease diagnosis, speckle noise has to be removed. In this paper
we evaluated some of the spatial domain
speckle reduction filters for OCT images. During the experiments, quantitative measures
like MSE, SNR, PSNR, homogeneity,
correlation, and energy were used to compare the denoising capability of various filters. Both
quantitative and visual inspection shows that
wiener filters were comparatively more effective in reducing speckle noise and also
successful in preserving the edges. Mean and
median filters performed worst. Frost and anisotropic diffusion filter reduced speckle
noise but could not preserve the edges and fine
details well. The result also indicates that different filters have different advantages over
another and the choice of particular filter may
depend on the type of use (Eg. enhancement for visual inspection or preprocessing for
segmentation of most prominent features in the
image). Most of the filters have smoothened the image with blurring effect.
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
33
REFERENCES
[1].Venkatraman. K, M.Sumathi., “Review of
optical coherence tomography image analysis
for retinal disorders”, International journal of Pharmacy and Technology, 2014, Vol. 6, no.
2, pp. 10.
[2]. Markus A.Mayer, AnjaBorsdorf, Martin Wagner, JoachinHornegger, Christian
Y.Mardin, Ralg P. Tornov., “Wavelet
denoising of multiframe optical coherence tomography data”, Optical Society of America,
2012, pp. 3-7.
[3]. L. Ramnath, G. Moreno, H. Mueller, T. Bonin, G. Huettmann, A. Schweikard.,
“Towards Multi-Directional OCT for speckle
Noise Reduction”, Springer-Verleg Berlin Heidelberg, 2008, pp. 1-4.
[4]. C.H KranthiRekha., “Speckle Noise
Reduction using Hybrid Image analysis Scheme-Review”, International Journal of
Science, Engineering and Technology
Research (IJSETR), 2013, Vol. 2, no. 1, pp. 2-4.
[5]. Jeehyun Kim, Donald T.Miller, Eunha
Kim, Sanghoon Oh, Junghwan Oh, Thomas Milner., “Optical coherence tomography
speckle reduction by a partially spatially
coherent source”, Journal of Biomedical Optics, 2005, pp.1.
[6]. RohitVerma, Jahid Ali., “A comparative
Study of Various Types of Image Noise and Efficient Noise Removal Technique”,
International Journal of Science, Engineering
and Technology Research (IJSETR), 2013, Vol. 3, no. 10, pp. 1-6.
[7]. Fang Qiu, Judith Berglund, John R. Jensen, Pathik Thakkar, Dianwei Ren.,
“Speckle Noise Reduction in SAR Imagery
Using a Local Adaptive Median Filter”, V.H. Winston & on, Inc. All rights reserved, 2004,
pp. 3-6.
[8]. Pierrick Coupe, Pierre Hellier, Charles Kervann, Christian Barillot., “Nonlocal
Means-Based Speckle Filtering for Ultrasound
Images”, IEEE Transactions on Image Processing, 2009, Vol. 18, No. 10, pp. 1,8-9.
[9]. Nezry E., Mougin E., Lopes A.,
GastelluEtchgorry J.P., Y. Laumonier., “Tropical Vegetation Mapping with Combined
Visible and SAR Spaceborne Data”,
International Journal of Remote Sensing,
1993. [10]. Jenny Rajan, K Kannan, M.R Kaimal.,
“An improved hybrid model for molecular
image denoising”, springer science+businessmedia,LLC, 2008.
[11]. C.H KranthiRekha, V.Vijaya., “Fourth
order PDE and adaptive filters for speckle noise reduction”, RASCET, Kerela, conference
proceedings, 2010.
[12]. Longzhi Wang, ZhuoMeng, X. Steve Yao, Tiegen Liu, Mingliang Qin., “Adaptive
speckle reduction in OCT volume data based
on block-matching and 3D filtering”, IEEE photonics technology letters, 2012, Vol. 24,
No. 20, pp. 1-2.
[13]. Mahboob Iqbal, Jie Chen, Wei Yang, Penbo Wang, Bing Sun., “SAR image
despeckling by selective 3D filtering of
multiple compressive reonstructed images”, Progress in Electromagnetics Research, 2013,
Vol. 134.
[14]. Mohammad R.N Avanaki, Ramona Cernat, Paul J. Tadrous, TaranTatla,
AdriahGh. Podoleanu, S. Ali Hojjatoleslami.,
“Spatial compounding algorithm for speckle reduction of dynamic focus OCT images”, IEE
photonics technology letters, 2013,Vol. 25,
No. 15. [15]. AshkanMasoomi, RoozbehHamzehyan,
NajmehCheaghiShirazi,.“Speckle reduction
approach for SAR image in satellite communication”, International Journal of
Machine Learning and Computing, 2012, Vol.
2, No. 1. [16]. KritikaBansali, Akwinder Kaur,. “A
review on speckle reduction techniques”, IOSR Journal of Computer Engineering(IOSR-JCE),
2014.
[17]. F. Luan, Y Wu., “Application of RPCA in optical coherence tomography for speckle
noise reduction”, Laser Physics Letters 10,
2013. [18]. A. Stella, Bhusan Trivedi.,
“Implementation of some speckle filters on
digital image and OCT image”, International Journal of Advanced Research in Computer
Science and Software Engineering, 2013,Vol.
3, No. 6.
34
IMPULSE NOISE REDUCTION FROM MAMMOGRAM IMAGES
USING NOVEL KT FILTERS
R. R. Subash Chandra Boss1, K. Thangavel
2
1,2 Department of Computer Science,
Periyar University, Salem
Abstract
One of the vital challenges in the field of image processing and computer vision is image denoising,
where the essential objective is to estimate the original image by suppressing noise from a noise-contaminated
version of the image. Image noise may be caused by different intrinsic (i.e., sensor) and extrinsic (i.e.,
environment) conditions which are often not possible to avoid in practical situations. Therefore, image
denoising plays an important role in a wide range of applications such as image restoration, visual tracking, image registration, image segmentation, and image classification, where obtaining the original image content is
crucial for strong performance. While many algorithms have been proposed for the purpose of image denoising,
the problem of image noise suppression remains an open challenge, especially in situations where the images are
acquired under poor conditions where the noise level is very high. This paper proposes different filtering
techniques based on frequency methods for the removal of impulse noise. The proposed filtering model is
validated by experimental analysis. In this study, the statistical quantity measures such as Signal-to-Noise Ratio
(SNR), Peak Signal-to-Noise Ratio (PSNR), and Root Mean Square Error (RMSE) are used to measure the
quality of enhanced images.
Keywords: Mammogram, SWT, Pre-processing, Impulse noise, PSNR, SNR and RMSE
1. Introduction
Breast cancer cases are spiraling world
over, and urban India is no exception. India’s
National Health Profile 2010 predicts that by 2020,
breast cancer will overtake cervical cancer as the
most common type of cancer among women in
India. Dr Rajni Mutneja, head of preventive
oncology at Rajiv Gandhi Cancer Institute, Delhi,
stated that almost one in 20 women in metropolitan cities are suffering from breast cancer and cases
have almost doubled in the last decade [1]. Indian
Council for Medical Research (ICMR) reported
that breast cancer has nearly doubled in the last 24
years. As per the International Agency for Research
on Cancer (IARC), India could see around 250,000
new cases of breast cancer by 2015 [2].
Medical imaging is the technique and
process used to create images of the human body
for clinical purposes (medical procedures seeking to reveal, diagnose, or examine disease) or medical
science (including the study of normal anatomy and
physiology). It is widely acclaimed as a hallmark of
modern medicine. Diagnostic imaging is an
umbrella term for a wide variety of scans,
examinations and images that are used in the field
of medicine such as X-ray, Computed Tomography
(CT), Magnetic Resonance Imaging (MRI),
Ultrasound (US) and Dermoscopy Images (DI) [3].
There is no doubt that medical imaging solution
technology plays a vital role in the diagnosis and
treatment of patients suffering from serious illness.
Unfortunately, medical images encounter a various
number of noises such as Gaussian, Poisson, Rician
and impulse noise (salt and pepper noise) [4]. This
study focuses on impulse noise as medical images
are mainly prone to this type of noise [5].
Digital mammograms are medical images
that are complicated to be interpreted, thus a preparation phase is needed in order to improve the
image quality and make the segmentation results
more accurate. The main objective of the digital
image pre-processing step is to improve the quality
of the image to make it ready to further processing
by removing the isolated and surplus parts in the
back ground of the mammogram [6, 5].
This paper is organized as follows: Section
2 describes the related work. Section 3 discusses
the Stationary Wavelet Transform. Section 4 explains the median filter techniques and M3 filter
techniques. Section 5 discusses the proposed
methods. Section 6 discusses experimental results
and Section 7 covers conclusion.
2. Related work
Impulse noise is frequently encountered in
acquisition, transmission, storage and processing of
images. The presence of impulse noise in an image
may be either relatively high or low. Thus, it could
severely degrade the image quality and cause some loss of image information details. Filtering a digital
image to remove noise and keeping the image
details is an essential part of this task. Various
35
filtering techniques have been proposed for
removing impulse noise in the literature: The
wavelet transform has become an important tool for
this problem due to its energy compaction property
[7]. Indeed, wavelets provide a framework for
signal decomposition in the form of a sequence of signals known as approximation signals with
decreasing resolution supplemented by a
sequence of additional touches called details
[8,9]. Denoising or estimation of functions,
involves reconstituting the signal as well as
possible on the basis of the observations of a
useful signal corrupted by noise [10-13]. The
methods based on wavelet representations yield
very simple algorithms that are often more
powerful and easy to work with than traditional
methods of function estimation [14].
Wang L et. al, proposed wavelet
filters[15,16]. Osman et. al., presented the image
noises are eliminated using the Stationary Wavelet
Transform, SWT, with time invariant characteristic
which is particularly useful in image denoising
[17]. Hasan Demirel et al., proposed an image
resolution enhancement technique based on
interpolation of the high frequency subband images
obtained by discrete wavelet transform (DWT) and
the input image. The edges are enhanced by
introducing an intermediate stage by using stationary wavelet transform (SWT) [18].
Hence, many models have been proposed
for noise removal from medical images. While
some of these models use complicated
formulations, others require deep knowledge about
image noise factors. Hence, a simple noise
reduction method that removes noise well and
preserves image details without relying on image
noise factors is desirable. The proposed methods
which are explained in section 5 removes any level
of impulse noise, is applicable for almost all noise models, these models do not use
complicated formulations and not required deep
knowledge on image noise factors.
3. Stationary wavelet method
Wavelet Transform is superior approach
to other time-frequency analysis tools like
Fourier Transform (FT) and Short Term Fourier
Transform (STFT) because its time scales width
of the window can be restricted to match the original signal especially in image processing
applications. This makes that it is particularly
useful for non-stationary signal analysis such as
noises and transients. For discrete signal, DWT
is a Multi-resolution Analysis (MRA) and it is
a non-redundant decomposition. The drawback of
non-redundant transform is their non-variance in
time [19]. The stationary wavelet transform (SWT)
was introduced in 1996 to make the wavelet
decomposition in time invariant [20, 21]. In order
to preserve the invariance by translation, the down-sampling operations must suppressed and
the decomposition obtained is redundant and is
called stationary wavelet transform, which is as
shown in figure 1.
Figure 1. Stationary Wavelet Transform
SWT has similar tree structured
implementation without any sub-sampling. This
balance of Perfect Reconstruction (PR) is
preserved through level dependent zero padding
interpolation of respective low pass and high
pass filters in the filter bank structure. SWT has
equal length of wavelet coefficients at each level.
The computational complexity of SWT is high as
compared to Discrete Wavelet Transform and need
larger storage space. And which is represented as O (n2). The redundant representation of SWT makes
shift-invariant and suitable for applications such
as edge detection, de-noising and data fusion
[22]. In stationary wavelet transform (SWT)
instead of down sampling, an up sampling
procedure is carried out before we separate the
variables x and y of image f (x, y) shown in the
following wavelets:
Vertical wavelet
(LH):
)()(),(1 yxyx
Horizontal wavelet (HL):
)()(),(2 yxyx
Diagonal wavelet (HH):
)()(),(3 yxyx
36
x+1, y
x-1, y-1 x, y-1 x+1, y-1
x-1, y x, y
x-1, y+1 x, y+1 x+1, y+1
wavelet is function and is the scaling
function. The detailed signals contained in the three
sub-images are as follows:
ly
kjlx
yxj lylxclyhlxgkkw j ),()()(),(2,
1
1
ly
kjlx
yxj lylxclyhlxgkkw j ),()()(),(2,
2
1
ly
kjlx
yxj lylxclyhlxgkkw j ),()()(),(2,
3
1
4. Median filter
It is possible to filter out the noise present
in image using filtering. A high pass filter
passes the frequent changes in the gray level
and a low pass filter reduces the frequent
changes in the gray level of an image. That is the
low pass filter smoothes and often removes the sharp edges. A special type of low pass filter is the
Median filter. The Median filter takes an area
of image (3 x 3, 5 x 5, 7 x 7 etc), observes all
pixel values in that area and puts it into the array
called element array. Then, the element array is
sorted and the median value of the element
array is found out. It has been achieved this by
sorting the array in ascending order using Bubble
sort. The middle in the sorted array is taken as the
median. The output image array is the set of all the
median values of the element arrays obtained for all
the pixels. Median filter goes into a series of loops which cover the entire image array [23].
Some of the important features of the
Median filter are: It is a non-linear digital filtering
technique. It works on a monochrome color
image. It reduces “impulse” and “salt and paper”
noise. It is easy to change the size of the Median
filter. It removes noise in image, but adds small
changes in noise-free parts of image. It does not
require convolution. Its edge preserving nature
makes it useful in many cases. The selected median
value will be exactly equal to one of the existing brightness value, so that no round-off error is
involved when we take independently with
integer brightness values comparing to the other
filters [23, 24].
M3- Filter
The M3-Filter is a hybridization of mean
and median filter [25]. This replaces the central
pixel by the maximum value of mean and median
for each sub images SXY. It is expressed as M3-
Filter, the intensity values are reduced in the
adjacent pixel and it preserves the high frequency
components in image. Therefore it may be suitable
for denoising the speckle noise in the ultrasound
medical image. It is a simple, intuitive and easy to
implement method of smoothing images. )}),({)},,({max(),(
),(),(tsgmeantsgmedianyxf
xyxy stssts
5. Kernel Threshold (KT) based Filtering
Techniques
A novel filter called KT filter is proposed
which is the enhancement of M3 filter. M3 filter
considers all the pixels in sub image but in this
method we consider twelve different schemes of
sub images. The figure 3 shows the KT Filter scheme models. This method replaces the central
pixel by the maximum value of mean and median
for each sub images SXY. It is expressed as KT-
Filter, the intensity values are reduced in the
adjacent pixel and it preserves the high frequency
components in image. Therefore it may be suitable
for denoising the impulse noise in the mammogram
medical image. It is a simple, intuitive and easy
method for smoothing images. The figure 2 shows
the Sub-image pixel co-ordination.
Figure 2. Sub-image pixel co-ordination
Scheme Models
KT1 KT2 KT3 KT4
KT5 KT6 KT7 KT8
37
KT9 KT10 KT11 KT12
Figure 3. KT Filter Scheme Models
6. Experimental and results
The proposed methods have been
implemented using MATLAB. The mammogram
images used for experimental analysis are taken
from the Mammographic Image Analysis Society
(MIAS). This corpus consists of 322 images, which
belong to three big categories: normal, benign and
malign. There are 208 normal images, 63 benign
and 51 malign, which are considered abnormal.
The sub-images of size 3 x 3 are used for
constructing KT filter. The important factors for
performance evaluation of the reconstruction algorithm are speed and the quality of the
reconstructed images. While the system cost and
speed are dependent on many factors including the
advancement in technology and the process in
hardware and software implementation, the image
quality is a technology-independent measure.
Image quality can be used to compare the
performance of the different system and select the
appropriate processing algorithm for a given
application. Hence, image quality assessment plays
an important role in image processing application. A great deal of effort has been made in
recent years, and several image quality matrices
have been developed in addition to visual analysis,
to predict the visible different between a pair of
image, the input image I(x,y) and the resultant
image I′(x,y). The figure 4 shows the proposed
system.
Different measures [26, 27] are used to
compare the performance of the filtering methods.
For quantitative evaluation, there common error
measurements, viz., MSE, RMSE, SNR and PSNR
are widely employed [28, 29]. The MSE, RMSE,
SNR and PSNR values for the Median, M3 and KT
Filters overall performance of statistical
measurements are tabulated in table1. It is observed
that the KT1 filter gives the best result for all tested
image. Figure 5 shows that the proposed method
results in images MDB030.
Figure 4. Proposed system
Median PSNR : 39.14461 M3 PSNR : 38.71739 KT1 PSNR : 46.90071 KT2 PSNR : 28.59734
KT3 PSNR : 38.68543 KT4 PSNR 38.69705 KT5 PSNR : 40.49473 KT6 PSNR : 43.86343
Origi
nal
image
KT Filter
Inver
se
Statio
nary
wavel
et
De-
noise
d
image
Statio
nary
wavel
et
38
KT7 PSNR : 35.9684 KT8 PSNR : 38.64685 KT9 PSNR : 34.64424 KT10 PSNR : 34.26017
KT11 PSNR : 33.43632 KT12 PSNR : 34.17703
Figure 5. Proposed method result in image MDB030
Table1. Overall performance of statistical measurements
for MIAS database
7. Conclusion
In this work, KT filter method is proposed as
pre-processing for mammogram images. Mammogram
images are susceptible to impulse noise, which results in inaccurate analysis that can prove
disastrous. Median filters and its variants are commonly
used for impulse noise removal, but they result in loss
of image details. Also, for high noise levels, they are
not sufficient in completely removing the noise. Using
our pre-processing method, noise is eliminated
effectively. The performance of the KT Filter is
evaluated using MSE, RMSE, PSNR and SNR
measures and compared with benchmark Median filter
and M3 filter. It was observed that KT1 Filter out
performs the benchmark Median and M3 filter. Further
the resultant mammogram can be used for accurately
detect region of interest (ROI), abnormalities in human
breast like calcification, circumscribed lesions etc.
S.No Filter MSE RMSE PSNR SNR
1 Median
6.881902 2.559121 39.44869
0.05251
3
2 M3
7.124516 2.622721 39.14855
0.05705
4
3 KT1
1.263565 1.108218 46.59112
0.01632
1
4 KT2
53.83242 7.162768 30.47873
0.04952
2
5 KT3
7.131364 2.623537 39.1469
0.04579
5
6 KT4
7.169801 2.632867 39.10955
0.04423
6
7 KT5
4.245671 2.003048 41.61886
0.03129
9
8 KT6
2.317237 1.49718 43.98814
0.00224
4
9 KT7
14.50898 3.749193 36.03314
0.07207
7
10 KT8 8.115979 2.80445 38.54341
0.066743
11 KT9 18.86642 4.285378 34.83178 0.06562
12 KT10
19.25543 4.332257 34.73163
0.06447
5
13 KT11
26.34454 5.061241 33.40312
0.08659
8
14 KT12
18.25871 4.205617 35.01777
0.08819
9
Statistical
Measurement
Formula
MSE
MN
jiFjif 2)),(),((
RMSE
MN
jiFjif 2)),(),((
SNR 2
2
10log10e
PSNR
RMSE
255log20 10
39
Acknowledgment
The first author immensely acknowledges the
partial financial assistance under University Research
Fellowship, Periyar University, Salem.
The second author immensely acknowledges the UGC, New Delhi for partial financial assistance
under UGC-SAP (DRS) Grant No. F.3-50/2011.
References
[1] http://idiva.com/news-health/breast-cancer-in-urban-india-
doubles-in-24-years/8532
[2] http://www.indiaprwire.com/pressrelease/health-
care/20130119142868.htm
[3] Xue-jun Z, Hong G, Yun-ting Z, Ying L, Li G. Educational
practice of computer science in medical imaging. IEEE
Proceeding. Education Technology and Computer Science. vol.
3. 2010. p. 546-9.
[4] Gravel P, Beaudoin G, De Guise JA. A method for modeling
noise in medical images. IEEE Trans Med Imaging
2004;23:1221-31
[5] Amir Reza Sadri, Maryam Zekri, Saeid Sadri, Niloofar
Gheissari “Impulse Noise Cancellation of Medical Images
Using Wavelet Networks and Median Filters”, Journal of
Medical Signals & Sensors, Vol 2 | Issue 1 | Jan-Mar 2012
[6] Abreu E, Lightstone M, Mitra SK, Arakawa K. A New
efficient approach for the removal of impulse noise from
highly corrupted images. IEEE Trans Image Process 1996;
5:1012-25.
[7] Michel Misiti, Yves Misiti, Georges Oppenheim, Jean-
Michel Poggi, “Wavelets and their Applications”, Published
by ISTE 2007 UK.
[8] C Sidney Burrus, Ramesh A Gopinath, and Haitao Guo,
“Introduction to wavelet and wavelet transforms”, Prentice
Hall1997.S. Mallat, A Wavelet Tour of Signal Processing,
Academic, New York, second edition, 1999.
[9] R. C. Gonzalez and R. Elwood‟s, Digital Image
Processing. Reading, MA: Addison-Wesley, 1993.
[10] M. Sonka,V. Hlavac, R. Boyle Image Processing ,
Analysis , And Machine Vision. Pp10-210 & 646-670
[11] Raghuveer M. Rao., A.S. Bopardikar Wavelet Transforms:
Introduction To Theory And Application Published By
Addison-Wesley 2001 pp1-126
[12] Arthur Jr Weeks , Fundamental of Electronic Image
Processing PHI 2005.
[13] Jaideva Goswami Andrew K. Chan, “Fundamentals Of
Wavelets Theory, Algorithms, And Applications”, John Wiley
Sons
[14] Donoho, D.L. and Johnstone, I.M. (1994) Ideal spatial
adaptation via wavelet shrinkage. Biometrika, 81, 425-455
[15] Karthikeyan K, Chandrasekar C. Speckle Noise Reduction of
Medical Ultrasound Images using Bayesshrink Wavelet
Threshold. Int J Comput Appl 2011; 22:8-14.
[16] Wang L, Lu J, Li Y, Yahagi T, Okamoto T. Noise reduction
using wavelet with application to medical X-ray image.
International Conference on Industrial Technology, vol. 20,
2005. p. 33-8
[17] A. M. Osman , Enhancement Of Photonand Neutron Images by
Denoising Using Stationary Wavelet Transform, Armenian
Journal of Physics, 2011, vol. 4, issue 3, pp. 154-164
[18] Hasan Demirel and Gholamreza Anbarjafari, “Image
Resolution Enhancement by Using Discrete and Stationary
Wavelet Decomposition” IEEE Transactions on Image
Processing, Vol. 20, No. 5, May 2011
[19] Mallat S. G., 1989, "Theory of Multiresoluiton Signal
Decomposition: The Wavelet Representation", IEEE Trans.
On Pattern Analysis and Machine Intelligence, Vol.11. No.
7,PP: 674-693.
[20] Pesquet J.C., Krim H. and Carfantan H., 1996, "Time
invariant orthonormal Wavelet
representation", IEEE Transactions on Signal Processing,
Vol. 44, PP: 1964-1970.
[21] Warin Chumsamrong and et. al., 1998, "Using Stationary
Wavelet Transform in the classification of SAR images",
19th Asian Conf.on Remote Sensing, 16-20.
[22] Sukla P. D., 2003, "Complex Wavelet Transforms and its
Application", M.Phil. Thesis, University of Strathclyde,
Scotland-UK.
[23] R.C. Gonzalez, R.E. Woods, “Digital Image processing”,
Pretice Hall. 2007.
[24] Boss, R. Subash Chandra, K. Thangavel, and D. Arul Pon
Daniel. "Mammogram image segmentation using fuzzy
clustering." In Pattern Recognition, Informatics and Medical
Engineering (PRIME), 2012 International Conference on, pp.
290-295. IEEE, 2012.
[25] K. Thangavel , R. Manavalan and I. Laurence Aroquiaraj,
“Removal of Speckle Noise from Ultrasound Medical Image
based on Special Filters: Comparative Study”, ICGST-GVIP
Journal, ISSN 1687-398X, Volume (9), Issue (III), June 2009
[26] J. Liu and Y.H. yang. Multiresolution Color Image
Segmentation. IEEE Trans. Pattern Anal. Mach. Intell.,
16:689-699,1994.
[27] M. Borsotti, P.Campadelli and R. Schettini. Quantitative
Evolution of Color Image segmentation Results. Pattern
Recogn. Lett, 19:741-747, 1998.
[28] K. Kanjanawanishkul and B. Uyyanonvara, Fast adaptive
Algorithm for time Critical Color quantization Application . In
C. Sun, H. Talpot, S. Ourselin, and T.Adriaansen editors, Proc.
VII th digital Image Computing: Techniques and Application
, pages 781-785, Sydney, December 2003.
[29] G. McGibney and M.R. Simith. A
[30] n Unbiased Signal- to – Noise ratio measure for MRI
Images. Med. Phy., 20: 1077-1078, 1993.
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
40
DETECTION AND EXTRACTION OF BLOOD VESSEL OF RETINAL IMAGES
IN DIABETIC RETINOPATHY USING FILTERS
Subhra Pattnaik, Asit Subudhi, Sunita Sarangi, Sukanta Sabut* School of Electronics Engineering
Institute of Technical Education & Research SOA University, Bhubaneswar, Odisha
ABSTRACT
Diabetic retinopathy is the result of damage to the tiny blood vessels that nourish the retina. They leak blood and other flu ids that cause swelling of retinal tissue and clouding of vision. The condition usually affects both eyes. The longer a person has diabetes, the more likely they will develop diabetic retinopathy. Diabetic Retinopathy (DR) is a medical condition that affects the microvasculature of the retina which leads to loss normal vision of a human. It is the most common causes of blindness in adult group of both developed and developing countries. Early detection of DR helps the ophthalmologists to advise proper treatment to save the vision of the patient. The change in appearance of retinal blood vessel structures can be detected by using edge detection and matched filter techniques. Segmentation of vascular structures of retina for implementation of Clinical diabetic retinopathy decision making systems is presented in this paper. The Canny edge detector, matched filter and matched filter with first order derivative of Gaussian filters have been implemented to find the abnormalities in diabetic retinal blood vessels and the results were compared by evaluated parameters between the original affected image and the output images of the work. The Peak Signal-to-Noise Ratio (PSNR) and Mean Square Error (MSE) values are computed for quantitative evaluation of the operators. The simulated result of segmented images shows that the vessel ext raction by MF-FDOG outperforms compared to MF and canny operators in detecting even small blood vessels of retina. The calculated PSNR value is also high for the MF-FDOG than the matched filter and Canny operator, it indicates the better performance of MF-DOG in extraction of blood vessels of DR retinal images.
Keywords: Diabetic retinopathy, retinal images, blood vessels, matched filter, MF-FDOG, Canny edge detector, PSNR, MSE
I. INTRODUCTION Diabetes is a disease that interferes with the body's ability
to use and store sugar, which can cause many health
problems. Diabetic retinopathy (DR) affects human eyes.
The blood vessels in a human are very small in size and
hence more susceptible to DR. The corrosion of blood
vessels starts occurring when the blood sugar levels are increased much above the normal levels for a prolonged
period of time. The Indian Council of Medical Research
estimated that around 65.1 million patients are affected by
diabetes, and by 2030, India will have 101.2 million
diabetic people [1]. Diabetic retinopathy is a vascular
disorder that affects the microvasculature of the retina. It
has been one of the foremost causes of blindness in the
working age group of both developed and developing
countries. About 33% of patients with diabetes have signs
of diabetic retinopathy. It damages the small blood vessel
of the retina, which leads to loss of vision [2]. Current research indicates that at least 90% of new diabetic
retinopathy cases could be monitored properly at the early
stages [3]. Artery-vein crossings and patterns of small
vessels can serve as a diagnostic indicator of diabetic
retinopathy. An accurate delineation of the boundaries of
blood vessels makes precise measurements of these features
that may be applied for diagnosis, treatment evaluation and
clinical study [4]. The retinal blood vessels can be directly
visualized non-invasively with very high resolution for
most clinical scenarios [5]. Edge detection is a common
starting step in many image processing applications. The
edge in an image provides useful structural information
about object boundaries [6]. Edge detection techniques
significantly reduce the amount of data and filter out useless
information while preserving the important structural
properties in an image.2-D Gabor wavelet and supervised
classification was used for automated segmentation of the
vasculature in retinal images by classifying each image
pixel as vessel or non-vessel, based on the pixel‟s feature
vector [7].Vessel extraction is basically a kind of line
detection problem and many methods have been proposed.
A class of popular approaches to vessel segmentation is
filtering based methods [8, 9] which work by maximizing the response to vessel-like structures. Mathematical
morphology [10] is another type of approach by applying
morphological operators. Trace-based methods [11] map-
out the global network of blood vessels after edge detection
by tracing out the center lines of vessels. Such methods rely
heavily on the result of edge detection. Vessel detection and
tracking by matched Gaussian and Kalman filters was used
with second-order derivative of Gaussian matched filter
which was designed according to the known blood vessel
feature, and used to locate the center point and width of a
vessel [12]. Vessel tracking methods are based on the principle of measuring some local image properties to
outline the vessel points which are further used to trace the
retinal vasculature. Feature extraction and recognition of
vessel structure are both performed simultaneously in these
types of algorithms. A new algorithm was presented which
follows matched filtering approach and is instigated by start
and end points followed by automatic detection of vessel
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
41
boundaries [13]. The classical matched filter (MF) method
has advantages of simplicity and effectiveness among the
various retinal vessel extraction methods, which detects
vessels by filtering and thresholding the original image [8].
Morphological operator was used to extract only the blood
vessel and then the edge of the blood vessels were clearly
detected by applying Laplacian and Gaussian operators
[14]. A multi-scale segmentation function assists in
detecting the retinal vessel structures [15]. The response of
the MF-FDOG method by using both the Matched Filter
(MF) and the First-Order-Derivative of the Gaussian (FDOG) was found to be more accurate in distinguishing
between the retinal vessels‟ structures compared to MF
alone [16]. An improvement in detecting retinal vessel both
qualitatively and quantitatively has been found by the
centre line extraction combining with matched filter
responses [17]. A new technique was proposed where all
together the edge detection, matched filtering and region
growing procedures has been used as a group for the
detection of retinal blood vessels in retinal images [18]. A
novel technique of adaptive local thresholding was
proposed by combining verification-based multithreshold scheme with classification procedures in [19], used for the
detection of vessels in retinal images. A technique used for
the segmentation of blood vessels in retinal images that
combines morphological filters and cross-curvature
evaluation [20]. The blood vessels are extracted from the
retinal images with better PSNR and 96% accuracy by
applying a curvelet transform and curvelet coefficients
[21].
In this paper, we have detected and segmented the
blood vessels by the canny edge detector, matched filter and
matched filter with first derivative of Gaussian (MF-FDOG)
for identifying the structures of the blood vessel in retinal
images, and the results are compared by the evaluating the
quantitative parameters PSNR and MSE between the gray
scaled image and the output image of canny operator and
filters.
II. METHOD
The retinal image is collected from an eye institute with permission from the ophthalmologist. One affected sample image is selected from ten sample retinal images. Blood vessels are extracted for the identification of damages in diabetic retinopathy and the canny operator, matched filter and matched filter with first derivative of Gaussian algorithms are applied to detect the blood vessel structures as per flow diagram. The RGB image is first converted into grayscale to strengthen the appearance of blood vessels. Initially, the original image was scaled to enhance the contrast of the image, and canny edge operators is applied to detect blood vessels by using MATLAB 9.0.Similarly, matched filter kernel is loaded then the original image is converted to gray scale image and Matched filter algorithm
is applied in order to find the threshold and segmented image. Similarly MF-FDOG algorithm is applied to get the segmented and threshold image. The methods of canny, MF and MF-FDOG have been described in segment A, B and C respectively.
Retinal RGB
Image
Gray Scale Image
Enhanced Image
Segmented image by Canny edge, Matched filter, MF-FDOG detection
Extracted Blood Vessel detection
Fig.1. Flow chart of blood vessel extraction of retinal image A. Canny operator
The Canny operator works in a multi-stage process. As it is susceptible to noise present in raw unprocessed image data, the raw images were smoothed out by convolving with a Gaussian filter [22]. Canny uses convolution mask in x and y directions, and after that bring it to compute gradient before a non-maximum suppression process. Computation of the gradient of image f(x, y) is carried out by convolving it with the first derivative of Gaussian masks in x and y directions. Smooth the image with a Gaussian filter to reduce noise and unwanted details and textures. The different processes of canny operator are applied to images and the results were presented in Fig. 2.
g(m,n)= (m,n)*f(m,n) ……………(2.1)
where = exp[-( ) /2 ]
B. Matched Filter
The matched filter is one of the template matching algorithms that are used in the detection of the blood vessels in retinal images and other application as well. The matched filter kernel may be expressed from [8] K(x, y) =
)
…………(2.2)
where L is the length of the vessel segment that has the same orientation, defines the spread of the intensity profile. To be able to detect vessels on all possible orientations, the kernel must be rotated to all possible vessel orientations and the maximum response from the filter bank is registered. Many papers found that rotating by an amount
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
42
of 15 is adequate to detect vessels with an acceptable amount of accuracy which results a filter bank with 12 kernels. The authors of [8] made some experiments on the values of L and and found that the best parameter values were those that gave the maximum response at L=9 and
=2. A Gaussian curve has infinitely long double sided trails; the trails are truncated at u=±3 .
The neighborhood „N‟ is defined such that
N={(u,v)
………………….(2.3)
The corresponding weights in the kernel i (i=1………12
which is the number of kernels) are given by
ki (x,y)=
/)
∀ Pi∈ N ….......................(2.4)
If A: number of points in N, the mean value of kernel is determined as
mi= ... ..............................(2.5)
Thus, the convolution mask used in this algorithm is given by
K´i(x,y)=ki(x,y)-mi, ∀ Pi ∈ N ……..………(2.6) C. MF-FDOG The matched filter is the zero-mean Gaussian filter. The first derivative of Gaussian is FDOG. According to [16] MF detects both vessels and non-vessels whereas MF-FDOG method is very simple and effective in detecting fine vessels that are missed by the MF. The first-order derivative of the Gaussian (FDOG) is G(x,y)=(- / exp(− / ) …………..(2.7) According to the paper in [16] the MF and the FDOG are
merged to form „„MF-FDOG‟‟ for vessel detection. Signal‟s response to the MF is denoted by h. A threshold T is then
applied to h to detect the vessels. The vessels and non-vessel edges can be better distinguished by thresholding their
responses to MF. The response of the input signal to the FDOG is denoted by d. The local mean of d is then calculated and
denoted by Dm. The local mean value of an element in d is defined as the average of its neighboring elements. In contrast,
in the neighborhood of the step edge there are also strong responses in h but the corresponding responses in Dm are very
high. Therefore, the local mean signal Dm can be used to adjust the threshold T to detect the true vessels while removing
the non-vessel edges. In other words, T should depend on Dm. If the magnitude in Dm is low, this implies that a vessel may
appear in the neighborhood, and hence the threshold T applied to h can be small to detect the vessels; if the magnitude in Dm
is high, this implies that some non-vessel edges may appear, and hence the threshold T can be high to suppress the non-
vessel edges. A thresholding scheme has been applied according to paper [16] by using the MF-FDOG for retinal
vessel detection. The threshold is applied to the retinal image‟s
response to MF but the threshold level is adjusted by the image‟s response to FDOG. After filtering the retinal image
with the MF-FDOG filters, two response images, H (by the MF) and D (by the FDOG) are obtained. The local mean image of D is calculated by filtering D with a mean
filter: Dm=D*W ……………. (2.8) Where W is a filter whose elements are all 1/w
2.The
local mean image is then normalized so that each element is within [0, 1]. The normalized image of Dm is denoted by
m. The threshold T is then set as
T= (1+ m).Tc ……...................(2.9) where Tc is the reference threshold and is defined as:
Tc=c. …………………………. (2.10)
: mean value of the response image H, and c is a constant which is set between 2 and 3 according to the paper [16]. By applying T to h, the final vessel map MH is obtained as: MH=1 H(x,y) T(x,y) ………………(2.11)
MH=0 H(x,y)<T(x,y) ………………(2.12) D. Evaluation Parameter The objective measures the PSNR used to evaluate the intensity changes of an image between the original affected image and the output images. The Peak-Signal-to-Noise Ratio (PSNR) can be computed as PSNR=10 ………….. (3.1)
and the Mean Square Error (MSE) is given by
1m n
2
MSE
Ia (i, j) Io (i, j) ..…… (3.2)
mn i 1 j 1
III. RESULTS AND DISCUSSION
An affected diabetic retinopathy image was chosen from
ten sample images collected from an eye institute. The images are preprocessed to extract the features in a more clear and proper manner. Hence lots of research has been carried out. The initial pre-processed methods are applied to attain better enhanced images according to the flow diagram is represented in flow diagram as shown in Figure 1. The canny edge operator was used to the affected DR images and the results are shown in Figures 2(a)–2(b). Similarly, the matched filter was applied to the image to get the output and the results are shown in Figures 3(c)-3(e). Then first derivative of Gaussian was applied to the image to get the segmented output and the results are shown in Figures 4(f)-4(h).Figure 5(i) represents the resultant of MF merged with FDOG. From the output images of canny edge detector, match filter and MF-FDOG it was observed that the MF-FDOG works better than MF and Canny edge detectors for detecting small blood vessels. The PSNR and MSE parameters were calculated between the grayscale images and the output images of canny edge detector, MF,MF-FDOG and the result is represented in Table1. The higher the PSNR value the better will be the performance and lower the MSE value better will be the performance hence the simulation result shows that the blood vessels extracted
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
43
from the retinal images has better PSNR values in the MF-FDOG as compared to MF and Canny operator.
(a) (b) Fig. 2.Simulation results of a diabetic image by canny operator (a) Gray Scale image; (b) Canny Operator.
(c) (d)
(e)
Fig. 3. Simulation results of a diabetic image by matched filter (c) Gray Scale image; (d) matched filtered image; (e) thresholded image at 0.35.
(f) (g)
(h) Fig. 4. Simulation results of an affected image by first derivative of Gaussian (f) gray scale image; (g)FDOG image; (h) thresholded Image at 0.35.
(i)
Fig. 5. (i)Simulation results of an affected image by merging matched filter with first derivative of Gaussian at threshold 0.35. Table 1: The PSNR and MSE results of different operators
Parameters Matched MF-FDOG Canny Filter(MF) Operator
PSNR 58.99 60.52 57.46
MSE 0.091 0.057 0.116
IV. CONCLUSION
Many different techniques could be used for extraction
of blood vessels in DR images. Initially the images are preprocessed to increase the contrast and also to enhance the image so that the vessels can be extracted in a better and efficient manner. In this paper, we have implemented the canny edge detector, matched filter, MF-FDOG filter for extraction of retinal blood vessel. The results were compared for different operators by visualizing the output images and evaluating the quantitative values obtained from input images and extracted output images. The result indicates that all the operators successfully detected the edges of the retinal blood vessels, but the MF-FDOG extraction method performed better compared to MF and canny operator for detecting small blood vessels of the retina in terms of better image quality, high PSNR and low MSE values. Thus, it was concluded that the MF-FDOG is a better choice in detecting the continuous and smooth edges of the blood vessels. It will provide better results for identifying the blockages of blood vessels in the retina for the diagnosis of diabetic retinopathy by ophthalmologists. Further work is necessary to validate the results by classifying the images in classifiers for more number of images.
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
44
Acknowledgment Authors would like to thank the ophthalmologists of the L.V. Prasad Eye Institute, Bhubaneswar, for providing the retinal images that has been used in this work.
References
[1] H. King, R.E. Aubert, and W.H. Herman, “Global burden of
diabetes, 1995-2025 prevalence, numerical estimates and projection”, Diabetes Care, vol. 21, pp.1414-1431, 1998.
[2] S.Wild, G.Roglic, A.Green, R. Sicree, and H. King, “Global
prevalence of diabetes,estimates for the year 2000 and projections for 2030”, Diabetes Care, vol. 27, pp.1047-1053, 2004.
[3] H.R. Taylor, andJ.E. Keeffe, “World blindness: A 21st century
perspective”, British Journal of Ophthalmology, Vol. 85, pp.261-266, 2001.
[4] A.Hoover, V. Kouznetsova, and M. Goldbaum, “Locating blood
vessels in retinal images by piece-wise threshold probing of a matched filter response”, IEEE Trans on Med Imaging, Vol. 19, No. 3, pp.203-210, 2000.
[5] G.Dougherty, “Image analysis in medical imaging: recent advances
in selected example”, Biomedical Imaging and Intervention Journal, Vol. 6, No. 3, pp.1-9, 2010.
[6] M.Ramalho, andK.M. Curtis, “Edge detection using neural network
arbitration, Image processing and its application”, Proceedings of the 5th International Conference on Image Processing and Its Applications, Edinburgh, vol 4, no. 6, pp.514–518, July 1995.
[7] J.V.B. Soares,J.J.G. Leandro, R.M. Cesar, Jelinek, M.J.Cree,
“Retinal Vessel Segmentation Using the 2-D Gabor Wavelet and Supervised Classification”, IEEE Trans Med Imaging, vol. 25, no. 9, sept. 2006.
[8] S.Chaudhuri, S. Chatterjee, N. Katz, M. Nelson, M. Goldbaum, “
Detection of blood vessels in retinal images using two-dimensional matched filters”, IEEE Trans. Med. Imaging, vol. 8 no. 3, pp. 263– 269, Sept. 1989.
[9] M. Al-Rawi, M.Qutaishat, M.Arrar, “An improved matched filter for
blood vessel detecetion of digital retinal images”, Comput. Biol. Med., vol. 37, no. 2, pp. 262-267,2007.
[10] M.M.Fraz, M.Y.Javed, A. Basit, “Evaluation of retinal vessel
segmentation methodologies based on combination of vessel
centerlines and morphological processing”, IEEE Int. Conf. on Emerging Technologies, pp. 18–19, Oct. 2008.
[11] Y.Tolias, S. Panas, “A fuzzy vessel tracking algorithm for retinal images based on fuzzy clustering”, IEEE Trans. Med. Imaging, vol. 17, no. 2, pp. 263-273, 1998.
[12] Chutatape, Liu Zheng, and S. M.Krishnan, “Retinal blood vessel
detection and tracking by matched gaussian and kalman filters”, Int. Conf. of the IEEE Eng. in Medicine and Biology Society, vol. 20, no. 6,1998.
[13] L.Zheu, M.S. Rzeszotarski, L.J.Singerman, J.M. Chokerff, “The
detection and qualification of retinopathy using digital angiograms”, IEEE Trans. Med Imaging, 1994.
[14] N.K.E.Abbadi, E.H.A. Saadi, “Blood vessels extraction using
mathematical morphology”, Journal of Computer Science, vol. 9, no. 10, 2013.
[15] E. Moghimirad, S.H. Rezatofighi andH.Soltanian-Zadeh, “Retinal
vessel segmentation using a multi-scale medialness functions”, Computers in Biology and Medicine, vol. 42, no. 1, pp.50–60, 2012.
[16] B.Zhang, L. Zhang andF. Karray, “Retinal vessel extraction by
matched filter with first-order derivative of Gaussian”, Computers in Biology and Medicine, vol. 40, no. 4, pp. 438-445, 2010.
[17] M. Sofka and C.V. Stewart, “Retinal vessel centerline extraction
using multiscale matched filters, confidence and edge measures”, IEEE Trans on Med Imaging, vol. 25, no. 12, pp.1531–1546, 2006.
[18] Y.Wang,Sc.Lee, “A Fast Method for automated detection of Blood
vessels in Retinal Images”, IEEE Comput. Soc. Proc. Asilomer Conf., pp. 1700-1704,1998.
[19] X.Jiang, D.Mojon, “Adaptive local thresholding by verification based
multithreshold probing with application to vessel detection in retinal images”, IEEE Transactions on Pattern Analalysis and Machine Intelligence, vol. 25, no. 1, pp.131–137, Jan. 2003.
[20] F. Zana, J.-C.Klein, “Segmentation of vessel-like patterns using
mathematical morphology and curvature evaluation”, IEEE Trans Med Imaging, vol. 10, no. 7, pp. 1110–1119, 2001.
[21] K.Jeyasri, P. Subathra,K. Annaram, “Detection of Retinal Blood
Vessels for Disease Diagnosis”, Int Journal of Advanced Research in Computer Science and Software Eng., vol. 3, no. 3, March 2013.
[22] J.Canny, “A computational approach to edge detection”, IEEE
Transactions on Pattern Analysis and Machine Intelligence, vol. 8, no. 6, pp.679-698, Nov. 1986.
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
45
A NOVEL EARLIER STAGE DETECTION SCHEME FOR GLAUCOMA
DISEASE USING AUTOMATED CLASSIFICATION METHOD
M.Kirthikavathi 1*
,M.Mahendraselvi 2*
S.Veluchamy3*
1,2PG Scholar of Electronics and Communication Engineering,Anna University Regional Office,Madurai 7,Tamil
Nadu, India.
3Faculty of Electronics and Communication Engineering, Anna University Regional Office, Madurai 7, Tamil
Nadu, India.
ABSTRACT Glaucoma is an eye disease which causes increased intraocular pressure (IOP) within the eye. This increased IOP in
the eye is usually caused by an imbalance production of aqueous humor in the eye which leads to vision loss. Almost 14 million people have glaucoma. There are no prior symptoms of glaucoma and therefore, early detection of the disease is essential for
preventing one of the most common causes of blindness. Glaucoma detection based on digital images of the retina has been performed in the past few years in the clinics but they still lack robust automated assistance. So there is a need for an automated system that detects glaucomatous eyes and classifies the glaucomatous stages based on acquiring fundus images. In contrast to other approaches, certain higher order cumulant features are extracted using radon transform and by using Linear Discriminant Analysis dimensionality of the extracted feature are reduced. Glaucoma stages are classified using the SVM classifier.
Keywords: Glaucoma, Radon transforms, LDA
I. INTRODUCTION
Glaucoma is a complicated disease that causes
increased fluid pressure inside the eye. This increased
fluid pressure (called intraocular pressure IOP) causes
damage to the optic nerve leads to permanent vision loss. Glaucoma is the second leading cause of
blindness. It is estimated that in India about 11.2
million people are suffering from glaucoma over the
age of 40. And it is surveyed that, in 2020
approximately 79.6 million people are diagnosed with
glaucoma. It has been reported that, people in the age
group of between 40 and 50 had elevated IOP of 2%
and people with an age of above 70 had elevated IOP
of about 8% which leads to blindness. Progression of
glaucoma is not detected until the optic nerve gets fully
damaged. So glaucoma screening at regular interval is necessary. It is recommended that persons at the age of
40 - 64 must undergo a regular screening of glaucoma
in every 2to 4 years and people at the age of above 65
are must undergo glaucoma screening for once in
every 2 years. Glaucoma detection based on manual
diagnosis system such as tonometry, ophthalmoscopy,
perimetry, gonioscopy, and Pachymetry may require
more manpower and more time. Medical management,
drainage implants, laser surgery and trabeculectomy.
Mass screening of patients helps for early stage
detection of glaucoma and helps to prevent patients
from surgery. A central reason to hope for automated glaucoma diagnosis is that it could be used as a tool for
mass screening and also minimizing the required
manpower and time required to test large number of
people.
A. RELATED RESEARCH WORK
Siamak Yousefi et al. Proposed method for
glaucoma using Optic nerve (ONH) features. From his
method Optic nerve head’s (ONH) geometric
parameters are also used to measure the progression of glaucoma. The geometric parameters of optic nerve
head measure any changes in the structure of ONH such
as the area of the optic disk (OD), the diameter of the
optic disk, area of the rim, cup diameter and mean cup
depth. High IOP, astrocytes and optic nerve fiber
deterioration are the main characteristics of glaucoma.
Optic nerve fiber deterioration causes decrease in
thickness of the retinal nerve fiber layer (RNLF).
Astrocytes and axon degeneration changes the ONH
configuration and this leads to the decrease in functional
capability of the retina. Stereo fundus images are mostly
used by ophthalmologists to measure the rim, disk diameter [1].
Sushma G.Thorat et al. uses digital fundus images
and proposed that digital fundus image analysis is very
useful in understanding the natural development of
disease which is based on computational technique to
make a qualitative assessment of the eye. These
methods reduce the inter and intra observer variability
errors which are arising during the screening of disease
by the doctors [2].
Sandra Morales et al. proposed glaucoma detection
method based on morphological features. Different morphological features such as cup, disk, disk diameter,
rim area and cup to disk ratio (C/D) obtained from a
fundus image can help to treat glaucoma [3]. C/D ratio
is one of the key parameters used by ophthalmologists
while diagnosing glaucoma. A normal OD contains
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
46
greater than 1.2 million fibers passing through it which
makes the cup size very small but glaucoma causes
ONH to lose the RNLF resulting in an increase in the
size of the cup. Progression of glaucoma leads to
enlarging the size of the cup this in turn leads to
increase in cup-to-disk ratio [6]. Depending upon the C/D ratio glaucoma can be
classified into three stages, namely Mild glaucoma,
Moderate glaucoma and severe glaucoma. At mild
glaucoma stage the C/D ratio is normally 0.4. Moderate
glaucoma have C/D ratio of about 0.7 and severe
glaucoma have C/D ratio of more than 0.7[4].Since
visual fields [1] and IOP measurement are not suitable
for mass screening, OD detection plays a major role in
diagnosing glaucoma. Geometric parameter
model[1],super pixel algorithm[2],mathematical
morphological method[3],wavelet based method [5]
fuzzy convergence[11],the contour model based approach[9]template matching[12],deformable models
and hough transform [8] are the few techniques used
automatic detection of ONH. Haralick texture [4],
fractal [8] and higher order cumulant [HOS] [7] features
are also used for glaucoma diagnosis.
This paper is organized as follows. The proposed
system is explained in section II, including the
preprocessing steps, feature extraction by radon
transform and dimensionality reduction by LDA.
Section III describes the results and discussion about
proposed work and section IV and V deals with conclusion and feature work respectively.
(a) (b)
(c) (
d)
Fig. 1.Fundus images: (a) normal, (b) mild glaucoma, (c) moderate
glaucoma (d) severe glaucoma
II. PROPOSED WORK
A. DATA COLLECTION
Human fundus images are the database used here.
Fundus images are obtained from the digital fundus
camera and are taken by the well trained
ophthalmologists. Fundus cameras are used to exam the
retinal disease. A modified digital back unit is
connected to the fundus camera to convert the fundus
image into a digital image. These digital images are processed and stored on the hard drive of a Windows
based computer with a resolution of 768 x 576 in JPEG
format. This consists of 8-bits of RGB layers with 256
levels each. The images are linked to the patient data
using the Visupac software, which is a patient database.
The images are usually obtained from the posterior
pole’s view, including the optic disc and macula. Three
different classes of fundus image are used there in this
project. Firstly normal fundus images without any
glaucoma infection. Secondly the fundus image with
mild glaucoma infection. Thirdly the fundus images
with severe glaucoma infection. Generally the database is obtained from the
department of ophthalmology in any reputed hospital or
the database is easily available on the internet in the
form of DRIVE database. In this paper totally 69
glaucoma (including normal, mild, severe glaucoma)
images are used.
Fig. 2. Block diagram for the proposed system
B. PRE-PROCESSING
Patient movement, bad positioning, poor focus,
inadequate illumination and reflections can cause a
significant proportion of images to be of such poor quality as to interfere with analysis. In the retinal
images there can be variations caused by the factors,
including differences in cameras, acquisition angle,
illumination and retinal pigmentation. Initially the
retinal color fundus image is converted into gray image
and then noise in the images is removed by using
filtering process. During the image capturing process,
photon noise may present due to the intrinsic property
of light. This noise is removed by using median filter.
Comparing to other filters, median filter removes the
noise as well as preserves the edges of the image.
Glaucom
a infected
images
Preprocessi
ng
Feature
Extraction
Feature
Ranking
Classificat
ion
Dimensional
ity
Reduction
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
47
We use 2D median filter for highlighting and removing
noise from the Morphological open function.
(1)
Here represents a neighborhood
centered on location (m, n) in the image.
C. FEATURE EXTRACTION
Higher order cumulant (HOC) features are extracted
using radon transform. HOC features are extracted for
every 10o.
a. Radon Transform
Radon transforms in 2D is an image projection
operation. It has a special property that it projects the
image intensity along radial line which are oriented in specific angle. This property is very helpful in finding
the features along line integrals. Applying radon
transform for a given set of specific angles in the image
is similar to computing the projection of image in
given angle. The obtained projection is the sum of
intensities of pixels in each direction.
Fig. 3.Radon transform
Let ƒ(x) = ƒ(x,y) be a compactly supported continuous
function on R. The Radon transform (Rƒ), is a function
defined on the space of straight lines L in R by the line
integral along each such line
Concretely, the parameterization of any straight line A
with respect to arc length t can always be written
Where p is the distance of A from the orgin and β is the
angle the normal vector to L makes with the x axis.
It follows that the quantities (β,p) can be considered as
coordinates on the space of all lines in R, and the
Radon transform can be expressed in these coordinates by
(5)
More generally, in the n-dimensional Euclidean
space R, the Radon transform of a supported
continuous function ƒ is a function Rƒ on the space
Σn of all hyperplanes in R. It is defined by
Where the integral is taken with respect to the natural
hypersurface measure, dσ and for 2-dimensional case
generalizes the dy term. Observe that any element of
Σn is characterized as the locus of an equation and p is
equal to product of and y.
Where β∈ pn−1 is a unit vector and p ∈ R. Thus the n-
dimensional radon transform may be rewritten as a
function on pn−1×R. It is also possible to generalize the
Radon transform still further by integrating instead
over k-dimensional affine subspaces of R. The X-ray
transform is The X-ray transform is obtained by
integrating it over straight line and is most widely used
for this construction.
b. Higher order cumulant features
Higher order cumulant features are used in proposed
system and are calculated using radon transform. The
first two order statistics have been used extensively in
bio- signal processing. The first two order moments
and power spectral density (PSD) derived from
moments are the first two order statistics. However,
most of the bio-signals are nonlinear, non-stationary
and non-Gaussian in nature, and thus need to model the
higher order statistics of the signal namely third and
fourth order statistics. The higher order spectra
cumulant are the higher order correlations of the given
signal. By using higher order moments higher order correlations are derived. HOS are used in the analysis
of epileptic EEG signals, automated analysis of sleep
stages, cardiac health diagnosis etc.
Let Y (k), k=0, ±1, ±2, ±3, ±4…is a digital signal to
be sampled. Their first four moments are defined as
(7)
(8)
(9)
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
48
(10)
The nth order moment is given by
, , )
=
The first 4 cumulants of a zero mean process are
defined as
(12)
(13)
(14)
(15)
The nth order cumulant can be calculated using nth
order moment as follows
(16)
Where is the nth order moment of an
equivalent Gaussian process and is
the nth order moment function.
Higher order cumulant features are thus extracted
and in order to reduce the dimensionality of obtaining
features, linear discriminant analysis (LDA),
independent component analysis (ICA) and principal
component analysis (PCA) techniques were used. Among three LDA provides better classification
accuracy
D. DIMENSIONALITY REDUCTION
LDA classifies the data by providing highest
possible discrimination between different classes of
data. When the data set is projected to different space
PCA modifies the location and shape of the data, but
LDA preserves it with maximum class of separability.
It maximizes the ratio of between class variance to
within class variance and hence it guarantees
maximum class of separability. Apart from the
dimensionality reduction LDA also provides maximum discriminating of classes.
LDA seeks to reduce dimensionality while
preserving as much of the class discriminatory
information as possible .Assume we have a set of -
dimensional samples (1, (2, … ( , 1 of which
belong to class 1, and 2 to class 2. To obtain a
scalar by projecting the samples onto a line =
.Of all the possible lines we would like to select the one that maximizes the separability of the scalars.
Good projection vector is obtained by defining a
measure of separation.
The mean vector is given by
We would then select the distance between the
projected means as our objective function
Maximum discrimination of classes can be obtained by
(i) Within class scalar matrix
(22)
(ii) Between class scalar matrix
(ii) Mixture scatter matrix
Similarly, we define the mean vector and scatter
matrices for the projected samples as
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
49
(28)
From our derivation, two-class problem can be written
as
Choose the projection that maximizes the ratio of
between-class to within-class scatter. Since the
projection is not a scalar (it has −1 dimensions) for
longer, we use the determinant of the scatter matrices
to obtain a scalar function.
is the sum of matrices of rank ≤1 and the mean
vectors is constrained
III. RESULTS AND DISCUSSIONS
Initially a fundus image of dimension 256X170 is
given as input. This fundus image is taken from
DRIVE database. Given a color input is converted into a gray image by using RGB to GRAY conversion
method and noise in the images is removed using
median filtering. Then by using radon transform higher
order cumulant features are obtained and the
dimensionality of these features is reduced using LDA.
Better sensitivity can be achieved from fundus
image using higher order cumulant features. Different
morphological features such as the ratio of the distance
between the cup portions of ONH to the diameter of the OD, C/D ratio is also used for glaucoma detection.
By using PCA only 80% of accuracy is obtained for the
SVM classifier. Texture analysis, fractal and power
spectral features, wavelet features and HOS are also
producing a lesser accuracy compared to higher order
cumulant features.
The result obtained from this experiment is as follows.
Fig. 4.RGB to GRAY Conversion
Figure 4 shows the conversion of RGB fundus image to gray color
image.
Fig. 5.Noise removed image
Figure 5 shows the noise removed image. .Fundus images
are having photon noise during capture process and are
removed using median filter.
Fig. 6.Radon transform output
Figure 6 shows the feature extraction using radon
transform. For every 10 degree features are extracted.
Fig. 7 LDA Output
Figure 7 shows the output of LDA technique. If there are
N classes, LDA reduces the dimensionality into N-1 classes. In this paper totally we have three classes and so
we get two sets of LDA features.
IV. CONCLUSION
If glaucoma is not treated in its initial stage it will lead to
permanent vision loss and also glaucoma does not have any
prior symptoms. In order to prevent vision loss earlier
detection of glaucoma is necessary. This paper provides a
new method for early detection of glaucoma using higher
order cumulant features. Higher order cumulant features are extracted using radon transform and the dimensions of the
obtained features are reduced using linear discriminant
analysis.
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
50
V. FUTURE WORK
As a future work, from the obtained higher order cumulant features some of the most significant set of features are selected by using feature selection technique. Feature selection technique ranks all the features and among them features which are having top rank are used for classification. Feature ranking enhances the accuracy of classification. Classification is done by using Support Vector Machine (SVM) classifier. The main aim of this paper is to find a maximum margin hyperplanes in the feature space which can separate negative and positive examples from each other. Among all classifier types SVM can classify the unknown data because it has higher generalization ability and so for better classification SVM classifier can be used.
References
[1] Siamak Yousefi, Michael Goldbaum, Madhusudhanan
Balasubramanian, Tzyy-Ping Jung, Robert N. Weinreb, Felipe A.
Medeiros, Linda Zangwill, Jeffrey M. Liebmann, Christopher A.
Girkin, and Christopher Bowd “Glaucoma Progression Detection
Using Structural Retinal Fiber Layer Measurements and Functional
Visual Field Points,” Ieee transaction On biomedical
engineering,2014 vol. 61, no. 4.
[2] Sushma G.Thorat “Automated Glaucoma Screening using CDR from
2D Fundus Images,” ISSN (e): 2014,2250 – 3005 Vol, 04 Issue, and 5
International Journal of Computational Engineering Research (IJCER).
[3] Sandra Morales, Valery Naranjo, Jesús Angulo, and Mariano Alcañiz
“Automatic Detection of Optic Disc Based on PCA and Mathematical
Morphology,” ieee transaction on medical imaging,2013 vol .32, no.4.
[4] Simon thomas.s, N.Thulasiram ”Automated diagnosis of glaucoma
using Haralick texture features,2013 “International Journal of scientific
research and management(IJSRM)||volume||1||Issue||7||pages ||376-
381||
[5] Sumeet Dua, Rajendra Acharya, Pradeep Chowriappa, and Vinitha
Sree.S “Wavelet-Based Energy Features for Glaucomatous Image
Classification”, ieee transaction on information technology in
biomedicine 2012 Vol. 16, no. 1.
[6] Yuji Hatanaka, Atsushi Noudo, Chisako Muramatsu, Akira
Sawada,Takeshi Hara, Tetsuya Yamamoto, and Hiroshi Fujita
,”Automatic Measurement of Cup to Disc Ratio Based on Line Profile
Analysis in Retinal images,” ieee transaction on medical Imaging
2011 vol. 112, pp.1661–1669.
[7] Rajendra Acharya .U, Sumeet Dua, Xian Du, Vinitha Sree S, and
Chua,” Automated Diagnosis of Glaucoma Using Texture and Higher
Order Spectra Features,” ieee transaction on information technology in
biomedicine 2011 vol. 15, no. 3.
[8] Radimkolar, Jiri Jan ”Detection of glaucomatous eye via color fundus
images using fractal dimensions”radio-engineering, 2008 Vol. 17,No.3
[9] Juan Xu ,Opas Chutatape, and Paul Chew,” Automated Optic Disk
Boundary Detection by Modified Active Contour Model, ”ieee
transaction on biomedical engineering 2007 vol. 23, no.10.
[10] Foracchia.M, Grisan.E, and Ruggeri. A,” Detection of Optic Disc in
Retinal Images by Means of a Geometrical Model of Vessel
Structure,” ieee transaction on medical imaging 2004 vol.23, no.10.
[11] Adam Hoover and Michael Goldbaum,” Locating the Optic Nerve in a
Retinal Image Using the Fuzzy Convergence of the Blood Vessels,”
ieee transaction on medical imaging 2003 vol.22, no.8.
[12] Marc Lalonde, Mario Beaulieu,Langis Gagnon,”Fast and robust optic
disc detection using pyramidal decomposition and Hausdorff-based
template matching, ieee transaction on medical imaging 2001
vol.20,No.11
[13] Arturo Aquino, Manuel Emilio Gegundez-Arias, and Diego Marín,”
Detecting the Optic Disc Boundary in Digital Fundus Images Using
Morphological, Edge Detection, and Feature Extraction Techniques,
”ieee transaction on medical imaging 2010 vol.29 No.11
[14] Kavith.S,K.Duraiswamy,”An efficient decision support system for
detection of glaucoma fundus image using ANFIS,”International
journal of advances in engineering and technology (IJAET) ISSN:
2012, 2231-1963
[15] Murthi.A, Madheswaran.M,”Enhancement of Optic Cup to Disc Ratio
Detection in Glaucoma Diagnosis, ”ieee transaction on medical
Imaging 2012 vol.34.pp.2246-2250.
[16) Rudiger Bock, Jorg Meier,”Glaucoma risk index: Automated
glaucoma detection from color fundus images,” Medical Image
Analysis 2010, 14 ( 471–481).
[17] Husandeep kaur,Amandeep kaur,”Early stage glaucoma detection in
diabetic patients:A review,”International journal of Advanced
Research in computer science and software Engineering 2014 volume
4,issue 5.
51
FEATURE EXTRACTION USING SIFT FOR MAGNETIC RESONANCE IMAGES
Sathiyaseelan. J1,, Mr. C. John Moses2
1,2 ECE dept. St. Xavier’s Catholic College of Engineering, Nagercoil,
Abstract— since the memory capacity of the images are high, storage of these images and files impose a huge
barrier in medical field. This project involves feature extraction using SIFT algorithm (Scale Invariant Feature
Transform) for MR (Magnetic Resonance) images. Here the quality of an image is increases and noise is reduced.
Feature extraction is the process of transforming input data into set of features. The study also investigates the feature
extraction and effective memory size reduction.
I. INTRODUCTION Magnetic resonance (MR) image is a type of scan
use to know the body defects with the help of signals
from normal and abnormal areas of the body which help
the doctors. To identify the defects this is the advance
technology over CT scan which neglects usage of X-
rays. It is more accurate and precise. Usually it is used
for brain, spine and joint imaging.
Feature extraction is nothing but defining or
identifying various features and extracting them from
the input data or image. Feature extraction is
transforming the input data into the set of features. It
integrates computation, programming is easy to use environment. Feature extraction will reduces the
resources need for desiring a data. MATLAB is a
language where the problems and the solutions were
represented in mathematical notation.
A robust set of methodologies to reduce the
design power consumption, X power for designed based
power analysis, web based power tools. The system
generator basics provides two key tools one is blocks for
building model and other hardware generator is model
into HDL code generator test vector with MATLAB and
Simulink block sets using HDL coder, generate VHDL
and Verilog code for Xilinx FPGA’s from MATLAB, Simulink and state flow model. Advantage of system
generator is to reduce the loss of data precision and save
time by skipping a step during code generation.
II. MATERIALS AND METHOD
In fig-1, the input source MR image should
have some other noise so that the noise has to be
smoothened by the Gaussian filter and the filtered
output will send to difference of Gaussian it will
identify the difference between two images. To increase the contrast of an image histogram equalization will be
used. The key point detection s used to discard low
contrast key points and then filters out those located on
edges. Each key points assigned one or more orientation
based on local image gradient direction the magnitude
and direction calculation for the gradient are done for
each and every pixel in a neighboring region. Then to
compute a descriptor vector for each key point such that
the descriptor is highly descriptive and partially
invariant to the remaining variations.
The Simulink blocks for MATLAB, where the
blocks are taken from various blocks from Simulink and
the Xilinx block for the gateway in, gateway out and system generator were used. The vertex 6 and other
logical devices were used to compute the design report,
RTL schematic, timing report and power analysis report.
This should calculate for various MR images.
Fig-1 Block diagram of SIFT for feature extraction
Alogrithm:
Step 1: Read the input Magnetic Resonance image of
brain.
Step 2: Convert RGB image into gray scale image
Step 3: Gaussian filtered is applied to an image.
Step 4: Histogram equalization is applied to increasing the contrast of an image.
Step 5: Edge detection is performed by key point
detection.
Step 6: Gradient magnitude and orientation is calculated
for image direction
Step 7: Finding key points and processing time for
various images.
A. Gaussian Filter
52
Gaussian filter is use to get original structure of the
input data by filtering the noise from the input data or
image. Gaussian filter is use to eliminate the noise and
to smooth the image. First convolution between source
image and Gaussian function, the result is a Gaussian
blurred image.
yxIyxGyxL ,*,,,,
Where, Gaussian function
2222/1,, yxezyxG
= Scaling factor
x, y =spatial coordinates
B. Difference of Gaussian
Difference of Gaussians (DoG) is an algorithm
which is used for finding the difference between two
filtered images. The subtraction between blurred image
and an original image from input, less blurred version of the original image. In the simple case of grayscale
images, the blurred images are obtained by convolving
the original grayscale images with Gaussian kernels
having differing standard deviations. The range of
frequency that preserves the blurred images in spatial
information.
,,,,,, yxLkyxLyxD
Where,
k – Kernel factor
C. Key point Detection
Localization of key points using DoG is carried
out here both high contrast and low contrast point are
detected, only the high contrast point are consider
whereas the low contrast point are neglected with the help of these high contrast point location, scale, rotation
have been done in order to find the key points in
accurate manner. After scale space extrema are detected
the SIFT algorithm and then filters out those located on
edges. Resulting set of key points is shown on output
image. The detection taken here by Scale-space extrema
produce many key points some of these are unstable.
D. Gradient Magnitude And Orientation
It is the necessary to identify the image even through the
scale, rotation and orientation are different for this
purpose the gradient magnitude and orientation is done with the help of invariance key points. Each key point is
assigned one or more orientations based on local image
gradient directions.
The Gaussian-smoothed image L at
the key point's scale σ is taken so that all computations
are performed in a scale-invariant manner. For an image
sample at scale σ, the gradient
magnitude, , and orientation, , are
precomputed using pixel differences:
Where,
Vertical Direction yxLyxLGx ,1,1
Horizontal Direction 1,1, yxLyxLGy
221,1,,1,1, yxLyxLyxLyxLyxm
yxLyxLyxLyxLsayx ,1,1,1,1,tan,
The magnitude was calculated for the gradient
were done for each pixel in the region of key points of
Gaussian blurred input image. The orientation histogram
of 36 bins is formed, each has 10 degrees. Each sample
in the neighboring window added to a histogram bin is
weighted by its gradient magnitude and by a Gaussian-
weighted circular window with a scale factor that is 1.5
times that of the scale of the key point.
E. Feature Descriptor
Key point locations of an image at particular
scales and assigned orientations. These ensures
invariance to image location, scale and rotation. Now
we want to compute a descriptor vector for each key
point such that the descriptor is highly distinctive and
partially invariant to the remaining variations such as
illumination, 3D viewpoint, etc.
The set of orientation histograms is created on
4x4 pixel neighborhoods with 8 bins each. These
histograms were computed from magnitude and
orientation values of samples in a 16 x 16 region around
the key point such that each histogram contains samples from a 4 x 4 sub region of the original neighborhood
region. The magnitudes are further weighted by a
Gaussian function with equal to one half the width of
the descriptor window. The descriptor then becomes a
vector of all the values of these histograms. Since there
are 4 x 4 = 16 histograms each with 8 bins the vector
has 128 elements. This vector is then normalized to unit
length in order to enhance invariance to affine changes
in illumination. To reduce the effects of non-linear
illumination a threshold of 0.2 is applied and the vector
is again normalized. Although the dimension of the descriptor,
seems high, descriptors with lower dimension than this
don't perform as well across the range of matching tasks
and the computational cost remains low due to the
53
approximate BBF method used for finding the nearest-
neighbor. It is also shown that feature matching
accuracy is above 50% for viewpoint changes of up to
50 degrees. Therefore SIFT descriptors are invariant to
minor affine changes. To test the distinctiveness of the
SIFT descriptors, matching accuracy is also measured against varying number of key points in the testing
database, and it is shown that matching accuracy
decreases only very slightly for very large database
sizes, thus indicating that SIFT features are highly
distinctive.
III RESULT
In fig-2 the input image is the brain images from
the image processing tool box. This image is filtered
using Gaussian filter and the noise will be removed.
Histogram is used for increases the contrast and enhance an images for the filtered image will increases
the intensity of an image by histogram equalization
process. The Canny edge detector is an edge
detection operator that uses a multi-stage algorithm to
detect a wide range of edges in images. The pixel values
should change edges length of an image. And then
enlarge size of an image.
Fig-2 Input brain MR image
Input MR
Images
Knee
(sv)
Knee
(lv)
Joint
(sv)
Joint
(fv)
Shoulder
(fv)
Brain
(cssv)
Skull
(cssv)
Spinal
cord
(cssv)
Brain
(cstv)
REAL time to
Xst completion
(seconds)
14 s 16 s 11 s
12 s 15 s 15 s 10 s 11 s 9 s
Memory size
(Kbits)
210900 209236 210900 208724 210644 208852 208532 210900 210900
CPU time to Xst
completion
(seconds)
13.70 s 16.24 s 10.80 s 12.45 s 15.02 s 15.20 s 10.39 s 11.22 s 9.47 s
Table- Result for computation time Various MR imageIn
this table the various input MR images are shown for
feature extraction. The real time computation CPU time
computation and memory size were calculated in these table
sing Simulink.
Fig-3 Output of key points extraction
The key point detection should carried out by
varying the scaling factor. The key point detection output.
The key points are extracted and shown in the workspace of
MATLAB the time to finding the key points and total no of
key points extracted
Fig-4 Key point extraction of an image
54
IV. DISCUSSION
In the system we have the software part which
reduces the memory size of an image and increases the
processing speed of an image. The memory size, real time and
CPU time will be computed in the table for various images.
Fast SIFT method can also be used for further improvement.
V. CONCLUSION
The SIFT algorithm has been a known extracting
features from a input image. Hardware acceleration is often
needed in many computer vision applications. In this project
the SIFT algorithm is faster for feature extraction and
processing speed is less. This design enables a streaming type
of data flow that uses a 50% less memory and calculate
processing time for various MR images. This project will be
develop by increasing the processing speed of the MR image.
VI. REFERENCES
[1] Chao-Yung Hsu, (2010)” Secure And Robust Sift with Resistance
to Chosen-Plaintext Attack” Proceedings of 2010 IEEE 17th
International Conference on Image Processing September26-29 pp
997-1000.
[2] Chao-Yung Hsu (2012) “Image Feature Extraction in Encrypted
Domain With Privacy-Preserving SIFT” IEEE transactions on
image processing, vol. 21, no. 11 pp 4593-4607.
[3] Claudia Chevrefils,(2009)” Texture Analysis for Automatic
Segmentation of Intervertebral Disks of Scoliotic Spines From MR
Images” IEEE transactions on information technology in
biomedicine, vol. 13, no. 4 pp 608-620.
[4] David G Lowe,(2004)”Distinctive Image Features from Scale-
Invariant Key points” International Journal of Computer Vision
pp 1-28.
[5] Feng-Cheng Huang,(2012) “High-Performance SIFT Hardware
Accelerator for Real-Time Image Feature Extraction” IEEE
transactions on circuits and systems for video technology, vol. 22,
no. 3 pp 340-351.
[6] Herbert Bay1, Tinne Tuytelaars(1998)” SURF: Speeded Up
Robust Features”IJCV 30(2) pp 1-14.
[7] Hong Liu, Hong Lu,(2010)”SVD-SIFT for Web near Duplicate
Image Detection” IEEE transactions on information technology
ICIP , vol. 13, no. 4 pp 1445-1448.
[8] Jayasimha Rao, Noora Partamies,(2014) ”Automatic Auroral
Detection in color Allsky Camera Image” IEEE transactions on
circuits and systems for video technology, vol. 24, no. 7 pp 1474-
1481.
[9] Jie Jiang, Xiaoyang Li,(2014)” SIFT Hardware Implementation for
Real-Time Image Feature Extraction” IEEE transactions on
circuits and systems for video technology, vol. 24, no. 7 pp 1209-
1220.
[10] Kazu Mishiba,(2013)”Image Resizing with SIFT Feature
Preservation” IEEE transactions on image processing ICIP, vol.
12, no. 10 pp 991-995.
[11] Kosuke mzuno et al. (2011) “A low power Real-time SIFT
Descriptor Generation Engine for Full-HDTV Video
Recognition”IEICE trans electron vol e94-c no 4 pp 448-457.
[12] Liang-Chi Chiu (2013)“Fast SIFT Design for Real-Time Visual
Feature Extraction” IEEE transactions on image processing vol.
22, no. 8 pp 3158-3167.
[13] M V S Lakshmi, Ravi Mathey, (2013)”Encrypted Feature
Extraction for Privancy SIFT” IEEE 17th International Conference
on Image Processing ISSN vol 1, issue 12 pp 23-26.
[14] Nithya Priya, Sasikumar,(2014)” Detection and Segmentation of
Brain Tumor using Adaboost SVM” IJIRCCE ISSN, vol. 2, special
issue1 pp 2323-2328.
[15] Peter I. Rockett,(2003)” Performance Assessment of Feature
Detection Algorithms: A Methodology and Case Study on Corner
Detectors” IEEE transactions on image processing, vol. 12, no. 12
pp 40-47.
[16] Rukun hu et al.(2009) “Investigating Visual Feature Extraction
Methods for Image Annotation” IEEE international conference on
system pp 3122-3127.
[17] Salim Lahmiri and Mounir Boukadoum,(2011)” Classification of
Brain MRI using the LH and HL Wavelet Transform Sub-bands”
IEEE 978-1-4244-9474-3/11 pp 1025-1028.
[18] Shuai Lu,Qinging Zhao,(2013)”MultiScale, Multi Level,
Heterogoneous Features Extraction and Classification of
Volumetirc Medical Images” IEEE transactions on information
technology in biomedicine, vol. 13, no. 4 pp 1418-1422.
[19] Wei Tu Chen, Wei Chuan Liu,(2010)”Adaptive Color Feature
Extraction Based on Image Color Distribution” IEEE transactions
on image processing, vol. 19, no. 8 pp 2005-2016.
[20] Xudong L U,(2012)” Perceptual Image Hashing Based on Shape
Contexts and Local Features Points” IEEE transactions on
information forensics and security, vol. 7, no. 3 pp 1081-1093.
[21] Yahao Zhau, Qiue Yu,(2013) “An Automatic Global to Local
Image Registration Based on SIFT and Thin Plate Spline” IEEE
transactions on image processing, vol. 21, no. 11 pp 2535-2538.
[22] Yao Shen, Parthasarathy Guturu, (2009)” Video Stabilization
Using Principal Component Analysis and Scale Invariant Feature
Transform in Particle Filter Framework” IEEE transactions on
consumer electronics, vol. 55, no. 3 pp 4343-4356.
[23] Ye zhang and Qingmao hu,(2008) “A PCA Based Approach to the
Representation and Recognition of MR Brain Midsagittal Plane
Images”IEEE EMBS conference pp 3916-3919.
[24] Atiq Islam, Syed M. S. Reza, (2013)” Multifractal Texture
Estimation for Detection and Segmentation of Brain Tumors”
IEEE transactions on biomedical engineering, vol. 60, no. 11 pp
3204-3215.
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
55
Simulation of Critical Home Health Monitoring & Drug Delivery
System
N. Abhinaya, G. Hari Krishnan, R. J. Hemalatha, G. Umashankar Department of Biomedical Engineering,
Sathyabama University, Chennai
ABSTRACT
The interface of enabling technology with advanced product design has shown radical development in the field of intelligent sensor embedded system design. Numerous applications are envisaged exploiting this inter-connectivity, particularly, in the field of biomedical applications. A need, for example, that is of growing demand,
is in the field of remote health monitoring and control of critically ill patients, with the help of networked sensors. The continuous monitoring of the health of a patient in a hospital, information fusion from multiple sensor data as well as broadcasting the recorded data on a network for the ease of access to the clinician and implementing the decisions of clinicians through automated drug delivery units could save millions of precious lives in a country with limited medical experts. In what follows is a brief detailed description of such a system that is developed. `Keywords: Home health care, Drug delivery system, embedded system, LabView, Patient, Healthcare.
INTRODUCTION
The monitoring of various parameters of a pa-
tient in critical care, like blood pressure, temper-
ature, heart rate is extremely important, to have a
continuous evaluation of the critically ill patients
[1]. The real time information of these parame-
ters is quite useful for medical services and if it
is easily accessible from a remote area for a cli-
nician then it can solve a major problem of phys-
ical absence of medical experts, in many places
of the developing World [3 & 4]. The measured
parameters are interfaced with LabView installed in a host PC using the Data acquisition unit of
National Instruments. The interfaced data are
further processed by using a suitable information
fusion algorithm. The net information is broad-
casted to client PC which is connected to the host
PC via use of web protocols of LabView [2&5].
This help in monitoring the parameters through
PC’s connected via intranet or internet with each
other. Finally drug delivery control system which
will have a data acquisition controlling and
broadcasting unit, display unit, and drug control-ling unit.
As the rapid development of information tech-
nology and automation technology, which in-
creases efficiency of industry greatly, is making
the patients of hospital wards and the health care
workers in the automatic association. Each bed
of the hospital wards and nurse room need long-
distance automatic call from time to time, which
is easy for patients in urgent of medical signal,
and the signal staff can also deal with relevant
affairs quickly and efficiently [7].
The proposed system approaches designing a
home health monitoring and drug control system
which can be considered as an integrated part of
three subsystems multi‐sensor data acquisition,
information fusion networking and data broad-
casting and automated system control [6]. This
multiple channel intelligent system would meas-
ure the parameters like blood pressure, tempera-
ture, heart rate etc Next, the measured parame-ters are interfaced with LabView installed in a
host PC using the Data acquisition unit of Na-
tional Instruments. The interfaced data are fur-
ther processed by using a suitable information
fusion algorithm. The net information is broad-
casted to client PC which is connected to the host
PC via use of web protocols of LabView. This
help in monitoring the parameters through PC’s
connected via intranet or internet with each other
[8].
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
56
For a patient who is in coma state, or who be-
came disabled due to some accidents or some
other cases need to be treated for a long time. It’s
difficult for people to afford the expense in the
hospital for such a long period. This project will
be more helpful for such people, where the pa-
tients can be monitored from the home itself.
Daily doctor can check the condition of the pa-tient without the physical presence and even if
some abnormalities are found the drug can be
given to the patient. The drug and the dosage of
the drug are controlled by the doctor through the
doctors control panel [9].
MATERIALS & METHODS
The project is done using LabView. LabVew
(short for Laboratory Virtual Instrumentation
Engineering Workbench) is a platform and de-
velopment environment for a visual program-
ming language from National Instruments [6].
The graphical language is named "G" originally
released for the Apple Macintosh in 1986.
LabView commonly used for data acquisition,
instrument control, and industrial automation on a variety of platforms including Microsoft Win-
dows, various flavors of UNIX, Linux, and Mac
OS X.
Algorithm
• Read the values from the files as indicat-
ed in figure 1.
• Update the current output status
• Indicate the abnormal values with re-
quired medicine
• Require doctor control
• Necessary actions take over by the physi-
cian
.
Figure 1 Diagramatic representation of algorithm
The main block diagram for reading and control-
ling parameter is as shown in figure 2. The pro-
ject of Home health monitoring system simula-tion results are obtained in LabVIEW develop-
ment system. Process result consists of one nor-
mal condition and six abnormal conditions. In
the file path control normal or abnormal files are
chosen to obtain the results.
Figure 2 Block diagram for reading the parame-
ters
The observed parameters are interfaced with LabVIEW installed in a host PC. The read
data are further processed by LabVIEW algo-
rithm and it will displayed by respective regions.
Doctor control unit is used to apply the values to
drug delivery unit. This also has been imple-
mented in simulation. Front panel consist of
blood pressure, temperature, heart rate and drug
level displays.
RESULT & DISCUSSION
The first condition is taken as normal, if the pa-
tient is in normal condition then no LED’s
glows. The normal temperature, pressure and
heart rates are displayed. In the status display it
won't display a need of medicine instead it dis-
plays the condition is normal. The next condition
is abnormal condition, it can be due to variation in any parameters, and it might be due to abnor-
mal temperature, pressure or heart rate. If there is
some variation in any parameters which are
measured, from the normal values then on the
corresponding front panel the LED glows indi-
cating an abnormal condition. If the patient’s
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
57
temperature is high then LED glows at tempera-
ture panel and status appears as abnormality in
temperature and also the medicine which is
needed is displayed in the status.
Likewise if there is any abnormality in other
parameters the same steps are followed. For
knowing the level of the drug which is stored
another indicator is present. If the drug level is
too low, then LED glows with an alert to refill
the drug. It also consists of a doctors control
panel which is the most important part where the
doctor from a remote area can control the drug
delivery to the patient. This consist of mainly
two parts one for selecting the drug and the other
for selecting the dosage.
Figure 3 Front panel display of body temperature
In this front panel the body temperature of the
patient will be displayed both in Celsius and
Fahrenheit as shown in figure 3. The temperature
will be also indicated as in a thermometer level
model. If the patient’s temperature is normal
then the status will be indicating normal. If the
patient is having a high temperature than the
normal level then LED glows like alarm. And
also the drug required and the abnormal condi-tion will be displayed. The normal body tem-
perature is 37C and 98.6F.
In this front panel the heart rate of the patient is
displayed as shown in figure 4. There are two
conditions for abnormal heart rate which are
tachycardia and Bradycardia. Normal heart rate
of an adult is 60-80 beats per minute. Tachycar-
dia is the increased heart rate and bradycardia is
the decreased heart rate. LED glows if there is
increased or decreased heart rate from the normal
condition. In the status the drug required an the
condition of the heart is indicate. In this front
panel blood pressure of the patient is displayed
as shown in figure 5. There are two abnormal
conditions high blood pressure and low blood
pressure. Normal blood pressure of an adult is 80/120 mmHg. If the pressure is more than
90/140 mmHg this indicates high blood pressure. If the pressure is lower than 60/90 mmHg this
indicates low blood pressure. If the patient is
having any pressure abnormalities the LED
glows. Status displays the condition whether it is
high blood pressure or low blood pressure and
also shows the medicine required.
Figure 4 Front panel display of heart rate
Figure 5 Front panel display of pressure
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
58
In this front panel drug level which is to be given
to the patient is displayed as shown in figure 6. A
tank indicates the level of the drug. When the
drug amount stored becomes low then an LED
glows this is an alarm for refilling the tank.
Figure 6 Front panel display of drug level
Figure 7 Front panel display of doctors control
panel
In figure 7 doctors control panel has been ex-
plained. This consists of tools for selecting the
required medicine and for selecting the dosage of
the medicine. The status of the medicine which is
used will be displayed. The level of the medicine
is also indicated which is all controlled by the
doctor.
SUMMARY & CONCLUSION
In this paper a home health monitoring with drug
delivery under the control of a doctor has been
introduced. Simulation of home health monitor-
ing & feed back control with drug delivery was
done using Lab VIEW. This project helps to monitor the patient from the home itself and doc-
tor can control the drug delivery even from a
remote area. This helps many people to get treat-
ed from their home itself without much expense. The real time implementation of this project can
bring drastic changes, and can save millions of
life.
REFERENCES
[1] Rifat Shahriyar, Faizul Bari MD., Kundu Gourab,
Sheikh Iqbal Ahamed and Mostofa Akbar MD., Sep-
tember 2009 ,“Intelligent Mobile Health Monitoring
System (IMHMS)”, International Journal of Control
and Automation, Vol.2,No.3.
[2] Bratislava, S.K Miroslav, Lehocki Fedor and Valky
Gabriel, 2-7 Jan 2012, “Multi-Platform Telemedicine
System for Patient Health Monitoring”, Proceedings of
the IEEEEMBS International Conference on Biomedi-
cal and Health Informatics (BHI 2012), Hong Kong and
Shenzhen, China.
[3] T. E. Doyle, M. Kalsi, B. Aiyush, J. Yousuf, and O.
Waseem,( 2009), “Non-Invasive Health Monitoring
System (NIHMS)”, Science and Technology for Hu-
manity (TICSTH), IEEE Toronto International Confer-
ence.
[4] Nitin P. Jain, Preeti N. Jain and Trupti P. Agarkar,
2012,“An Embedded, GSM based, Multiparameter,
Realtime Patient Monitoring System and Control -
AnImplementation for ICU Patients”, Information and
Communication Technologies (WICT).
[5] Shubhangi M. Verulkar, Maruti Limkar, June 2012
“Real Time Health Monitoring Using GPRS Technolo-
gy”,International Journal of Computer Science and
Network (IJCSN) Volume 1, Issue 3,
[6] S. Deepika, V.Saravanan, May-2013 “An Implementa-
tion of Embedded Multi Parameter Monitoring System
for Biomedical Engineering”, International Journal of
Scientific & Engineering Research, Volume 4, Issue 5.
[7] Chris A. Otto, Emil Jovanov, and Aleksandar
Milenkovic, 2006, “WBAN-based System for Health
Monitoring at Home”, Journal of Mobile Multimedia,
vol. 1, pp. 307-326.
[8] Lodaya P.N,Wadkar S.P, “Movable patient health mon-
itoring using GPS”Advances in Recent Technologies in
Communication and Computing (ARTCom 2011), 3rd
International Conference on 14-15 Nov. 2011
[9] Alexandros Pantelopoulos and Nikolaos G. Bourbaki,
January 2010, “A Survey on Wearable Sensor-Based
Systems for Health Monitoring and Prognosis”, IEEE
Transactions onSystems, Man, and Cybernetics—Part
C: Applications and Reviews, vol. 40, no.1.
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
59
DESIGN AND DEVELOPMENT OF DIAGNOSTIC TOOL FOR STRAIN
INJURIES OF FINGER JOINTS USING FLEX SENSOR
Balakumaran.V1, Bhuvaneshwari.S
2, Nithyaa.A.N
3, Premkumar.R
4
1,2,3,4Department of Biomedical Engineering, Rajalakshmi Engineering College,Chennai-602105, India
Abstract
Repetitive strain injuries (RSIs) are injuries to the musculoskeletal and nervous systems that may be caused by repetitive
tasks, forceful exertions, vibrations or mechanical compression. In recent days majority of the population uses computers
and it is an inevitable part of day to day life. People working as programmers tend to work for long hours. This work culture
strains the joints in the fingers beyond their normal capacity leading irreversible joint damage. If a person loses his fingers
functionality it is a great impairment for him and also to people around. To prevent a person from losing his fingers
functionality an early detection of the condition is important. This research focuses on designing a glove with flex sensor and
a microcontroller to diagnose and measure the extent of joint deformities in hand.
Keyword- flex sensor, microcontroller, glove, Graphical User Interface
INTRODUCTION
REPETITIVE STRAIN INJURY (RSI)
Repetitive strain injuries also known as
cumulative trauma disorders refers to the injuries of
musculoskeletal and nervous systems that may be
caused by strain forceful exertions, vibrations,
awkward position and mechanical compression.
Some of the measures used to asses RSI's are grip
and strength and diagnostic tests such as
Finkelstein's test. General exercise is recommended
for reducing the risk of developing RSI.
Different Causes for repetitive Strain Injury
• Tendinitis
• Tendinosis
• Carpal Tunnel Syndrome(CTS)
Tendinitis is the inflammation of a tendon that
involves larger-scale acute injuries which occurs
mostly in limbs. The severity of Tendinitis vary
according to different individual like rock climbers
develop in fingers , swimmers in their shoulders.
Symptoms are tenderness near the joint, mild
swelling and muscle stiffness.
Tendinosis also known as chronic tendinitis is the
damage of tendon at cellular level. It can be
detected visually or by touch for simple swelling.
The increased water content is detected by
ultrasonography. Symptoms can vary from pain
and stiffness of the tendon, or burning around the
inflamed tendon. Physical therapy and rest is a
common experienced therapy. Surgery is done for
extreme cases.
Carpel tunnel syndrome is a median entrapment
neuropathy that causes pain and symptoms in the
distribution of the median nerve due to its
compression at the wrist in the carpal tunnel. It
may be caused by a combination of genetic and
environmental factors. The main symptoms of CTS
is intermittent numbness of the fingers. Its
treatments include use of corticosteroid injection
and night splints.
MATERIALS AND METHOD
The first step is designing a smart glove which is
mounted with 4.5 inch flex sensors on each finger.
This sensor has a characteristic property of
showing a change in the resistance when it is bent.
By exploiting the change in resistance in the sensor
due to the bend, a smart glove has to be designed
and fabricated for diagnostic purpose.
Flex sensor, a input device whose input is a
constant 5V supply and the corresponding voltage
from the sensor is the output. The output voltage is
affected by the variations in resistance these
variations are to be measured. During zero bend
state, in order to control the voltage from the
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
60
sensor, the output is passed through a voltage
divider circuit. Then its output is fed to the ADC of
the microcontroller to get the corresponding digital
value for the voltage. ADC used is the internal 10
bit present in the microcontroller. The
microcontroller Atmega 328, sends the data to the PC and it is represented as a bar graph in a GUI.
The data is transferred as individual byte from each
ADC channel in real time. The GUI is designed to
read the values that the microcontroller transmits
through its Rx and Tx lines UART to the
FTDI232RL that sends these values continuously
to the PC on a COM serial port. The values on the
COM are read by the GUI and it is interpreted in
bar graph representation.
Voltage divider is a linear circuit that produces an
output voltage (Vout) that is a fraction of its input
voltage (Vin). Voltage division refers to the partitioning of a voltage among the components of
the divider.
Impedance Buffer provides electrical impedance
transformation from one circuit to another Voltage
buffer used to transfer voltage to the
microcontroller to make reading from the ADC and
also transfer a voltage from a first circuit, having a
high output impedance level, to a second circuit
with a low input impedance level.
Microcontroller- ATmega328 is a single chip
micro-controller created by Atmel which operates between 1.8-5.5 volts and belongs to the megaAVR
series. The high-performance Atmel 8-bit AVR
RISC-based microcontroller combines 32 KB ISP
flash memory with read-while-write capabilities, 2
KB SRAM ,1 KB EPROM, 23 general purpose I/O
lines and 32 general purpose working registers.
Graphical User Interface is designed using
processing whose program can be written in the
text editor as sketches. It reads the real time values
in the ADC that the controller reads and it sends to
the PC and interpreted as a real time bar graph plot
for better diagnosis.
Fig 2. Typical GUI of the system
Fig 4. Working of the system
RESULT AND DISCUSSIONS
The glove was worn by the subject and they were
asked to bend their individual fingers one by one in
increments of 10º till the point where they could
bend the maximum. The corresponding voltage
output from the sensor is recorded for each angle.
Angles of bend are
10,20,30,40,50,60,70,80,90,100,110 till the
maximum the person can bend. There is variability
exhibited by people with fatter hands not able to
bend to the extent that a person with thin fingers
can. It is not an abnormality and it is just because
of the more amount of flesh in their fingers that this
variation arises. The data is tabulated and the graph
is plotted for discussions. This type of study was
carried out for different age and sex. The
recordings here were interesting and have
significant importance.
Subject 1
Age: 21
Ang
le
Ind
ex
Mid
dle
Rin
g
Litt
le
Thu
mb
0 4 4 4 4 4
10 3.9
87
3.94
4
3.9
01
3.9
22
3.92
2
20 3.9
44
3.92
2
3.8
58
3.8
58
3.81
5
30 3.8
79
3.90
1
3.7
85
3.8
15
3.75
40 3.8
75
3.87
9
3.7
72
3.7
93
3.70
7
50 3.7
5
3.83
6
3.7
07
3.7
07
3.66
4
60 3.7
07
3.77
2
3.5
34
3.6
85
3.66
4
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
61
70 3.6
64
3.71
7
3.4
7
3.6
25
3.54
2
80 3.6
21
3.47
8
3.2
97
3.5
34
3.49
9
90 3.5
13
3.38
4
3.1
9
3.4
91
3.49
1
100 3.4
05
3.31
9
3.1
68
3.3
19
Table 5.1 Subject 1 data recorded
Fig 5.1 Graph for Subject 1
From the study it is clear that there is a significant
increase in the resistance offered by the sensor with
the bending of the fingers. Here it is clear that
when a person suffers an abnormality and is not
able to bend the finger beyond a point and not able
to clinch fingers, there is a significant low
resistance offered by the sensor than compared to
condition were the person is able to close fingers
fully. This results in a smaller voltage drop in case
of abnormality and this is the data provided for
diagnosis and the person can be suggested with
treatment. This method is purely a screening test
and after detecting the abnormality the doctors will
have to find the exact nature of the problem and
provide treatment to the patients.
Fig 7. Output of the system
CONCLUSION AND FUTURESCOPE
Many people nowadays suffer from RSI even
without knowing they are having a problem with
their fingers. Most neglect not being able to bend
fingers as a problem. A majority of population
work in IT field and are suffering various
complications due to their work nature. A
diagnostic method is not available to diagnose RSI
of the fingers. Thus our system will fill the gap in
providing an easy and reliable system for diagnosis
of RSI. The resistance change are represented in a
GUI, which provides a means for easy diagnose in
real time. The glove is user friendly and can be
worn and removed with ease. It does not affect any
natural degrees of motion of the fingers. It can be
worn during a subject is carrying out a work and
bending can be recorded. This system provided real
time data thus making the process easy and
removes delays in interpreting the data and result
generation.
The system uses wires to connect to the computer
and the glove this can be replaced with a wireless
connectivity to improve convenience and user
friendliness. It can be used to record data to a vast
population and the data process using database and
we can use pattern recognition system to generate
an automated diagnosis in computer without a
doctor to interpret the data.
REFERENCES [1] Ali, A.M.M., Parit Raja, Ambar, R. ; Jamil, M.M.A. ; Wahi,
A.J.M. ; Salim; Artificial hand gripper controller via Smart
Glove for rehabilitation process, Biomedical Engineering
(ICoBE), 2012 International Conference, 27-28 Feb. 2012,pp
300 - 304
[2] Axisa, F.Gehin, C. ; Delhomme, G. ; Collet, C. ; Robin, O. ;
Dittmar, A.; Wrist ambulatory monitoring system and smart
glove for real time emotional, sensorial and physiological
analysis Engineering in Medicine and Biology Society, 2004.
IEMBS '04. 26th Annual International Conference of the IEEE
(Volume:1), 1-5 Sept. 2004,pp:2161 - 2164
[3] FTDI 232 R datasheet
[4] Getting Started with Processing by Casey Reas and Ben Fry.
[5] Laura Dipietro, Angelo M. Sabatini, Paolo Dario; A Survey
of Glove-Based Systems and Their Applications IEEE
transactions on systems, man, and cybernetics—part c:
applications and reviews, Vol. 38, no. 4, July 2008
[6] Learning Processing: A Beginner's Guide to Programming
Images, Animation, and Interaction by Daniel Shiffman.
[7] Lisa K Simone and Derek G Kampe; Design considerations
for a wearable monitor to measure finger posture, Journal of
NeuroEngineering and Rehabilitation March 2005,2:5
doi:10.1186/1743-0003-2-5
[8]http://www.atmel.com/images/atmel-8271-8-bit-avr-
microcontroller- atmega48a-48pa-88a-88pa-168a-168pa-328-
328p_datasheet.pdf
[9] http://www.bionicgloves.com/default.asp
[10] http://www.cyberglovesystems.com/products/cyberglove-
ii/overview 52
[11] http://en.wikipedia.org/wiki/Power_Glove
[12] http://www.ti.com/lit/ds/symlink/lm124-n.pdf
[13] http://en.wikipedia.org/wiki/Buffer_amplifier
[14] United States Patent, Patent Number: 5,347,843, Date of
Patent: September, 10, 1991
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
63
CLUSTER ANALYSIS OF GENE EXPRESSION DATA USING
OPTIMIZATION TECHNIQUES: A SURVEY
B. Nithya 1 1, R. Rathipriya 2
2
1,2, Department of Computer Science, Periyar University, Salem -636 011, Tamilnadu, India.
Abstract
The microarray data analysis is one of the most important things for functional genomics research. The development of microarray technology produces massive gene expression data sets. Clustering is the process of grouping genes into set of
disjoint classes called gene clusters so that genes within a cluster are highly similar with one another and dissimilar with the genes in other clusters. Applying traditional clustering techniques such as Hierarchical clustering, k-means, C-means, Self-Organized Maps Clustering (SOM) for this type data is tedious. Therefore, this paper studies the available clustering techniques using optimization techniques like, Particle Swarm Optimization (PSO), ACO, BCO, etc., in the literature and summaries the clustering using optimization techniques for microarray data.
KEYWORD: Gene Expression data, clustering, PSO, K-means, C-means.
Introduction:
Gene Expression [14] is the process by
which information from gene is used in the synthesis
of a functional gene product. These products are
often proteins, but in non protein coding genes such
as rRNA genes or tRNA. Genes, the product is a
functional RNA. Each data point produced by a DNA microarray hybridization experiment represents the
ratio of expression levels of a particular gene under
two different experimental conditions.
Clustering is the unsupervised learning
problem. A clustering is the set of clusters, usually
containing all genes in the microarray datasets. It is
defined as the process of organizing genes into
groups whose members are similar in some way. A
cluster is a collection of genes which is “similar”
between them and are “dissimilar” to the genes
belonging to other clusters. There are mainly two categories of clustering. The first one is hierarchical
method and the other one is partitional method.
Partitional method will divide the genes to various
clusters based on some conditions. Many diverse
clustering techniques have extensively been under
development of the clustering. The most widely used
techniques in analysis of gene expression data which
are applied in the early stages and proven to be useful
are Hierarchical clustering , K-means clustering and
Self-organized maps (SOM).
Microarray data clustering analysis:
Clustering gene expression data can be divided into the three groups,
1) gene-based,
2) sample-based and
3) Subspace clustering as both
genes and samples is required to clustered
genes.
Gene-based clustering: The gene-based clustering is divided into
group of genes together in co- expressed genes which
indicate co-function, co-regulation and reveals the
natural datasets of genes. The genes are treated as the
cluster, while the samples of the features. Clustering
algorithms for gene expression data should be
competent of extracting the microarray genes from a
high level of background noise. A clustering
algorithm should depend as possible on prior
knowledge and also gives a graphical representation
of the cluster structure and the other than partitioning
the gene expression datasets.
Sample-based clustering:
To find the substructure of the microarray
data, regards the samples as the genes and the group
of genes as the features. Genes are generally related
to different infection or medicine effects within a
gene expression matrix. Only a minute subset of
genes whose gene expression levels are powerfully
correlated with the class similarity, rise and fall
coherently and exhibiting fluctuation of a similar
shape under a subset of the gene conditions, called
the informative genes that participate in any
microarray data. The remaining gene data are
regarded as noise in the microarray data as they are irrelevant to the gene expression datasets. By
focusing on a subset of genes and the conditions of
significance, the noise levels are inducing by new
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
64
genes and condition can be lowered, which is
characterized by co-clustering. As a result, to identify
useful genes and decrease of gene dimensionality
used for clustering samples to detect their
substructure particular methods should be applied in
microarray datasets
Subspace clustering:
To find a subset of microarray data, such that the genes are appearing as a cluster in a subspace
produced by a subset of the gene expression data.
The subset of microarray data for different subspace
clusters can be different in a subspace clustering.
Genes and sample datasets are treated symmetrically,
such that each gene or sample genes can be regarded
as clusters or Biclustering features. A particular gene
may contribute in many Pathways that may or may
not be collective under all conditions Subspace
clustering techniques confine consistently exhibit by
the blocks within gene expression datasets.
K-means Clustering K-means clustering [6] is an easy and quick
method used commonly due to its simple
implementation and small number of genetic
expression datasets. The distance between every gene
and the center of every cluster is promptly calculated
to result in a grouping of gene data to clusters where
the genes inside all clusters are as close to the center
of the clusters as possible while at the equal point there is maximal distance connecting genes for
different clusters. This method is useful if different
values of k are attempted and it only gives the
amount of clusters not the relationship between them
like hierarchical clustering. The drawbacks of this
method are the lack of prior knowledge of the
number of gene clusters in a gene expression data
which results in the changing of consequences in the
altering of microarray data in successive runs because
the original clusters are selected randomly and the
quality of the clustering has to be assessed from the
gene expression datasets.
S.
N
O
AUTHOR CLUSTERING
TECHNIQUES
CLUSTERING
ANALYSIS
1. Erfaneh
Naghieh and
Yonghong
Peng[6]
K-means Clustering,
Hierarchical
clustering, Self-
Organised Maps
Clustering (SOM)
Gene based
clustering
2. LopamudraD
ey,
Anirban
Mukhopadhy
K means, FCM,
PSO, clustering
microarray gene
expression data.
Sample based
clustering
ay[11]
3. KA Abdul
Nazeer, MP
Sebastian &
SD Madhu
Kumar
Novel Harmony
Search-K means
Hybrid (HSKH)
algorithm.
Sub space
clustering
method
4. Adil M.
Bagirov
Karim
Mardaneh[2]
K means, global k
means algorithm
Gene based
clustering
method
Fuzzy c means clustering: In fuzzy, c means[5] clustering data
elements can belong to more than one cluster, and associated with each element is a set of membership
levels of the microarray data. They indicate the
strength of the association between that data element
and a particular cluster in the gene expression data.
Fuzzy clustering [12] is a method of assigning the
gene data membership levels in the clustering, and
then using them to allocate data elements to one
cluster or more clusters in the gene expression
datasets.
S.NO AUTHOR CLUSTERING
TECHNIQUES
CLUSTER
ANALYSIS
1. Matthias E.
Futschik and
Nikola K.
Kasabov[13]
Fuzzy c-means
(FCM)
clustering,
Robust analysis
of gene
expression
Time-series.
Sample
based
clustering
method
2. LopamudraDey,
Anirban
Mukhopadhyay[12]
K means, FCM,
PSO, clustering
microarray gene
expression data.
Gene based
clustering
method
3. Xlaobo zhou,
Xiaodong wang,
Eduward R.
Dougherty, Daniel
Ruls, Eduwadr
Shu[5]
Novel clustering
strategy using
fuzzy c means.
Subspace
and sample
based
clustering
4. Matthias E.
Futschik and
Nikola K.
Kasabov[14]
A robust
analysis of gene
expression
Time-series.
Gene based
clustering
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
65
Hierarchical clustering:
Hierarchical clustering [6] is the original and
mainly common clustering method applied to gene
expression data which is developed on the basis of a single layered neural network. A hierarchical
sequence of group of clusters is generated by
grouping genes with similar patterns of expression
across a range of samples located near the microarray
data to each other. Hierarchical clustering calculates
all pairs-wise space relationships between genes and
experiments to merge pairs of values that are most
similar to the formation of a node of gene datasets.
The inter-cluster distance groups together these
clusters to make a higher level cluster in the
microarray data, which can be graphically illustrated
by a tree, called a dendrogram representing the clusters and the relationship between genes and
microarray data. This is repeated, comparing genes or
new clusters until all clusters are joined in the
microarray datasets.
Advantages of hierarchical clustering: 1. Embedded flexibility concerning the level
of granularity in the genes.
2. Simplicity of behavior in any forms of similarity or distance from the microarray data.
3. Applicability to any attributes type of
clusters.
Disadvantages of hierarchical clustering 1. Ambiguity of termination criteria of the
gene clustering.
2. Generally hierarchal algorithms do not
return to once constructed clusters with the purpose
of gene expression datasets.
S.NO AUTHOR CLUSTERING
TECHNIQUES
CLUSTERING
ANALYSIS
1. Erfaneh
Naghieh and
Yonghong
Peng[6]
K-means
Clustering,
Hierarchical
clustering, Self-
Organised Maps
Clustering
(SOM)
Sample based
clustering
Self-Organized Maps Clustering (SOM): Self organized map clustering [16] is a
logically high-speed and simple to execute a method
in the microarray data, scalable to large gene
expression datasets. It is closely related to
multidimensional scaling and its genesis to represent
all microarray data in the basis space by points in a
target gene where the distance and nearness
relationships are preserved in the clustering. At the
input, the gene expression data are presented and
output genes are organized with an example
neighborhood gene structure in the clustering. The
significant features of SOM are that it generates a high dimensional microarray dataset and places
similar clusters near each other clusters in the genes,
so that the nearest clusters in this grid are more
related than clusters that are not nearest in the
microarray data. SOM [6] is trained through
competitive learning for the distribution of the input
data set which provides a relatively robust approach
than k-means in the clustering of highly noisy data.
However SOM requires users to input the number of
clusters and the grid structure of the neuron map.
After the completion of the training, Clusters are identified by mapping all data points to the output
gene expression datasets. The drawbacks of this
method is that it is not effective since the main
interesting patterns may be merged into only one or
two clusters and cannot be identified.
Particle swarm optimization: Particle Swarm Optimization (PSO)[14]
representation be first described in 1995, as a latest method for purpose microarray data. In PSO,
particles flow in virtual gene expression data are
looking for feasible solutions in the clustering, which
in our case are the same particles in the gene datasets.
Not like older separation of the clustering algorithms
where the number of partitions genes has to be
specified earlier to run in the microarray data, PSO
algorithm is able to forming clusters on its individual
depending on the level of similarity or dissimilarity
S
.
N
O
AUTHOR CLUSTERING
TECHNIQUES
CLUSTERING
ANALYSIS
1
.
Xiang Xiao,
Ernst R. Dow,
Russell
Eberhart, Zina
Ben Miled,
Robert J.
Oppelt[16]
The rate of
convergence is
improved by adding a
conscience factor to
the Self-Organizing
Maps algorithm
It is based on
the subspace
clustering
method
2
.
Erfaneh
Naghieh and
Yonghong
Peng[6]
K-means Clustering,
Hierarchical
clustering, Self-
Organised Maps
Clustering (SOM)
Gene and
sample based
clustering
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
66
between the gene expression datasets. The PSO K-
means algorithm [18] is really different from
SOM/PSO algorithm in the microarray data sense
that the latter uses PSO to optimize the weights of
gene expression data once it has been trained with
training microarray datasets.
Limitation in clustering of micro array data
It is meaningful to cluster genes and samples in a
gene expression dataset. Owing to this special characteristics of gene expression data, and the
particular requirements from the biological domain,
gene-based clustering face several new challenges
and is still an open problem. They are:
Owing to the complex procedures of microarray
experiments, gene expression data often contains
a huge amount of noise. Therefore, clustering
algorithms for gene expression data should be
strong to handle noisy data.
Overlapping gene clusters or sample clusters plays vital role in the biological function which
is a great challenge in clustering.
Users of microarray data may not only be
interested in the clusters of genes, but also
interested in the inter and intra relationship
between the gene clusters.
References:
[1] A. Abraham, C. Grosan and V. Ramos (2006) (Eds.),
Swarm Intelligence and Data Mining, Studies in
Computational Intelligence, Springer Verlag, Germany,
pages 270, ISBN: 3-540-34955-3.
[2] Adil M. Bagirov Karim Mardaneh, Modified global k-
means algorithm for clustering in gene expression data sets,
Australian Computer Society, Inc
[3] Ben-Dor, A., Shamir, R., and Yakhini, Z. 1999. Clustering
gene expression patterns. J. Comp. Biol. 6(3–4), 281–297.
[4] D. Jiang, C. Tang, A. Zhang, “Cluster Analysis for Gene
Expression Data: A Survey”, IEEE, vol. 16, no. 11, Nov.
2004.
[5] D. Dembele, and P. Kastner, “Fuzzy C-means method for
clustering microarray data”, Bioinformatics, 19, 973-980,
2003.
[6] E. Shay, “Microarray cluster analysis and applications”,
Available at:
http://www.science.co.il/enuka/Essays/Microarray-
Review.pdf, Jan, 2003.
[7] Eisen, M. Spellman, PL, Brown, PO, Brown, D. “Cluster
Analysis and Display of Genome-wide expression
patterns,” Proc. Natl. Acad. Sci. USA 95: 14863-14868,
1998.
[8] H. Turner, T. Bailey, W. Krzanowski, Improved
biclustering of microarray data demonstrated through
systematic performance tests, Comput. Stat. Data Anal. 48
(2) (2005) 235–254.
[9] J. C. Bezdek, R. Ehrlich and W. Full, “FCM: Fuzzy c-Mean
Algorithm”, Comp. and Geo-Science, 1984
[10] J. Liu, J. Yang, W. Wang, Biclustering in gene expression
data by tendency, in: Proceedings of the 2004
Computational Systems Bioinformatics Conference (CSB
2004), 2004, pp. 1–12
[11] K. E. Parsopoulos, M. N. Vra, “Particle Swarm
Optimization and Intelligence: Advances and
Applications”, Information science reference, Hershey,
New York, 2010.
[12] K. Premalatha, A. M. Natarajan, “A New Approach for
Data Clustering Based on PSO with Local Search”,
Computer and Information Science, Vol. 1, No.4,
November 2008.
[13] N.R.Pal, J.C.Bezdek,” On cluster validity for the fuzzy c-
means model”, IEEE Trans. on fuzzy systems, pp.370-379,
1995
[14] Qiang, N., Xinjian, H., “An Improved Fuzzy Cmeans
Clustering Algorithm based on PSO”, Journal of Software,
Vol. 6, No. 5, 2011.
S.
N
O
AUTHOR CLUSTERING
TECHNIQUES
CLUSTERING
ANALYSIS
1. Ajith
Abraham,
Swagatam
Das, and
Sandip
Roy[1]
A family of bio-
inspired
algorithms, well-
known as Swarm
Intelligence (SI)
Sample based
clustering
2. V. Kumutha,
S.
Palaniammal[
14]
Particle Swarm
Optimization, Gene
Expression Data,
Fuzzy c means
Algorithm.
Gene based
clustering
3. R.
Balamurugan,
A. M.
Natarajan, K.
Premalatha[1
8]
The Particle Swarm
Optimization (PSO,
have been analyzed
for the four
benchmark Gene
expression dataset.
It also based on
the subspace
clustering
method
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
67
[15] P.T¨or¨onen, M.Kolehmainen, G.Wong, E.Castr´en,”
Analysis of gene expression data using self-organizing
maps”, FEBS Letters, Vol.451, pp. 142-146,1999
[16] S.C. Madeira and A.L. Oliveria, “Biclustering Algorithms
for Biological Data Analysis: A survey”, IEEE, vol. 1, no.
1, Jan- March 2004.
[17] K. Y. Yeung, W. L. Ruzzo. “Principal Component Analysis
for Clustering Gene Expression Data,” Bioinformatics,
17:763-774, 2001.
[18] R.Balamurugan A.M.Natarajan and K. Premalatha,
“Comparative Study on Swarm Intelligence Techniques for
Biclustering of Microarray Gene Expression Data.”, World
Acad. Sci. Eng. Technol., Int. J. Comput. Inf. Sci. Eng.
Vol.8, No. 2, pp. 4619-4625, 2014.
[19] T. Kohonen, Self-Organizing Maps, 3rd edition, New York:
Springer- Verlag, 2001
[20] Wolfgang Huber, Anja von Heydebreck, Martin Vingron,
Analysis of microarray gene expression data, April 2, 2003.
[21] X. Cui and T. E. Potok, "Document Clustering using
Particle Swarm Optimization", IEEE Swarm Intelligence
Symposium, Pasadena, California, 2005.
[22] Y. Cheng, G.M. Church. Biclustering of gene expression
data, in: Proceedings of ISMB 2000, 2000, pp. 93–103.
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
68
AUTOMATED AMBU BAG SYSTEM USED IN EMERGENCY SITUATION
Archana J 1 1*, Mary Helta Diasy 2
2
1, 2 St. Xavier’s Catholic College of Engineering, Nagercoil, India
ABSTRACT
AMBU stands for Artificial Manual Breathing Unit. Ambu bag otherwise called as manual resuscitator or self-inflating bag. It is hand held device used to provide positive pressure ventilation to a patient with breathing problem. Ambu bag is used for an artificial respiration device consisting of a bag that is squeezed by hand. Bagging is necessarily regular in medical emergencies when the patient’s breathing is insufficient and also to provide automated ventilation in preference to mouth to mouth resuscitation. In all the ambulances and in emergency wards of hospitals manually operated ambubag is used. The problems involved in this type of manual ambubag is (i) the negligence of the caretakers required quantity of oxygen is not carried over to the lungs, (ii) if the caretakers are alert one cannot assure that they expel a constant quantity of oxygen into the patient’s lungs and (iii) this type of mechanism basically is a stress to the caretakers. To overcome this problem, an automated ambu bag system is used. At the bottom of mechanical setup ambu bag is kept, cam mechanism is attached with the shaft of motor. While motor start rotating in clockwise direction cam mechanism will convert circular motion into
linear motion so that ambu bag will be compressed and expand and oxygen will be delivered to patient. Speed of the motor will be controlled by micro controller. Sensors are attached in this mechanical setup to monitor speed, pressure and oxygen flow to the patient. And speed will be displayed by LCD display. This automated ambu bag system is used for providing constant level oxygen to the needy patient.
Keywords: Ambu bag, DC motor, cam mechanism, pressure sensor, proximity sensor, LCD display and micro controller
I. INTRODUCTION
The human respiratory system allows one to obtain oxygen, eliminate carbon dioxide. Breathing consists of two phases, inspiration and expiration. Inspiration is the process of taking in air. Expiration is the process of blowing out air. Respiration rate varies according to age.
A resuscitator is a compact, portable device for on-the-scene, short term ventilator support and delivery of oxygen. It is used with a face mask and perhaps an oropharyngeal airway to keep the tongue from blocking the airflow. Also, an esophageal obdurator may be used to prevent the stomach from being inflated and to prevent regurgitation and aspiration of stomach contents. The bag resuscitator delivers a breath of oxygen to the patient when the operator squeezes the bag. Upon the release, the bag fills with oxygen again, while the patient exhales to the air through the inspiratory/expiratory (I/E) valve. The demand valve is essentially a simple artificial ventilator. It consists of a "smart"(I/E) valve coupled to an oxygen source through a flow resistance. The I/E valve is designed to open following the pressure drop of an inspiratory effort, after a built-in expiratory interval has elapsed, or when manually triggered by the operator. It remains open until a preset pressure is delivered. Two types of emergency resuscitators are shown in Figure 1.1(a) and 1.1(b) these ambu bag are varies according to the types of valve.
Other types of respiratory therapys are Gas-Delivery
Equipment, Vacuum-Delivery Systems and Suction
Devices, Humidifiers and Nebulizers, Intermittent Therapy
Equipment, Ventilators, Extracorporeal Oxygenators,
Monitoring Instrumentation, Neonatal Equipment, Patient-
Training Devices and Environmental Control Equipment.
Figure 1.1(a) Manual bag resuscitator and 1.1(b)
Demand valve
II. MATERIALS AND METHOD
Fig 2.1 shows the block diagram of the model, where
ambu bag is kept at the bottom for bagging operation.
Bagging will be done by cam mechanism. Speed of the
motor is controlled by PIC microcontroller. Proximity
sensor is used to sense the speed of the motor and that is
displayed on LCD. Pressure sensor will sense the pressure
of outlet air. And if the pressure exceeds it will produce the
alarm sound for indicating it. Heart beat sensor will
measure the heart rate of the patient at the same time. Switches are fixed for controlling the speed of the motor.
That is according to the age group the lung capacity will be
increased and the size of the ambu bag also increased.
According to the patient requirement the ambu bag must be
changed and exact button must be pressed. Incase if it is
adult, adult bag must be kept and adult button must be
pressed. Everything was controlled by microcontroller so
that ambu bag is compressed or expanded and oxygen is
delivered.
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
69
Fig 2.1 shows the block diagram of the system and the
various materials used are discussed below:
A. Ambubag:
Ambu bag otherwise called as Bag-valve-mask (BVM)
ventilation, it is an essential during emergency. This basic
airway management technique will allows for oxygenation
and ventilation of patients until more definitive airway can
be established and in some cases where endotracheal intubation or other definitive control of the airway is not
possible. For the emergency situation medical technician
will prefer this basic BVM ventilation for airway
management. In the pediatric population, BVM may be the
best preference for prehospital airway support. BVM
ventilation is also appropriate for elective ventilation in the
operating room when intubation is not required, but it is
now often replaced by the laryngeal mask airway.
The masks come in many sizes, including newborn, infant, child, and adult as the same way ambu bag size also
differ according to age. Choosing the appropriate size helps
to create a good seal and will provide effective ventilation.
Nowadays the bags are equipped with a pressure valve.
Some bags have one-way expiratory valves to prevent the
entry of room air; these allow for delivery of greater than
90% oxygen to ventilated and spontaneously breathing
patients. Bags lacking this feature deliver a high
concentration of oxygen during positive pressure ventilation
but only deliver 30% oxygen during spontaneous breaths.
B. PIC16F877A:
It has 100,000 erase/write cycle enhanced Flash
program memory. It is self-reprogrammable under the
software control. In-circuit serial programming via two pins
is possible. Low power, high speed Flash/EEPROM
technology is used here. In-circuit debugging via two pins
is possible. Programmable code protection can be done. It provides watchdog timer with its own on-chip RC oscillator
for reliable operation. It has up to 8-channel analog to
digital converter. It has wide operating voltage range.
C. Wiper Motor
This is a type of DC motor. The construction is single
speed motor. The armature with 8-slots is mounted on self-
lubricating sintered bushes. Two carbon brushes, set 180
degrees apart, rub on an 8 segment commutator generally
installed at the driving end. Two strong permanent magnets
are bonded to the steel yoke using an adhesive, which is
sometimes coated externally with non-ferrous metal to
protect it against corrosion. A steel worm, formed on the
end of the armature, drives a plastic worm wheel at a speed
of about l/10th the speed of the armature. The motor has the
output drive through a pinion gears, driven directly by the worm wheel. At the joint faces of the motor, rubber seals
are fitted to protect it from moisture. A polythene pipe is
used to vent the gases formed by arcing at the brushes.
Figure 2.2 Single-speed motor.
Typical values for wiper motor speed and hence wipe
frequency are 45 rpm and 65 rpm at normal and fast speed
respectively. The motor must overcome the starting friction
of each blade at a minimum speed of 5 rpm.
D. CAM
A ‘cam’ is a mechanical device with a surface or groove
that controls the motion of a second part called a ‘follower’
in order to convert rotary motion to linear motion.
Fig 2.3 Cam
Figure 2.3 shows the cam, this is specially designed cam.
At the bottom of cam, rubber ball is fixed to avoid the wear
and tear of the ambu bag.
E. Proximity sensors:
Proximity sensors is the metal sensing sensor. When
metal is approaching the sensor its output will be high.
When a no metal approaches the sensor its output will be
low.
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
70
F. Pressure sensor:
This is attached in the mechanical setup. It is used for
measure the pressure of expiratory air. Because the pressure
cause most dangerous situation. Patient may tend to vomit-
additional airway problems. To provide automated assist
device for breathing such that the oxygen supplied at
constant amount for the patients with pulmonary failure.
G. Heart beat sensor:
. Heart rate is calculated as the number of heart beats per
minute (beats/min). In the ECG, heart rate can be
determined by counting the number of P waves (atrial rate)
or QRS complexes (ventricular rate) in a specified time
period. One easy method involves counting the number of
QRS complexes in a 6-second period and multiplying that
number by ten. This gives the number of heartbeats (heart
rate) in 60 seconds. Six seconds in a standard ECG
comprises 30 large boxes (30 times 0.20 second per large square equals 6 seconds). Most ECGs have small vertical
hatched lines in the margins, which represent 3-second
intervals (6 seconds) and multiply by ten to determine the
heart rate per minute.
Fig 2.4 Block Diagram of Heart Beat Detection
III RESULT
Fig. 3.1 AutoCAD output
The above fig 3.1 shows the 3dimentional model of the
mechanical setup using AutoCAD software.
Fig. 3.2 Mechanical setup
As shown in fig 3.2 Cam mechanism is used and at the
end of cam rubber ball is fixed. When the motor starts
rotating in clockwise direction, it will compress and expand
the ambu bag. This rubber ball will prevents damage of
ambubag [9]. In this system we have the mechanical setup which has
wood plates at the bottom, over which the ambu bag is been
kept. The bagging operation is done with the help of cam
arrangement which is operated using DC shunt motor. The
speed of the DC motor is controlled by microcontroller
programming [7]. According to patient age (i.e) adults,
children, infants the output will be changed. Volume of air
will be different for adults, they need 12 to 16 breathing per
minute, for children they need 16 to 20 bpm, for infants 20
to 25 breathing per minute and the size of the ambu bag
varies according to the ages.
The motor speed is controlled by microcontroller. The PIC
Microcontroller board consists of circuits necessary to
operate a Microcontroller with PC interface. The board
contains provisions for interfacing 8 analog inputs and 23
Digital level signals [15].
IV. DISCUSSION
Totally 3 stepdown transformer is used. For operating
D.C motor 12 volt stepdown transformer is used. Another
12 volt transformer is used for giving the supply for all
sensors. And 5volt stepdown transformer is used for
microcontroller. ADC is used for converting analog signal
into digital signal. To activating pressure sensor AD620 IC
is used, and which needed supply of +12 and -12 volt. Driver circuit is also used.
V. CONCLUSION
This fabricated automated breathing device will have a
significant role in every ambulance and in hospitals. It may
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
71
possibly use for save the life of the patients who is suffering
from breathing disorders in a better manner when compared
with conventional type. This project provides automated
assist device for breathing such that the oxygen is supplied
at constant amount for the patients with pulmonary failure.
Hence vomiting or other complications do not occur using this automated type.
VI. REFERENCES
[1]. A Elsharydah 2002 “ Manual Resuscitators (Ambu Bags) Can
Ventilate The Lungs Adequately Despite Big Sub atmospheric
Pressure In The Breathing Circuit ” The Internet Journal of
Anesthesiology/1092406X, 20031101 Volume 7 Number 2.
[2]. Albert cook.M and John G Webster 2002 “Therapeutic medical
devices & its application and design” second edition book.
[3]. Arne Sieber, 2012 “Smart Electrochemical Oxygen Sensor for
Personal Protective Equipment” IEEE sensors journal, vol. 12,
no. 6.
[4]. C.Jacq, 2010 “Ultra-low pressure sensor for neonatal
resuscitator” Proc. Euro sensors xxiv, 532–535.
[5]. D. F. Reyes Romero, 2011” Utilizing frequency response for
medium-independent flow sensing” Procedia Engineering 25,
599 – 602.
[6]. Florian Petit, 2014 “Analysis and Synthesis of the Bidirectional
Antagonistic Variable Stiffness Mechanism” IEEE/ASME
transactions on mechatronics.
[7]. J. Capelli1, N. Ferlo1 2011 “A Novel System to Assist in
Manual Resuscitation and Detect Spontaneous Breathing” 978-
1-61284-8928-0/11/$26.00 ©IEEE.
[8]. Jean Lévine, 2004 “On the Synchronization of a Pair of
Independent Windshield Wipers” IEEE transactions on control
systems technology, vol. 12, no. 5,787-795.
[9]. Jin-Woo Jung, 2008” TX Leakage Cancellation via a Micro
Controller and High TX-to-RX Isolations Covering an UHF
RFID Frequency Band of 908–914 MHz” IEEE microwave and
wireless components letters, vol. 18, no. 10, 710-712.
[10]. Joshua Cysyk, 2011 “Rotary blood pump control using
integrated inlet pressure sensor” 33rd Annual International
Conference of the IEEE EMBS Boston, Massachusetts USA,.
[11]. J. Stocks , Ph.H. Quanjer 1995, 8, 492–506 “Reference Values
For Residual Volume, Functional Residual Capacity And Total
Lung Capacity” Eur Respir J, DOI:
10.1183/09031936.95.08030492 Printed in UK - all rights
reserved.
[12]. Jun Zhang 2012 “Self-Righting, Steering and Take-off Angle
Adjusting for a Jumping Robot” IEEE/RSJ International
Conference on Intelligent Robots and Systems. 2089-2094
[13]. Kiyotaka Ho 2009, “Fuzzy Logic Approach to Respiration
Detection by Air Pressure Sensor” fuzz-IEEE korea,911-915.
[14]. Kyoungchul Kong 2011 “A Compact Rotary Series Elastic
Actuator for Human Assistive Systems” IEEE/ASME
transactions on mechatronics, vol. 17, no. 2, 288-297.
[15]. Kyu-Chan Lee,2003” Design and Analysis of Automotive High
Intensity Discharge Lamp Ballast Using Micro Controller Unit”
IEEE transactions on power electronics, vol. 18, no. 6, 1356-
1364.
[16]. LI Faxin 2011” The Design of Parallel Combination for Cam
Mechanism” 3rd International Conference on Environmental
Science and Information Application Technology. 1343 – 1349.
[17]. Nithin.S, 2011 “Smart Grid Test Bed Based on GSM”
International Conference on Communication Technology and
System Design 258 – 265
[18]. Neil R. MacIntyre, MD 2006 “Current Issues in Mechanical
Ventilation for Respiratory Failure” Downloaded from
www.chestjournal.org at Taichung Veterans General Hospital.
[19]. Nguyen Van Tuong and Premysl Pokovny “A Case Study of
Modelling Concave Globaidal Cam” www.intechopen.com.
[20]. Olivier Legendre, 2012 “High-Resolution Micro-Pirani
Pressure Sensor With Transient Response Processing and Time-
Constant Evaluation” IEEE sensors journal, vol. 12, no. 10,
3090-3097.
[21]. Ramesh, G.Narasimharao, Robinson.A. 2010, 151-155.
Sathyabama Univ., Chennai, India , “Intelligent engine with
micro controller valve actuation and eliminating the cam
linkage arrangement” , Frontiers in Automobile and Mechanical
Engineering (FAME).
[22]. Riichiro Tadakuma 2013 “The Gear Mechanism with Passive
Rollers: The Input Mechanism to Drive the Omnidirectional
Gear and Worm Gearing” IEEE International Conference on
Robotics and Automation (ICRA).
[23]. Takeshi Takaki 2011,” High-Performance Anthropomorphic
Robot Hand With Grasping-Force-Magnification Mechanism”
IEEE/ASME transactions on mechatronics, vol. 16, no. 3,583-
589.
[24]. Yang Kaihua, 2012 “Research on Method of Fault Diagnosis
about Vehicle Wiper DC Motor” Intelligent System Design and
Engineering Application, Second International Conference.
[25]. Yunseog Hong, “Noncontact Proximity Vital Sign Sensor
Based on PLL for Sensitivity Enhancement” IEEE transactions
on biomedical circuits and systems1.
[26]. Zhenghao Ge,2010 “On the Hybrid Cam-linkage Mechanism
Realizing Variable Trajectory” International Conference on
Computer, Mechatronics, Control and Electronic Engineering
(CMCE), 272-276.
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
72
DISCRIMINATION OF SEMG SIGNALS BASED ON TEMPORAL AND
SPECTRAL APPROACH FOR FRAILTY ANALYSIS
Vidya K V1 and E. Priya
2*
1PG scholar, 2Assistant Professor
Department of Electronics and Communication Engineering, Sri Sai Ram Engineering College, Chennai, India
ABSTRACT Surface electromyography (sEMG) signal is the electrical manifestation of neuromuscular activities and depends on the anatomical
and physiological properties of the contracting muscles beneath the skin. In this work, a single channel surface EMG amplifier is designed to acquire the signals non-invasively from the skin by using bio-potential electrodes for three different hand movements. Subjects of age group 1 (18-30 years) and group 2 (60-78 years) are considered for this analysis. The recorded signals are pre-processed using Empirical Mode Decomposition (EMD) method to remove unwanted noises. Various error measures and performance index are calculated from the pre-processed sEMG signals to compare the performance of EMD with conventional digital filters. Relevant time and frequency domain features are extracted from the pre-processed signals. It is observed that the statistical analysis performed over the extracted features show
distinct variation between the age groups. Further, the classification of hand movements is done for both age groups independently using linear discriminant analysis classifier. Results demonstrate that training and testing accuracy for age group1 are 94.44% and 80.64% and that for age group2 are 93% and 76.47% respectively. Thus the methodology proposed in this work could be useful for the analysis of frailty especially for the subjects above 60 years.
Keywords: surface electromyography signal; empirical mode decomposition; digital filters; temporal and spectral features; linear discriminant analysis
I. INTRODUCTION
Frailty is a complex phenomena associated with ageing and has become a main area of research in gerontology. Frailty can be recognized by frailty phenotype and frailty index [1]. The criteria used for frailty analysis are weight loss, reduced endurance, physical slowness, muscle weakness and physical exhaustion. An individual is termed frail if at-least three of these conditions are satisfied. This work uses surface Electromyography (sEMG) signals for comparative analysis of young with elderly subjects. The sEMG signal is an electrical signal produced as a result of muscular exertion. This signal is recorded non-invasively by applying conductive elements or electrodes to the skin surface. Since sEMG signals detect the electrical activity associated with muscular exertion, they reveal important neural changes and decline in physical functions associated with ageing [2].
The sEMG signal is a very weak signal ranging from 20–500 Hz and is affected by number of noises [3-5]. Its amplitude is stochastic (random) in nature making it difficult to get the intrinsic properties of these signals from a single feature. Also the instantaneous value of the sEMG signal is not useful because of its random nature [6]. So the sEMG signals are segmented to form windows from which features are extracted. Allowable window length ranges from 50 to 400ms and the optimal window length is between 150ms and 250ms [7, 8].
Several signal analysis approaches have been reported in literature to characterize the sEMG signals [2, 9-12]. The sEMG signal analysis have wide range of applications such as prosthetic devices control, human machine interaction, functional electrical stimulation, kinematic parameter prediction, neuromuscular disease diagnosis and so on [13-16]. EMG measurements are used for the control of
prosthetic devices like artificial limbs. This involves picking up of EMG signals from the terminated nerve endings and activating an artificial arm using these signals [17].
Feature extraction is an important step in the sEMG signal processing and the classification accuracy depends mainly on the choice of feature set [17]. Waveform length (WL), time windowed Root Mean Square (RMS), Auto Regression (AR) coefficients, cepstrum coefficients, Mean Absolute Value (MAV), MAV slope, Slope Sign Changes (SSC) and zero crossings (ZC) are the useful time domain features. The common frequency domain features include mean and median frequency and spectral moments [5, 7, 16, 18].
Linear Discriminant Aanalysis (LDA) classifier has high computational efficiency for real-time operation. It is simple and has classification performance similar to more complex algorithms [18].
II. MATERIALS AND METHODS
A. EMG signal acquisition system
In this work, a single channel sEMG amplifier circuit is developed to acquire the signal. The EMG signal acquisition system consists of the important blocks shown in Fig. 1.
Fig. 1. Block diagram of EMG signal acquisition system
Electrodes EMG
amplifier
Analog to
digital
converter
PC
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
73
Fig. 2. Different hand movements used in this work
The sEMG signal is acquired from two age groups, age group 1 (18-30 years) and age group 2 (60-78 years).
Twenty volunteers, including seniors (with an average age
of 70 years) and young people (with an average age of 23
years) are recruited for this work. The subjects are asked to
perform a ‘rest-motion-rest’ pattern for three different hand
movements. The hand movements chosen are closed fist,
spherical grasp and pointing gesture as shown in Fig. 2. The
subjects are instructed to perform five repetitions of each hand motion and are allowed to rest for one minute before
performing the next hand movement.
The disposable surface electrodes are secured over the relevant muscles on the body surface, after initial skin
preparation. These electrodes pick up the voltages produced
by the contracting muscle. A single channel sEMG
amplifier consists of three electrodes, of which two
electrodes are placed over the relevant muscle and the third
act as the reference electrode. The reference electrode is
placed on the wrist and the signal electrodes on the flexor
digitorum superficialis muscle [6].
Low frequency components in the sEMG signal are
mainly noise components. Once the sEMG signals are
acquired using the pre-amplifier, RC high pass filter with a
cutoff frequency 12 Hz is used. This removes most of the
movement artifacts and low frequency noises. The high
pass filter is followed by a second stage of amplification for
bringing the signal to transistor transistor logic level. This
stage is designed to give an adjustable gain with a maximum of 20 times amplification to the filtered signal.
The bias adjustment is provided to resolve offset problems
in the amplified signal. This is used to change the reference
level of the signal.
The output of the EMG amplifier is given to the Analog to Digital Converter (ADC) whose resolution is of 10 bits.
The sampling frequency chosen in this work is twice the
highest frequency (Nyquist rate) and thus the signals are
sampled at 1000 Hz. The amplified and filtered sEMG
signals are sampled and recorded using LabView program
in PC for further processing.
B. EMG signal processing and analysis
Fig. 3. Block diagram of sEMG signal processing
The raw sEMG signal obtained after ADC conversion is pre-processed using EMD algorithm. The frequency
components below 10 Hz are removed after the
decomposition procedure. The sEMG signal is then
reconstructed with the frequency components 20-500 Hz.
The functional blocks of sEMG signal processing are
shown in Fig. 3.
Pre-processing is a crucial and essential step in sEMG
signal processing. One of the methods of pre-processing is
by using conventional digital filters but they distort the
filtered sEMG signal after pre-processing. In this work,
EMD algorithm is used as the pre-processing technique. EMD is able to deal with non-stationary and non-linear
signals thus making it suitable for sEMG signals. EMD is a
purely data driven, signal-dependent procedure and makes
no assumptions about the input signal [19].
EMD is a technique which decomposes a given signal
into a finite number of one dimensional functions called
Intrinsic Mode Functions (IMFs). Given any signal ( )x t ,
the IMFs are found by an iterative procedure called sifting
algorithm. EMD algorithm involves finding all the local
maximai
M , for 1,2,...i and minimak
m , for
1,2,...k in ( )x t and are interpolated as
( ) ( , )M i
M t f M t and ,m k
m t f m t .
Pre
-pro
cess
ing
Decomposition using EMD
algorithm
Removal of very low
frequencies
Feature extraction
Reconstruction of the signal
after noise removal
Raw sEMG
signal
Classification
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
74
These interpolated signals form the upper and lower
envelopes of the signal. The mean of these envelopes are
calculated as
2
M t m te t
and is subtracted from
the original signal as x t x t e t . The procedure is
iterated until x t remains nearly unchanged forming an
IMF. The obtained IMF t is then subtracted from the
signal x t x t t . The whole process is repeated
if x t has more than one extremum (neither a constant nor
a trend).
Large baseline wandering components are present in
higher order IMFs. Thus the de-noised signal is
reconstructed with the lower order IMFs leaving the three higher order IMFs. This reconstructed signal is used for
extracting both time and frequency domain features [20].
The performance of EMD is compared with
conventional infinite impulse response causal and non-
causal filters. Error measures such as Mean Square Error
(MSE), Root Mean Square Error (RMSE), Mean Absolute
Error (MAE), Relative Error (RE) and performance index
measure, Signal to Error Ratio (SER) are used for
validating the EMD algorithm. The mathematical
formulations are presented in Table 1, where 0 ( )x t is the
original, ( )fx t is the pre-processed signal defined for N
samples, ( )o
P f and ( )f
P f are the spectral density of
original and pre-processed signal [21].
TABLE I. MATHEMATICAL FORMULATIONS OF ERROR
MEASURES AND PERFORMANCE INDEX
Performance
measures Formula
MSE 2
1
1( ) ( )
N
o f
i
x t x tN
RMSE 2
1
1( ) ( )
N
o f
i
x t x tN
MAE
1
1( ) ( )
N
o f
i
x t x tN
RE
2
2
( ) ( )
( )
o f
o
P f P f
P f
SER
var
10 logvar
o
o f
x t
x t x t
C. Feature extraction
Feature extraction delivers useful information from
the signal. Mean absolute value gives the mean of absolute
value of signal x in a time window with N samples. It is a
simple measure to detect muscle contraction levels [22].
Mean Absolute Value (MAV) is formulated as
1
1 N
k
k
MAV xN
(1)
where kx is the thk sample in the analysis window.
Integrated Absolute Value (IAV) is defined as
1
N
k
k
IAV x MAV N
(2)
IAV is used as an onset index to detect muscle activity and
to measure total muscular effort [23]. RMS is expressed as
2
1
1 N
k
k
RMS xN
(3)
RMS is a measure of the power of sEMG signal that is
related to constant force and non-fatiguing contraction [3], [18]. Waveform length is defined as the cumulative length
of the EMG signal within the analysis window. WL
provides a measure of the complexity of the signal [24].
1
1
where
N
k k k k
k
WL x x x x
(4) (4)
Auto Regression coefficients provides information about the muscle's contraction state. AR models individual EMG
signals as a linear autoregressive time series. It is defined as
1
p
k i k i k
i
x a x e
(5)
where ia represents auto regressive coefficients, p is the
order of AR model, and ke is the residual white noise.
Zero crossing is the number of times signal x crosses
zero within an analysis window. It is associated with the
frequency of the signal. To avoid signal crossing counts due
to low-level noise, a threshold is included. The ZC count
is increased by 1 if
1 1
1
and 0 or and 0
and
k k k k
k k
x x x x
x x
(6)
Median Frequency, MDF is a frequency at which the EMG
power spectrum is divided into two regions with equal
amplitude. It can also be considered as half of the total
power. The definition of MDF is given by
1 1
1
2
MDF M M
j j j
j j MDF j
P P P
(7)
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
75
Spectral moments are the alternative possibility to extract
the features from a power spectral density and are defined
as
0
, 0,1,...,
W
m
mM f P f df m L (8)
Equation (5) defines the spectral moments derived from
power spectral density for a bandwidth W. The first three
moments of PSD are 0 1 2, and MM M
D. Classification
Classification of the three hand movements is done for
age group 1 and 2 independently after the extraction of
relevant features. The LDA classifier is used for the
classification in this work. Out of the total samples in each
hand movement, features from 70% of the samples are used as training data and the rest is used for testing. The
procedure is done for both age groups separately.
In LDA, three scatter matrices, called the within-class
wS , between-class bS , and total scatter tS matrices are
computed. It involves computing the eigen values ( )i of
tS so that1
dT
t i i i
i
S u u
. Then the eigen vectors ( )iu of
t bS S is computed where tS
denotes the pseudo inverse
of tS and 1
dT
t i i i
i
S u u
[25].
III. RESULTS
The raw sEMG signal acquired using the sEMG amplifier for group 1 and 2 are shown in Fig. 4 (a) and (b)
respectively. The signal shows a single burst acquired from
‘rest-motion-rest’ action and an offset shift in voltage.
The corresponding power spectrum density is presented
in Fig. 5 (a) and (b) respectively. The plot shows that the
dominant energy component of sEMG lies within the
frequency range 50-150 Hz. Distinct difference is observed
between the age group 1 and 2 indicating the onset of frailty
in elderly.
0 200 400 600 800 10000
1
2
3
4
5
time (ms)
sEM
G a
mpl
itude
(V
)
(a)
0 200 400 600 800 10000
1
2
3
4
5
time (ms)
sEM
G a
mpl
itude
(V)
(b)
Fig. 4. sEMG signal of age (a) group 1 and (b) 2
0 100 200 300 400 500
0
1
2
3
4
5
6x 10
-3
Pow
er/
Hz
Frequency (Hz) (a)
0 100 200 300 400 5000
1
2
3
4
5
6x 10
-3
Pow
er/H
z
Frequency (Hz) (b)
Fig. 5. Power spectral densities of age (a) group 1 and (b) 2
The sEMG signals are further decomposed by EMD
algorithm to de-noise the signal.
0 100 200 300 400 500 600 700 800 900 1000
IMF 1
IMF 2
IMF 3
IMF 9
IMF 8
IMF 7
IMF 6
IMF 5
IMF 4
IMF10
IMF11
(a)
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
76
0 100 200 300 400 500 600 700 800 900 1000
IMF 1
IMF 10
IMF 11
IMF 9
IMF 2
IMF 3
IMF 4
IMF 5
IMF 6
IMF 7
IMF 8
(b)
Fig. 6. sEMG signal decomposition into IMFs for age group (a) 1and (b) 2
The decomposed signals with eleven IMFs are shown in
Fig. 6 (a) and (b). The three higher order IMFs are excluded
to reconstruct the pre-processed sEMG signal.
Fig. 7 (a) and (b) shows the reconstructed sEMG signal
with offset being nullified and their corresponding power
spectral density is presented in Fig. 8 (a) and (b). Plots
show that the low frequency (noisy) components are
reduced after EMD pre-processing.
0 200 400 600 800 1000-3
-2
-1
0
1
2
3
time (ms)
sEM
G a
mpl
itude
(V)
(a)
0 200 400 600 800 1000-3
-2
-1
0
1
2
3
time (ms)
sEM
G a
mpl
itude
(V)
(b)
Fig. 7. Reconstructed sEMG signal of age group (a) 1and (b) 2
0 100 200 300 400 5000
1
2
3
4
5
6x 10
-3
Pow
er/H
z
Frequency (Hz)
(a)
0 100 200 300 400 5000
1
2
3
4
5
6x 10
-3
Pow
er/H
z
Frequency (Hz)
(b)
Fig. 8. Power spectral densities of reconstructed sEMG signal of age group
(a) 1and (b) 2
The EMD algorithm is validated by comparing with
conventional digital filters. Table 2, 3 and 4 shows the mean, standard deviation of the error measures and
performance index, SER. The comparison across the age
group is also presented in the table.
TABLE 2. COMPARISON OF ERROR MEASURES, MSE AND RMSE
VALUES (MEAN ± SD) FOR EMD AND CONVENTIONAL DIGITAL
FILTERS
Performance index MSE RMSE
Gro
up
1
IIR causal 0.092 ± 0.047 0.297 ± 0.071
IIR non-causal 0.0062 ± 0.004 0.075 ± 0.026
EMD 0.003 ± 0.003 0.049 ± 0.026
Gro
up
2
IIR causal 0.074 ± 0.03 0.267 ± 0.061
IIR non-causal 0.004 ± 0.001 0.061 ± 0.008
EMD 0.002 ± 0.0005 0.039 ± 0.007
TABLE 3. COMPARISON OF ERROR MEASURES, MAE AND RE
VALUES (MEAN ± SD) FOR EMD AND CONVENTIONAL DIGITAL
FILTERS
Performance
index MAE RE
Gro
up
1
IIR causal 0.211 ± 0.054 0.005 ± 0.02
IIR non-causal 0.059 ± 0.021 0.004 ± 0.02
EMD 0.043 ± 0.023 0.00052 ± 0.0006
Gro
up
2
IIR causal 0.195 ± 0.051 0.005 ± 0.003
IIR non-causal 0.048 ± 0.006 0.0037 ± 0.002
EMD 0.033 ± 0.004 0.0008 ± 0.0009
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
77
TABLE 4. COMPARISON OF ERROR MEASURES, SER VALUE
(MEAN ± SD) FOR EMD AND CONVENTIONAL DIGITAL FILTERS
Performance index SER
Gro
up
1
IIR causal 2.474 ± 0.64
IIR non-causal 15.761 ± 1.102
EMD 23.557 ± 3.171
Gro
up
2
IIR causal 2.066 ± 1.432
IIR non-causal 15.853 ± 2.56
EMD 23.372 ± 2.43
It is observed from Table 2, 3 and 4 that the error measures are found to be lower with higher SER value for
EMD based pre-processing when compared to IIR causal
and non-causal filters.
The line plot shown in Fig. 9 (a), (b) presents the average values of MAV, RMS, M0 and Fig. 9 (c), (d)
presents IAV and WL at ‘rest-motion-rest’ for age group 1
and 2 respectively.
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
Zeroth Spectral Moment
Mean Absolute Value
Root Mean Square
Ave
rag
e V
alu
es
rest max contraction rest
Group 1
hand movement (Closed fist)
(a)
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
Zeroth Spectral Moment
Mean Absolute Value
Root Mean Square
Ave
rag
e V
alu
es
rest max contraction rest
Group 2
hand movement (Closed fist)
(b)
20
40
60
80
100
120
Group 1
hand movement (Closed fist)
max contraction restrest
Ave
rag
e V
alu
es
Integral Absolute Value
Waveform Length
(c)
20
40
60
80
100
120
Group 2
hand movement (Closed fist)
max contraction restrest
Ave
rag
e V
alu
es
Integral Absolute Value
Waveform Length
(d)
Fig. 9. (a), (b) MAV, RMS, M0 and (c), (d) IAV, WL of age group 1 & 2
for closed fist hand motion
Fig. 10 (a)-(d) shows the values for maximum contractions
for the other two hand motions. It is observed that MAV,
RMS, zeroth spectral moment (M0), IAV and WL are
higher for young (group 1) compared to elderly (group 2).
0.0
0.1
0.2
0.3
0.4
0.5
Group 1
Group 2V
alu
e fo
r m
ax c
on
trac
tio
n
MAV RMS M0 (a)
0
10
20
30
40
50
60
70
80
Group 1
Group 2
Va
lue
s f
or
ma
x c
on
tra
cti
on
IAV WL (b)
0.0
0.1
0.2
0.3
0.4
0.5 Group 1
Group 2
Val
ues
fo
r m
ax
co
ntr
acti
on
MAV RMS M0
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
78
(c)
0
10
20
30
40
50
60
70
80
Group 1
Group 2
Va
lue
s f
or
ma
x c
on
tra
cti
on
IAV WL
(d)
Fig. 10. (a) , (c) MAV, RMS, M0 and (b), (d) IAV, WL of age group 1 &
2 for spherical grasp and pointing gesture
Table 5. Feature number and names notations
Feature
numbers
Feature
names
1 M0
2 M1
3 M2
4 MAV
5 RMS
6 WL
7 AR
8 MDF
9 ZC
Table 5 shows the numerical notations for different features that are used in Table 6. Table 6 shows the
classification accuracies of age group 1 and 2 for different
combinations of features.
Table 6. Classification accuracy of age group 1 and 2 for different feature
combinations
Features Classification accuracy (Percentage)
Age group1 Age group2
Training
accuracy
Testing
accuracy
Training
accuracy
Testing
accuracy
1 50 45.16 44 35.29
1,2 56.94 45.16 32 35.29
1,2,3 72.22 74.19 72 47.05
1,2,3,4 75 61.29 76 58.82
1,2,3,4,5 77.77 54.83 77 55.88
1,2,3,4,5,
6 79.16 61.29 92 55.88
1,2,3,4,5,
6,7 86.11 77.41 92.12 67.64
1,2,3,4,5,
6,7,8 93.05 80.64 92.4 76.47
1,2,3,4,5,
6,7,8,9 94.44 80.64 93 76.47
IV. DISCUSSION
Frailty is a phenomena associated with ageing. The
musculoskeletal system declines as normal ageing
progresses. The reduction in muscle mass and muscle
function is one of the outcomes of this decline [26]. The
sEMG signals are electrical manifestation of the
neuromuscular activation due to muscle contractions and,
can reveal decline in physical functions associated with
frailty [2]. There is no conclusive work reported in
literature dealing with EMD pre-processing in frailty analysis. This work also compares the performance of EMD
across age-groups.
The power spectrum density of the raw sEMG signals
(Fig. 4 (c) and (d)) shows that the dominant energy
components of sEMG are within the frequency range 50-
150 Hz. Also the power of the sEMG signal for elderly is
found to be lower compared to the young. It is observed
from the power spectral density of the reconstructed signal (Fig. 6 (c) and (d)) that the noises at lower frequencies are
considerably reduced after EMD. Also the reconstructed
signal is devoid of the offset voltage. The EMD algorithm is
validated by comparing with conventional digital filters
(Table 2, 3 and 4). The error measures are found to be
lower with higher SER value for EMD based pre-
processing when compared to causal and non-causal filters.
Error measures and SER for group 1 and group 2 are found
to be similar. This shows that EMD is able to eliminate
baseline wandering artifacts from the sEMG signals
irrespective of the age group making it a suitable pre-processing technique for sEMG signals.
MAV, IAV, RMS and WL are features that are
directly related to force of muscle contractions and the
lower values of these features in age group 2, could be due
to the weaker muscle contractions in elderly. The lower
values of moments of power spectral density and their
gradual variation with the muscle contraction in elderly
could be due to age related decrease in the number and size of fast twitch fibers in elderly. Results demonstrate that the
above observations are consistent for all the three hand
movements.
The classification of hand movements done for
both the age groups separately. The table shows that both
the testing and training accuracies are lower in age group 2.
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
79
V. CONCLUSION
It can thus be concluded that the lower sEMG
signal power in elderly is due to lower muscle contraction
and muscular effort. The MAV, IAV, RMS and WL in time
domain and spectral moments in frequency domain are
more relevant features that depict the frailty analysis
between the two age groups. Thus the methodology used in
this work appears to be suitable for analysis of muscle
weakness which is one of the most significant characteristics of frailty in elderly.
References
[1] Fried LP et.al, ” Cardiovascular Health Study Collaborative Research Group: Frailty in older adults: evidence for a phenotype.”, J.
Gerontology: Med. Sci., vol. 56A, no. 3, M146-M156, 2001.
[2] Kaitlyn P Roland, Gareth R Jones and Jennifer M Jakobi,” Daily electromyography in females with Parkinson’s disease: A potential
indicator of frailty,” Archives of Gerontology and Geriatrics, vol. 58, pp. 80-87, 2014.
[3] Carlo J. De Luca, Surface Electromyography: Detection and
Recording, DelSys, Inc, 2002.
[4] Ulvi Baspinar, Huseyin Selcuk Varol and Volkan Yusuf Senyurekj, “Performance Comparison of Artificial Neural Network and
Gaussian Mixture Model in Classifying Hand Motions by Using sEMG Signals,” Biocybernetics and Biomed. Eng., vol. 33, no. 1, pp.
33–45, 2013.
[5] Rubana H. Chowdhury, Mamun B. I. Reaz, Mohd Alauddin Bin Mohd Ali, Ashrif A. A. Bakar, Kalaivani Chellappan and Tae G.
Chang, “Surface Electromyography Signal Processing and Classification Techniques,” Sensors, vol. 13, pp. 12431-12466, 2013.
[6] Nianfeng Wang, Yulong Chen and Xianmin Zhang,” Realtime
recognition of multi-finger prehensile gestures,” Biomed. Signal Process. and Control, vol. 13 pp. 262–269, 2014.
[7] Smith L H, Hargrove L J, Lock B A, Kuiken T A, “Determining the Optimal Window Length for Pattern Recognition-Based Myoelectric
Control: Balancing the Competing Effects of Classification Error and Controller Delay,” IEEE Trans. Neural Syst. Rehabil. Eng., vol. 19,
Issue 2, pp. 186-192, April 2011.
[8] An-Chih Tsai, Tsung-Han Hsieh, Jer-Junn Luh, Ta-Te Lin,” A comparison of upper-limb motion pattern recognition using
EMGsignals during dynamic and isometric muscle contractions,” Biomed. Signal Process. and Control, vol. 11, pp. 17-26, 2014.
[9] Mojgan Tavakolan, Zhen Gang Xiao and Carlo Menon, “A
preliminary investigation assessing the viability of classifying hand postures in seniors,” BioMedical Eng. OnLine, vol. 10, no. 79, 2011.
[10] V. Monaco, A. Ghionzoli, P. Dario and S. Micera,” Features of the
EMG patterns during walking in young end elderly people: Preliminary study “Gerontechnology, vol. 7, no. 2, pp. 165-170,
2008.
[11] Malgorzata Klass, Stephane Baudry and Jacques Duchateau,”Age-related decline in rate of torque development is accompanied by
lower maximal motorunit discharge frequency during fast contractions” J. Appl. Physiology, vol.104, pp. 739-746, 2008.
[12] Roberto Merletti, Philip A. Parker, Electromyography: Physiology,
Engineering, and Non-Invasive Applications, John Wiley & Sons, 2004, pp. 419-430.
[13] Chandrasekhar Potluri, Madhavi Anugolu, Yimesker Yihun, Parmod
Kumar, Steve Chiu, Marco P. Schoen and D. Subbaram Naidu,” Implementation of sEMG-Based Real-Time Embedded Adaptive
Finger Force Control for a Prosthetic Hand”, 50th IEEE Conf. Decision and Control and European Control, ISSN : 0743-1546, pp.
861-866, 2011.
[14] N.S.K. Ho, K.Y. Tong,X.L. Hu, K.L. Fung, X.J. Wei, W. Rong and E.A. Susanto, "An EMG-driven Exoskeleton Hand Robotic Training
Device on Chronic Stroke Subjects Task Training System for Stroke Rehabilitation", IEEE Int. Conf. Rehabilitation Robotics, ISSN :
1945-7898, pp. 1-5, 2011.
[15] C. Frigo, M. Ferrarin, W. Frasson, E. Pavan and R. Thorsen,"EMG
signals detection and processing for on-line control of functional electrical stimulation", J. Electromyography and Kinesiology, vol.
10, pp.351-360, 2000.
[16] Shaikh Anowarul Fattah, A. B. M. Sayeed Ud Doulah and Marzuka Ahmed,” Evaluation of Different Time and Frequency Domain
Features of Motor Neuron and Musculoskeletal Diseases,” Int. J. Comput. Applicat., vol. 43, no. 23,pp. 34-40, 2012.
[17] R S Khandpur, Handbook of biomedical instrumentation, 2nd
ed.,Tata McGraw-Hill, 1987, pp. 178-185.
[18] Dennis Tkach, He Huang and Todd A Kuiken, “Study of stability of time-domain features for electromyographic pattern recognition”, J.
NeuroEng. and Rehabilitation, 2010, vol. 7, no. 21, 2010.
[19] Xu Zhang and Ping Zhou, "Filtering of surface EMG using ensemble empirical mode decomposition", Medical Eng. & Physics, vol. 35,
pp. 537-542, 2013.
[20] R.T. Rato, M.D. Ortigueira and A.G. Batista, “On the HHT, its problems, and some solutions,” Mech. Syst. and Signal Process., vol.
22, pp. 1374–1394, 2008.
[21] Zhou Wang and Alan C. Bovik,"Mean Squared Error: Love It or
Leave It? A new look at signal fidelity measures", IEEE Signal Processing Mag., pp.98-117, 2009.
[22] M. B. I. Reaz, M. S. Hussain and F. Mohd-Yasin, “EMG Analysis
Using Wavelet Functions To Determine Muscle Contraction,” 8th Int. Conf. e-Health Networking, Applicat. and Services, pp. 132-134,
2006.
[23] Dionysios-Dimitrios Koutsouris and Athina A. Lazakidou, Concepts and Trends in Healthcare Information Systems, Ann. Inform. Syst.
16, Springer, pp. 205-229, 2014.
[24] Angkoon Phinyomark, Chusak Limsakul, Pornchai Phukpattaranont,” EMG feature extraction for tolerance of white Gaussian noise,” Int.
Workshop and Symp. On Sci. and Technology, pp. 178-183, 2008.
[25] Mohammed Z. Al-Faiz and Sarmad H. Ahmed, “Discriminant Analysis for Human Arm Motion Prediction and Classifying”,
Intelligent Control and Automation, vol. 4, pp. 26-31, 2013.
[26] Zachary J. Palace, and Jennifer Flood-Sukhdeo, “The Frailty Syndrome”, Today’s Geriatric Medicine, vol. 7, no. 1, pp. 18, 2014.
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
80
MODEL IDENTIFICATION AND CONTROLLER DESIGN OF A PERFUSION
SYSTEM DURING AN ECMO SUPPORT
M.Dhinakaran 1*, Dr S.Abraham Lincon
2, P.Praveen Kumar
3
1Assistant Professor, Department of Instrumentation Engineering, Annamalai University, Tamil Nadu, India.
2Professor, Department of Instrumentation Engineering, Annamalai University, Tamil Nadu, India.
3M.E.Student, Department of Instrumentation Engineering, Annamalai University, Tamil Nadu, India.
ABSTRACT Extracorporeal Membrane Oxygenation (ECMO) is a temporary life support system used for patients who’s Heart or Lungs is not
working properly. ECMO is a modified form of Heart Lung Machine and it can be used for a longer period. ECMO system must be managed by a Perfusionist to maintain proper Blood flow and Blood pressure. Much of the time, the Perfusionist makes small adjustments in the ECMO system to maintain flow and pressure. This maintenance process can be tedious and is prone to human error. So by the introduction of an Automatic control in an ECMO system the variables can be perfectly controlled. In this work model identification and
controller design of a perfusion system done by real time BGA Reports collected from Various Cardiopulmonary bypass surgery Patients. For that an identified model is estimated and tested in MATLAB System Identification Toolbox.The Proposed Controller is designed and tested in MATLAB Simulink for an ECMO conditions. So this control strategy presented improves the patient’s safety.
Keywords: Extracorporeal Membrane Oxygenation ECMO, Blood Gas Analyzer BGA, Partial Pressure of Oxygen (pO2),
Partial Pressure of Carbon dioxide pressure (pCO2).
I. INTRODUCTION
Extracorporeal membrane oxygenation (ECMO) is a
unique form of cardiopulmonary bypass. It has become
standard therapy in the management of neonatal respiratory
failure in the past decade, and is also being used selectively
in the support of respiratory and cardiac failure in the
pediatric and adult populations.
Fig. 1. Blockdiagram of an ECMO Setup
The first adult ECMO was used by Hill for aortic
rupture in the year 1972 and the first report of successful
ECMO support was applied by Robert Bartlett to a newborn
with respiratory failure in the year 1975.The extracorporeal
membrane oxygenation (ECMO) apparatus consists of a venous reservoir, blood pump with tubing, a membrane
oxygenator, and a countercurrent heat exchanger. The blood
pump is either a simple roller pump (most common) or a
constrained vortex centrifugal pumps. The oxygenator is
responsible for exchanging both oxygen and carbon
dioxide. Three types of commercial oxygenator are
available are membrane oxygrnator, bubble oxygenator, and
hollow-fibre oxygenator. The heat exchanger used to warm
the blood by exposing to warm water that circulates within
metal tubing.
II. ECMO SETUP
ECMO support currently comes in two varieties:
VenoArterial (VA) ECMO, and VenoVenous (VV) ECMO.
VA ECMO - It is used to support both heart and lungs
function. Here a cannula is placed through the right jugular
vein into the right atrium.The Deoxygenated Blood is
passed to a venous reservoir and pumped by a pump
through the oxygenator. In oxygenator gas exchange occurs
by oxygen added to the blood and carbon dioxide is
removed, next, the blood is warmed by the heat exchanger
to a body temperature before returning to the patient through a cannula placed through the right carotid artery
into the aorta.
VV ECMO - It is used to support lungs function. Here a
double-lumen cannula is placed through the right jugular
vein into the right atrium. Deoxygenated blood is
withdrawn from the right atrium through the outer
fenestrated venous catheter wall, and through the inner
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
81
lumen of the catheter the oxygenated blood is returned
across the tricuspid valve.
The VAECMO and VV ECMO cannula placement in Heart
is shown in Figure2
Fig. 2. Types of ECMO
III. ECMO MONITORIG AND CONTROL
ECMO Process Monitoring and Control done by the
coordination between perfusionist and anesthetist,
Fig. 3. Chid placed in a VAECMO setup
ECMO Monitoring and Measurement Includes: Perfusion pressure, venous return, Urine output, Temperature, Blood
gas, Electrolytes, ACT (Clotting time),Partial pressure of
Oxygen PO2 and Partial Pressure of Carbon dioxide
PCO2,ECG, EEG, Pump speed etc
ECMO Important Control Includes:
PO2 - obtained by varying the combination of FiO2 and
Medical air
PCO2 - Obtained by Blood/Gas Flow
Hemoglobin saturation - achieved by the increase in blood
flow or Hemoglobin Concentration
Anticoagulation - Obtained by increase in Heparin level Platelets - Achieved by Platelets Transfusion
Oxygen uptake -Varied by changing the temperature
Etc..
IV. BLOOD GAS ANALYZER
The major function of the pulmonary system is to
deliver oxygen to cells and remove carbon dioxide from the
cells. Blood Gas Analyzer tests are performed to assess
respiratory status because it helps to evaluate gas exchange
in the lungs. BGA test can also measures the working
condition of lungs and kidneys. Blood gas analyzer present
in JIPMER Hospital Puducherry is shown in Figure 4.
Fig. 4. Blood Gas Analyzer (Model RapidLab-348)
BGA Performs the diagnostic test on blood drawn from
venous side or arterial side. Some of the components
measured by BGA are PH: measures hydrogen ion concentration in the blood, it
shows acidity or alkalinity of blood.
(Normal range - 7.35 to 7.45)
PCO2: It is the partial pressure of CO2 that is carried by the
blood for excretion by the lungs, known as respiratory
parameter.
(Normal range - 35to 45 mmhg)
PO2: It is the partial pressure of O2 that is dissolved in the
blood and it reflects the body ability to pick up oxygen from
the lungs
(Normal range - 80 to 100 mmhg)
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
82
HCO3: it reflects the kidney’s ability to retain and excrete
bicarbonate, known as the metabolic parameter,
(Normal range - 22 to 28 meq/L)
V. REAL TIME DATAS OF CPB SURGERY PATIENTS
Various CPB Surgery Patients BGA Reports are
collected . This arterial blood gases report is measured for
every fifteen seconds with an ECMO support. The input
level and output level simultaneously changes, which
depends on patients age and their physical conditions. This
patients ABGA report is completely analyzed by
comparing with past recovered patients records.
TABLE I. AVERAGE OF BGA PATIENTS REPORTS COLLECTED
VI. PATIENT DATA ESTIMATION USING SYSTEM
IDENTIFACTION TOOL BOX
For the Average patient data and with the suggestions
from perfusionist and Specialzed Doctors’ a strong process
model at various time delay is designed by using system
identification toolbox in matlab environment.
Model Identification of Patients Data: Initially ten patients
ABGA Report is analyzed and its average taken. This
average data is used for designing the linear model of the
system with the system identification toolbox in MatLab
environment.
Inspired oxygen gases FiO2and pressure of oxygen pO2 can
be modeled as the linear transfer function. The Oxygen
Plant is given as
)1( 059.177.1
2
3322.735.59
2)(
SS
S
OsG
Total blood flow rate and pressure of carbon dioxide pCO2
can be modeled as the linear transfer function. The Carbon
dioxide Plant is given as
)2( 059.1S77.1
2S
08023.0S087.3
2CO)s(G
VII. CONTROLLER DESIGN
The Conventional Controller is used for Controlling
CPB Gases. So here the PID Controller is chosen as a
Conventional Controller . The PID controller takes the present, the past, and the future error into considerations.
Fig. 5. Block diagram of Control mechanisms of the CPB Gases
PID controller is constructed from various combinations
of proportional, integral and derivative in terms of patients
Number
of
Patients
Input Output
FiO2
[%]
Blood
gases
Flow[LPM]
pO2
[mmHg] pCO2
[mmHg]
Patients 1
42.22222 1.780556 316.825 40.9
Patients2 48.61111 1.620278 312.0833 43.2
Patients 3 42.22222 1.780556 316.825 40.9
Patients 4 47.33333 1.652333 313.2688 42.63
Patients 5 42.05556 1.692389 315.6396 41.48
Patients 6 42.08333 1.707083 315.8767 41.36
Patients 7 44.13889 1.732472 315.442 41.58
Patients 8 45.05556 1.686389 314.4542 42.05
Patients 9 44.08796 1.705532 315.0864 41.75
Patients 10 44.09524 1.709381 315.1372 41.72
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
83
measured BGA report model which is required to meet
specific performance requirement functions. The General
formula of the PID Transfer function given as
)3( d
K S
iK
pK)s(GC
Controller selection:
ECMO Support : The PI Control mode is selected
CPB Surgery : PID Control mode is selected
VIII. PID CONTROLLER RESPONSES OF CPB GASES
DURING AN ECMO SUPPORT
0 100 200 300 400 500 600 700 800 9000
50
100
150
200
250
300
350
Time (sec)
Pressure of O
xygen (pO
2)
pO2
Fig. 6. Partial pressure of oxygen during an ECMO
Support
0 100 200 300 400 500 600 700 800 9000
10
20
30
40
50
60
Time (sec)
Pre
ssure
of
Carb
an d
ioxid
e (
pC
O2)
pCO2
Fig. 7. Partial pressure of Carbon dioxide during an ECMO Support
IX. PID CONTROLLER RESPONSES OF CPB GASES
DURING CPB SURGERY
0 20 40 60 80 100 120 140 160 180 2000
50
100
150
200
250
300
350
pO
2 [
mm
Hg
]
Time (Sec)
PID Controller Responses for CPB Gases
Fig. 8. Partial pressure of oxygen during CPB Surgery
0 20 40 60 80 100 120 140 160 180 2000
5
10
15
20
25
30
35
40
45p
CO
2 [
mm
Hg
]
Time [Sec]
PID Controller Responses for CPB Gases
Fig. 9. Partial pressure of Carbon dioxide during CPB Surgery
X. CONTROLLER RESULTS
TABLE II. CONTROLLER PARAMETERS FOR AN ECMO
SUPPORT
Controller Parameters pO2 pCO2
Proportional Gain (Kp) 0.017 3
Integral Gain (Ki) 6.18 9
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
84
TABLE III. CHARACTERISTICS OF CONTROLLERS FOR
ECMO SUPPORT
Controller Characteristics pO2
[mmHg]
pCO2
[mmHg]
Maximum Saturation limit 600 55
Set Point (mmHg) 315 45
Over shoot (%) 13.5 13.24
Settling Time (sec) 80 10
Rise Time (sec) 35 2.5
TABLE IV. CONTROLLER PARAMETER FOR CPB
SURGERY
Controller Parameters pO2 pCO2
Proportional Gain (Kp) 0.69 1.23
Integral Gain (Ki) 2.65 0.04
Derivative Gain (Kd) 0.0084 -0.17
TABLE V. CHARACTERISTICS OF CONTROLLERS
FOR CPB SURGERY
Set point
pO2
mmHg
Time (sec) Controller Characteristics
min max
Over
shoot
(%)
Rise
time
(sec)
Settling
Time
(sec)
150 0 15 5.3 0.5 2.7
280 15 30 5.1 15.6 17.2
300 30 45 3.2 30.4 31.9
320 45 60 2.7 45.4 46.6
310 60 75 2.1 60.2 62.1
300 75 90 2 75.2 76.2
250 90 105 1.8 90.3 91.8
300 105 120 1.5 105.2 106.2
320 120 135 1.4 120.3 121.1
300 135 150 0.7 135.2 136.09
TABLE VI. CHARACTERISTICS OF CONTROLLERS
FOR CPB SURGERY DISCUSSION
Real time BGA Reports collected from Various
Cardiopulmonary bypass surgery Patients. For FiO2 and
Partial pressure of oxygen PO2, the Blood Gases flow and
Partial pressure of carbon dioxide PCO2 is Identified into
second order model and it is developed in MATLAB
System Identification Toolbox. The Proposed Controller is
designed and tested in MATLAB Simulink for an
ECMO/CPB Surgery conditions.
XI. CONCLUSION
The main proposal of this research is to analyze the Real
time BGA Reports collected from Various Cardiopulmonary
bypass surgery Patients and Controller Characteristics are
obtained for ECMO conditions. This controller
characteristics can used in future developments of robustic
algorithm for an ECMO support.
Acknowledgment
We wish to express our sincere thanks to Mr Mathavan
Senior Perfusionist CTVS Department JIPMER Hospital
Pudhucherry and Mr Yogesh Perfusionist (Heart Lung
Machine) KMCH Hospital Erode. Who provide us valuable
guidance and timely help for carrying this Research work.
Set Point
PCO2
mmHg
Time (sec) Controller Characteristics
min max
Over
Shoot
(%)
Rise Time
(Sec)
Settling
Time
(Sec)
40.9 0 15 2.9 3.8 4.2
43.2 15 30 0.2 15.2 15.6
40.9 30 45 0.08 30.2 30.6
42.62 45 60 0.04 45.1 45.3
41.47 60 75 0.01 60.07 60.1
41.36 75 90 0.08 75.02 75.07
41.57 90 105 0.03 90.00 90.02
42.05 105 120 0.001 105.0 105.05
41.74 120 135 0 120.0 120
41.72 135 150 0 135 135
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
85
References
[1] Berno J.E. Misgeld, Steffen Leonhardt and Martin Hexamer
“Multivariable control design for artificial blood-gas exchange
with heart-lung machine support” 2012 IEEE International
Conference on Control Applications (CCA) Part of 2012 IEEE Multi-
Conference on Systems and Control October 3-5, 2012. Dubrovnik,
Croatia
[2] B. J. E. Misgeld J. Werner M. Hexamer “Robust and self-tuning
blood flow control during extracorporeal circulation in the
presence of system parameter uncertainties”Medical & Biological
Engineering & Computing 2005, Vol. 43
[3] M.Dhinakaran, S.Abraham Lincon, “A Study and Development of
Auto Tuning Control in a Perfusion System for Extracorporeal
Membrane Oxygenator” International Journal of Computer
Applications (0975 – 8887) Volume 106 – No. 16, November 2014
[4] M.Dhinakaran, S.Abraham Lincon, “Perfusion System Controller
Strategies during an ECMO Support” International Journal on
Soft Computing (IJSC) Vol. 5, No. 3, November 2014
[5] M.Dhinakaran, S.Abraham Lincon,P.Praveen Kumar “Modeling and
Controller Design of Perfusion System for Heart Lung Machine”
International Conference on Current Trends in Engineering and
Technology, ICCTET’14 IEEE 2014IEEE - 33344July 8, 2014,
Coimbatore, India,
[6] Hexamer M.,& Werner(2003). “A Mathematical model for the gas
transfer in an oxygenator” IAFC conference on modeling and
control in biomedical systems (pp.409-414).Australia: Melbourne.
[7] Cardiopulmonary Anatomy & PhysiologyEssentials for
Respiratory Care, 4th EditionTerry DesJardins COPYRIGHT ©
2002 by Delmar, a division of ThomsonLearning
[8] Introduction to Extracorporeal Circulation Maria Helena L.
Souza &Decio O. Elias
[9] Scott I. Mars’, Robert H. Bartlett’, Janice M. Jenkins3, & Pierre
Ababa. Controller design for extracorporeal life support (1996)
18th Annual International Conference of the IEEE Engineering in
Medicine and Biology Society, Amsterdam 1996 .1733-1735.
[10] Hammon J Wi . Extracorporeal Circulation: Perfusion System
Cohn Lh, ed. Cardiac Surgery in the Adult. New York: McGraw-Hill,
2008:350-370.
[11] Aidan O'Dwyer “Handbook of PI and PID Controller Tuning
Rules” 2nd Edition Published by Imperial College Press
[12] Mohammad H.Moradi “PID Control New Identification and
Design Methods” Springer-Verlag London Limited 2005
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
86
A TECHNIQUE TO IMPLEMENT BIMODAL BIOMETRICS FOR
PREVENTING INFANT MIX-UPS USING RASPBERRY PI
S.Sivaranjani1*
, Dr. S.Sumathi 2
1 M.E Scholar, Embedded System Technologies, Sri Sai Ram Engineering College, Chennai, India.
2 Associate Professor, ECE, Sri Sai Ram Engineering College, Chennai, India.
ABSTRACT In developing countries the problem of swapping and abduction of new born is a challenging issue and occurs all over the
world. Traditional methods do not provide required level of security for the newborn. Hence, a newborn personal authentication system is been proposed for this issue based on multi biometrics. In this paper, a bimodal biometric system is been proposed to authenticate the newborn with their mother, based on footprint of newborn and finger print of their mother. This concept is further been enhanced by developing a prototype to be implemented on a Raspberry Pi a single board computer. The Raspberry Pi has System on chip (SoC) denoted as Broadcom BCM2835 SoC, ARM1176JZFS application processor. It increases performance, consumes less power, reduces overall system cost and size. The Raspberry Pi is been controlled by a modified version of Debian Linux optimized for ARM architecture. The required algorithm for processing images for biometric authentication is been performed using open source OpenCV library in Linux platform. Thereby the proposed system improves the security system in
hospitals / birth centers and also provides a low cost solution to the newborn Mix-Ups rather than the expensive DNA procedures. In this paper, discusses about the research works carried on hardware as a biometric module to enhance the performance of a standalone device.
Keywords: OpenCV, Footprint, Fingerprint, Raspberry PI, ARM1176JZFS;
I. INTRODUCTION
Biometric system is a pattern-recognition system
recognizes a person based on feature vector derived
from a specific biological characteristics such as
Physiological biometric identifiers include fingerprints,
hand geometry, ear patterns, eye patterns (iris and
retina), facial features, and other physical
characteristics. Behavioral identifiers include voice,
signature, key stroke, and others.
The Present method of footprint, fingerprint acquisition
in hospitals is inked footprint of the newborn along with
the fingerprint of the mother. This is stored in a file
which forms the medical database. This method of
image acquisition is offline. Another method of practice
is to tie a number band around the hands/legs of the
newborn as a measure of identity. This number band is
same as the one which is also tied to the mother of the
infant. At the time child kidnapping or abduction,
mixing of babies, multiple claims for an infant in any hospitals, birthing centers causes emotional breakdown
and confusion. This raises a question on the
effectiveness of the offline method and the method of
tying number bands (ID bands). This eventually leads to
the DNA test at times. Hence, biometrics can be used to
solve such identity issues.
In the online system, by a digital source and computers
are used for processing and storage. The newborn’s footprint images captured using a high resolution
camera, whose type is Canon EOS 7D. The fingerprint
of the newborn’s mother is also collected by means of a
fingerprint scanner. Bimodal authentication provides
better security when compared with unimodal
authentication. Further, implementation of bimodal
authentication in hardware as embedded system
enhances the overall performance of the system as a
standalone device.
A complex IC that integrates the major functional
elements such as programmable processor, on-chip
memory, accelerating function hardware eg: GPU, both
hardware and software, analog components into a
single chip or chipset is called system on chip (SoC).
Thus, reduce overall system cost, increase
performance, lower power consumption and reduce
size. Raspberry Pi has (SoC) denoted as Broadcom
BCM2835 SoC multimedia processor. The chip used is
ARM1176JZF-S 700MHz, RISC architecture and draws low power. The Raspberry Pi is to build a low-
cost standalone device. It works in Linux platform, In
this paper, discusses about the implementation of
bimodal biometric (fingerprint and footprint) in
Raspberry Pi. Debian Linux platform optimized for
ARM architecture is been used. An open source
computer vision OpenCV is used for biometric
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
87
programming. It has C++, C, Python and Java
interfaces and supports Windows, Linux, Mac OS, iOS
and Android. It takes the advantage of the hardware
acceleration of the heterogeneous compute Platform.
The OpenCV library is installed from repository by
following few commands to execute the installation of
dependencies, downloading and decompression of
OpenCV, compilation and installation of OpenCV in
Debian Linux. The programming is written in text file
with .cpp extension, the text file is added as executable
files in CMake file and executed using GCC compiler.
Appropriate algorithms are used for reliable
recognition process of Newborn is employed.
II. RELATED WORKS
Some related Research studies have been done on the
field of implementing biometrics into hardware and
Biometric recognition studies related to Infant Mix-
Ups.
S.Balameenakshi, S.Sumathi describes about Biometric
recognition of Newborns Identification using
Footprints concept using Matlab and Labview
software. It uses match score level fusion [7].
Mariano Fons, et.al. Describes about design of
embedded fingerprint matcher system also about the
hardware software co-design of a computational platform responsible for matching two fingerprint
minutiae sets. A novel system concept is suggested by
making use of reconfigurable architectures (FPGA) [8].
Jang-HeeYoo et.al. Describes the design of embedded
biometric systems that identify person by using face-
fingerprint or iris-fingerprint multimodal biometrics
technology and to implement real-time system, the
most time consuming biometric algorithms are implemented in a field programmable gate array. The
hardware platform to demonstrate biometric algorithms
are consisted of a low-power ARM920T core processor
(400MHz). The biometric algorithms have been
effectively optimized for embedded systems and the
results show the possibility of the real-time application
of stand-alone and mobile biometrics [16]. The digital
image processing algorithms in embedded systems
increasingly employ DSP chip or FPGA modules. The
whole system is verified using ARM GNU C under the
embedded Linux operating system.
Sung Bum Pan et al. describes about a VLSI
implementation of minutiae extraction for secure
fingerprint authentication. It presents a system on chip
implementation of the fingerprint feature extraction
algorithm. It is been developed for SoC targeted for
ARM CPU and AMBA bus can also be extended for
many other smart card configurations[17].
Maritane Barrenechea et al. describes about a low-cost
fingerprint minutiae extraction and matching system
based on a Spartan3 family FPGA with embedded
Leon2 open core processor. The architecture
incorporates a floating point unit and a discrete Fourier
transform coprocessor to accelerate the minutiae
extraction process [18].
Stevan o. n. silva, lucianosilva discuss about A
reduced structure of system was created following
the standards of the FHS, containing the necessary
system utilities, supporting file handling, device
configuration, network settings and support for runtime configuration. The structure still takes the
required libraries for applications that use OpenCV.
Startup scripts were rewritten to set up an
environment for implementation of OpenCV and to
start necessary services. Thus, the development of
an embedded architecture based on the Linux
system and the OpenCV library, focused on the
effective use of the hardware, through the selection
of specific features required by this library and aims
for performance improvement[1].
Priyanga .M, Raja ramanan.V discuss about
Unmanned Aerial Vehicle for Video Surveillance
Using Raspberry Pi. Wherein it describes about
PuTTY software is a free and open-source terminal
emulator, serial console .PuTTY's Key Generator is
broken into three main functions: generating,
importing, and exporting keys [3]. PuTTY
generator known as PuTTYGen will automatically
generates a key and provides the key pair for future file transfer.
R N daschoudhary, rajashreetripathy discusses
about real time Face detection and tracking the head
poses position’s from high definition video using
Haar Classifier through Raspberry Pi BCM2835
CPU processor which is an combination of SoC
with GPU based Architecture. It provides only 4.5
frames per second on 4-core CPU it becomes too slow to process HD stream in real time [9].
Therefore, a solution to this problem is been
discussed by parallel modification of OpenCV
algorithm for GPU.
Patricia Melin, Diana Bravo, and Oscar Castillo
describes about pattern recognition using modular
neural networks with a fuzzy logic method for
response integration. It proposes a new architecture for modular neural networks for achieving pattern
recognition in the particular case of human
fingerprints. Response integration is based on the
fuzzy Sugeno integral to combine the outputs of all
the modules in the modular network [13].
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
88
Ajay Kumar, Sumit Shekhar describes about the
Personal recognition using rank level combination
of multiple palm print representations. It describes
about nonlinear rank level fusion which is helpful
in combining multibiometrics biometric fusion [14].
Pankaj Bhowmik et al. Discusses about Fingerprint
Image Enhancement and its Feature Extraction for
fingerprint recognition in OpenCV (visual c++) for
extracting the minutiae points where a curve track
finishes, intersect with other track or branches off.
A critical step in studying the statistics of
fingerprint minutiae is to reliably extract minutiae
from the fingerprint images [4]. The paper discuss
about image enhancement techniques prior to minutiae extraction to obtain a more reliable
estimation of minutiae locations.
Md Kawser Jahan Raihan et al. describes about
Raspberry Pi image processing based economical
automated toll system. In developing countries
RFID for each car does not exist. RFID is still a
costly solution. Hence, a image processing
technique to detect license plate for auto toll system. The raspberry pi is been used a
minicomputer has the ability of image processing
and control a complete toll system [15]. Thus, it
implies that Raspberry pi provides low-cost
solution for image processing computations.
G.Senthil Kumar et al. describes about embedded
image capturing system using raspberry pi system. A system is been implemented by considering the
requirements of image capturing using Raspberry
pi. Experimental results have shown that designed
system is fast enough to run the image capturing,
recognition algorithm and data stream can flow
smoothly between the camera and Raspberry Pi
board [19].
J. Fierrez-Aguilar, L. -M. Munoz-Serrano, F Alonso-Fernandez and J. Ortega-Garcia discuss
about the effects of image quality degradation on
the verification performance of automatic
fingerprint recognition is investigated. It studies the
performance of two fingerprint matchers based on
minutiae and ridge information under varying
fingerprint image quality. The ridge-based system is
found to be more robust to image quality
degradation than the minutiae-based system for a
number of different image quality criteria is been
concluded [11].
Kazuki Nakajima describes about Footprint-Based
Personal Recognition wherein an input pair of raw
footprints is normalized, both in direction and in
position for robustness image-matching between the
input pair of footprints and the pair of registered
footprints[12]. In addition to the Euclidean distance
between them, the geometric information of the
input foot print is used prior to the normalization,
i.e., directional and positional information. In the
experiment, the pressure distribution of the footprint was measured with a pressure-sensing
mat.
Shepard et.al. (1966) collected footprint of 51
newborns analyzed the sample and were only able
to identify babies, resulting in approximately 20%
identifiable footprints. However it was felt that the
majority of these 20% correctly matched prints
would not stand up under legal scrutiny in courts.
Singh et.al. Describes minutiae-based fingerprint
minutiae matching algorithm to solve two
problems correspondence and similarity
computation. For the correspondence problem,
assigns each minutia two descriptors texture-based
and minutiae-based descriptors and use an
alignment-based greedy matching algorithm to establish the correspondences between minutiae,
extracts a 17-D feature vector from the matching
score using support vector classifier. The
algorithm is tested on FVC2002 databases.
Ajay Kumar, Sumit Shekhar describes about the
Personal recognition using rank level combination
of multiple palm print representations. It describes
about nonlinear rank level fusion which is helpful in combining multi biometrics biometric fusion
[14].
Koichi Ito et al. proposed an efficient fingerprint
recognition algorithm using the phase-based image
matching. The proposed technique is particularly
effective for verifying low-quality fingerprint
images.
III. PROPOSED WORK
The required database is been collected for
authentication consists of 6 samples of same newborn
footprint and corresponding 6 samples of their mothers
fingerprint. Then the collected samples undergoes 5
main steps namely (1) Image Acquisition (2) Image
enhancement (3) Binarization (4) Thinning (5) Feature
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
89
Extraction. The extracted features in the form of
template are stored in database known as enrollment
stage. When the input image is given to the proposed
system it undergoes all the preprocessing stage
followed by pattern matching stage finally a decision is
been made based on match score if the given input is
authenticated or not. Implementation of mothers fingerprint recognition in Raspberry Pi is been
employed. All the data required for such recognition is
been stored in SD card of 16 GB (varies depending
upon application) memory. Initially the hardware is
been booted in Linux platform via SD card ideally pre-
installed with NOOBS (New Out Of the Box Software).
Linux terminal can be used to write the command to
download and install open source tool for image
processing called OpenCV library via Keyboard
connected to USB port of Raspberry Pi. OpenCV is a
set of libraries targeted for real time computer vision applications. The programming related to biometrics is
processed on ARM Processor. The result of the
processing after decision level is been displayed on
monitor via HDMI port of Raspberry Pi. Thus, this
hardware helps in developing a chip by verifying and
implementing it using Raspberry Pi and helps in
implementing it as an Application specific integrated
circuits (ASIC).
Advances in Computer Vision have introduced a set of
useful features namely localization, pattern recognition,
counting of objects and people, face recognition tracking
also in the field of Biometrics. The OpenCV library
allows these features to be implemented on computers
with relative ease, but general purpose hardware prevents
a large number of applications, makes it too expensive,
with large physical size, high power consumption and low
tolerance to vibration and changes in environmental
temperature [1]. OpenCV program, which is integrated
with Raspberry Pi to build a low-cost stand-alone device is been employed. The Raspberry Pi is controlled by a
modified version of Debian Linux optimized for ARM
architecture. The core frequency is been set to 700MHz.
GPU is capable of Blu-ray quality playback, using H.264
at 40Mbits/s. It has GPIO where its header consists of 17
pins [3]. The front End Graphical interface can be used for
developing application in C++ and cross compiled for
ARM architecture.
Stressing this hardware by selecting specific features of
the system that provide support to applications based on
this library in order to minimize the processing overhead
and reduce the use of computing resources such as main
and storage memories [1] is been employed. Raspberry Pi
board depicted in fig 1.
Fig 1: The Raspberry Pi board
Source: www.raspberrypi.org
3.1. OPENCV LIBRARY
The OpenCV (Open Source Computer Vision Library) is
a Computer Vision library written in C and C++ compatible with Linux. It has been designed to be
computationally efficient, with a strong focus on real-
time applications and additionally it takes the advantages
of CPUs with multiple cores [Bradski, &Kaehler, 2008].
The purpose of this library is that it provides a simple
computer vision infrastructure to prototype quickly for
sophisticated applications. It has over 2500 optimized
algorithms, includes both a set of classical algorithms and
the state of the art algorithms in Computer Vision, which
is been used in image processing, detection and face
recognition, object identification, classification actions, traces, and other functions. The library is used
extensively by companies like Google, Yahoo, Microsoft,
Intel, IBM, Sony, Honda, Toyota, and startups area as
Applied Minds, Video Surf and Zeitera, and research
groups and government .OpenCV provides better
portability [1].
3.2 EMBEDDED VISION: Fingerprint recognition in
Raspberry Pi
The boot of raspberry pi is done via SD card ideally
preinstalled with NOOBS and the option of Debian Linux
is been chosen during boot process. Few commands have been used in Linux terminal to install OpenCV, GCC,
Numpy, Scipy, Matplotlib, Cmake, files necessary for
developing biometrics Programming on Debian Linux
Platform.
$ Sudo apt-get update
$ Sudo apt-get upgrade
$ Sudo apt-get install build-essential $Wget fossies.org/linux/misc/opencv-2.4.9.zip
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
90
$Sudo apt-get installlibav format-devlibgtk 2.0-dev pkg-
configCmake-libswscale-dev bzip2
$unzip opencv-2.4.9
$cd opencv-2.4.9.zip
$mkdir build
$cd build
$Cmake –D CMAKE_BUILD_TYPE=BUILD –DCMAKE_INSTALL-PREFIX=/URS/LOCAL-D
$sudoviopencv.conf I <
/etc/ld.so.conf.d/ /usr/local/lib Esc: wq
$Make
$Sudo make install
$sudoldconfig
$sudoviopencv.conf /etc/bash.bashrc.
$apt-get install matplotlib
All the libraries are downloaded via Ethernet port of
raspberry Pi connected to internet. Results of biometric
implementation of both using OpenCV with .cpp text file
and with python shell by importing OpenCV library is
been implemented. In case of executing it in python shell
additional dependencies are required to be installed such
as matplotlib, scipy, math, numpy. Tkinter tool which is inbuilt in python is been used for creating GUI.
Reliability of fingerprint recognition depends on the
performance of minutiae extraction algorithm relies on
the input quality of images. Thus, enhancement of image
is processed by Histogram equalization, this method
usually increases the global contrast of the image, and the
intensities can be better distributed on the histogram by effectively spreading out the most frequent intensity
values.
Fig 2: Depicts the Histrogram Equalization plot of a
finger print in Raspberry pi. Fig 3: shows the enhanced
and thinning of a fingerprint using OpenCV in Pi board
where threshold is set to 127/255. Fig 5 shows the
Adaptive equalized image of a fingerprint.
Fig 2: Histogram equalization of finger print.
Fig 3: Enhanced and thinning of a fingerprint using OpenCV.
The analysis of images are further done by extracting the
high frequency components of a fingerprint is been
shown in fig 4.
Fig 4: High frequency component of a fingerprint image.
Fig 5: Adaptive Equalized image of a fingerprint.
After image enhancement process the image undergoes
binarization of an image. Binarization is really an
important step in the process of ridge extraction. The
image is divided into blocks (16X16) matrix and the mean intensity value is calculated, if each pixel values in
the matrix is larger than mean intensity value of current
block then the pixel value is been made high ‘1’
otherwise, considered as low ‘0’. Thinning of an image is
been performed which is total of one pixel width. The
thinning of a fingerprint result using morphological
operation is been shown in fig 6.
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
91
Fig 6: Binarization and Thinning of a fingerprint.
It is been further analyzed that Guo hall’s parallel
Thinning Algorithm in OpenCV provides better in
detecting the end points in images. Fig 3 shows the
fingerprint thinning using Guo hall’s Algorithm using
OpenCV in c++. Parallel thinning with two sub-iteration
algorithms is described below:[20]
3x3 Template Used For Pixel Removal
p8 p1 p2
p7 P p3
p6 p5 p4
Algorithm: Let C (P) be the number of distinct 8-
connected components of 1's in Ps 8-neighborhood.
N(P) =Min(N1(P), N2(P))
N 1(P) = (p1vp2)+(p3vp4)+(p5vp6)+(p7vp8)
N2 (P) = (p2vp3)+(p4vp5)+(p6vp7)+(p8vp1)
An edge point will be deleted if it satisfies: a) C(P)=l;
b) 2<=N(P)<=3;
c) Apply one of the following:
1) (P2vP3vP5)*P4=0 in odd iterations; or
2) (P6vP7vP8)*P8=0 in even iterations
Where "v" expresses the logic "OR" operation. C(P)=1
Means P is 8-simple.
There is only one group of 8-connected 1's around P. Due
to this condition, deletion of P will not break the
connectivity of the elements in the 3X3 window under processing. Condition (a) guarantees P is not a break
point. The use of N (P) allows one to identify the end
points if or not they have one or two 1's 8-neighbors [20].
To obtain unique features of fingerprint minutiae is been
calculated and stored for template match. After the
process of minutiae extraction false H breaks (False
minutiae) in fingerprint is been removed using Euclidean
distance algorithm. Fig 7 shows the minutiae extraction
of a fingerprint image using python-OpenCV interface.
Fig 7: Minutiae Extraction of a fingerprint.
Minutiae extraction works after image thinning process.
Different types of minutiae are ridge ending and
bifurcation. These minutiae are detected using the
concept of Crossing Number (Cn) based on Rutovitz’s
definition of crossing number for a pixel P is given
below[21]:
Where Pi is the binary pixel value in the neighborhood of
P with Pi = (0 or 1). The crossing number Cn (P) at a
point P is defined as half of cumulative successive differences between pairs of adjacent pixels belonging to
the 8- neighborhoods of P. If Cn(P) = = 1 it’s a ridge
ending and if Cn(P) = = 3 it’s a ridge bifurcation point. A
3X3 matrix slide over the thinned fingerprint image to
detect candidate minutiae. For example, if the matrix
matches the pattern shown in fig 8, it can be concluded
that whether P is a ridge ending or bifurcation.
Fig 8: Two sample 3X3 matrix pattern of ridge ending and bifurcation.
The ridge endings are marked in blue color and
bifurcation with red color in programming result of
which is shown in figure 9. After detecting all the
possible minutiae, the next step is to validate all minutiae to eliminate false minutiae before matching stage.
Fig 9 also depicts the filtered minutiae after the removal
of false minutia structures (e.g. spikes, bridges, holes,
breaks, spurs, ladder structures) by keeping minimum
Euclidean distance equal to 10.0.
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
92
Fig 9: Shows the Minutiae detection and filtered minutiae of a fingerprint in Raspberry Pi.
Biometric systems can be configured to make a match or
no match decision, based on a predefined number,
referred to as a threshold, which establishes the
acceptable degree of similarity between the trial template
and the enrolled reference template. After the
comparison, a score representing the degree of similarity
is generated and this score is compared to the threshold to make a match or no-match decision.
Similarly, the work on foot print images has to be
employed and feature vector has to be calculated and
appropriate fusion schemes as to be employed. The six
samples of images has been collected out of which 3 are
trained images and remaining 3 are considered to be
testing images, based on which a statistics as to be evaluated based on False Acceptance Rate (FAR) and
False Rejection Rate (FRR). Such that genuine
acceptance rate can be calculated which is equal to 1-
FAR.
IV. CONCLUSION
This paper discusses the implementation of low cost ambient new born authentication System based on
biometric traits of the Infant mothers fingerprint. On
implementing this concept on Raspberry Pi further
enhances the performance. The biometric processing is
been employed by installing open source OpenCV library
into a Linux platform. Appropriate algorithm for
fingerprint enhancement, feature extraction is been
employed in Raspberry pi. Hence this method is a low
cost solution to the newborn swapping rather than the
expensive DNA procedure. A review on different
hardware implementation as a biometric module also been discussed.
References
[1] Stevan O.N. Silva, Luciano Silva ”A Linux Microkernal Based Architecture For Opencv In The Raspberry Pi Device”
International journal of scientific Knowledge (IJSK), June 2014. Vol.5, No.2.
[2] Ricardo Neves, Anibal C. Matos “Raspberry PI Based Stereo
Vision For Small Size ASVs” INESC TEC, 2014.
[3] Priyanga .M, Raja ramanan . V “ Unmanned Aerial Vehicle for video surveillance Using Raspberry Pi” International journal of
Innovative Research in Science, Enginerring and Technology,
March 2014.
[4] Pankaj Bhowmik, Kishore Bhowmik, Mohammad NurulAzam,
Mohammed WahiduzzamanRony “ Fingerprint Image Enhancement And its feature Extraction for recognition”,
International Journal of Science and Technology Research Volume 1,Issue 5, June 2012.
[5] Anil jain, Karthik Nandakumar and Arun Ross,”score
Normalisation in multimodal biometric systems”, Pattern Recognition 38(2005)2270 2285.
[6] S. Zhang, R janaKiraman, T. Sim and S. Kumar, “Continuous
Verification Using Multimodal Biometric System” Proc. Second Int’l conf. Biometrics, pp.562-570,2006.
[7] S.Balameenakshi. S.Sumathi, “Biometric Recognition of
Newborns Identification using Footprints”, Proceedings of 2013 IEEE International Conference on Information and
Communication Technologies. Tamilnadu, India .
[8] Mariano Fons, Francisco Fons, Enrique Cantó,” Design of an Embedded Fingerprint Matcher System”, 2006 IEEE.
[9] R.N daschoudhary, rajashree tripathy, “Real time Face Detection
and tracking using haar classifier on SOC, SARC-IRF International conference ,12th April 2014, New Delhi,India.
[10] Koichi Ito, Ayumi Morita, TakafumiAok, Tatsuo Higuchi Hiroshi Nakajima$, and Koji Kobayashi “A fingerprint Recognition
Algorithm using Phase-Based Image Matching for Low-Quality Fingerprints”,2005,IEEE.
[11] J. Fierrez-Aguilar, L. -M. Munoz-Serrano, F Alonso-Fernandez
and J. Ortega-Garcia, ”On the effects of image Quality Degradation on minutiae and ridge-based automatic fingerprint
recognition”,2005.
[12] Kazuki Nakajima ”Footprint –Based Personal Recognition “,IEEE Transactions on Biomedical Engineering, vol
47.No.11,November 2000.
[13] Patricia Melin, Diana Bravo, and Oscar Castillo “Fingerprint Recognition using modular neural networks and fuzzy integrals
for response Integration”, International joint conference on neural networks, Montreal,Canada,july 31- august 4 2005.
[14] Ajay Kumar, Sumit Shekhar “ Palmprint Recognition using Rank
Level fusion”, 2010 IEEE 17th International Conference on image processing , September 26-29,2010,Hong Kong.
[15] Md. Kawser Jahan Raihan, Mohammad Saifur Rahaman,
Mohammad Kaium Sarkar and Sekh Mahfuz “Raspberry Pi Image processing Based Economical Automated Toll system” Global
journal of researches in engineering electrical and electronics Engineering. Vol 13 issue 13 version 1.0 year 2013.
[16] Jang-Hee Yoo, jong-Gook Ko, Yun-Su chung, Sung-Uk Jung, Ki-Hyun Kim, Ki-Young Moon anf Kyoil Chung. ETRI-information
security research division. “ Design of embedded multimodal Biometric systems” IEEE.
[17] Sung Bum Pan, Darsung Moon, KichulKim, Yongwha Chung “
A VLSI implementation of minutiae extraction for secure fingerprint authentication” IEEE 2006.
[18] Maitane Barrenechea, jon Altuna, Miguel San Miguel “A low-
Cost FPGA-based embedded fingerprint verification and matching system” signal theory and communications group, department of
electronics university of Mondragon, spain.
[19] G.Senthil Kumar, K.Gopalkrishnan, V.Sathish Kumar “ embedded image capturing system using Raspberry Pi sytem” International
journal of emerging treand and technology in computer science. April 2014.
[20] Harish Kumar and Paramjeet Kaur, ”A comparative study of
iterative Thinning Algorithm for BMP images” International journal of computer Science and information Technologies
(IJCSIT) 2011.
[21] Wen Liu, Nigel whyte “ Fingerprint Recognition ” Research Manual , institute of technology CARLOW, 2009 .
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
93
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
93
FEATURE EXTRACTION AND COMPARISION OF MOTOR ACTIVITIES
FOR CURSOR CONTROL
Sasweta Pattnaik1, Manasa Dash
2, Sukant Sabut
1*
1Department of Electronics & Instrumentation Engineering
Institute of Technical Education & Research
SOA University, Bhubaneswar, Odisha
2Department of Mathematics
Silicon Institute of Technology
Bhubaneswar, Odisha *Corresponding author
ABSTRACT
Brain-computer interface (BCI) is a hardware and software communication system that permits cerebral activity alone to control computers or external devices. The immediate goal of BCI research is to provide communications capabilities to severely disabled people who are totally paralyzed or ‘locked in’ by neurological neuromuscular disorders, such as amyotrophic lateral sclerosis(ALS), brain stem
stroke, or spinal cord injury. The use of Electroencephalogram (EEG) signals in the field of BCI has obtained a lot of interest with diverse applications ranging from medicine to entertainment. One major challenge in current BCI research is how to extract features of time-varying EEG signals and classify the signals as accurately as possible. In this paper, we have extracted the features of EEG signal obtained from the database of BCI project by discrete wavelet transform (DWT) for backward imagery and backward movement with task of left and right hand. The results were compared for mean, standard deviation and peak power values.
Keywords: Brain Computer Interface (BCI); Electroencephalograph (EEG); Wavelet Transform, Movement with task,
Imagery Movement
I. INTRODUCTION
In BCI (Brain Computer Interface), people can send
information simply by thinking. This is used to enable
communication for severely disabled user. All interfaces
(such as keyboards, mice, or voice recognition systems)
require motor movement. So, BCIs are designed to enable
communication for severely disabled people like Spinal
Cord Injuries.
The electrophysiological phenomena
investigated most in the quest for an automatic
discrimination of mental states are event-related potential
(ERP) [1], and localized changes in spectral power of
spontaneous EEG related to sensorimotor processes [2, 3].
Beta is the brain wave usually associated with active
thinking, active attention, and focus on the outside world or solving concrete problems. The original EEG signal is time
domain signal and the signal energy distribution is
scattered. The signal features are buried away in the noise.
In order to extract the features, the EEG signal is analyzed
to give a description of the signal energy as a function of
time or/and frequency. Based on previous studies, features
extracted in frequency domain are one of the best to
recognize the mental tasks based on EEG signals [4]. The
first analysis method was the Fast Fourier Transform (FFT)
by applying the discrete FFT to the signal and finds its
spectrum. As EEG signals are non-stationary, traditional
feature extraction method such as amplitude-frequency analysis based on FFT only uses amplitude and frequency
information and not time domain information. However,
much research has shown that the information of time
domain is very important for improving the classification performance of EEG signals. Wavelet decomposition is a
good time-frequency analysis method since wavelets allow
decomposition into frequency components while keeping as
much time information as possible. Event-related
Desynchronization (ERD) and Event-related
Synchronization (ERS) are the underlying
neurophysiological phenomena accompanying motor
imagery. It provides the theoretical basis of the motor
imagery-based BCI research [5-7]. The rhythm µ (8-12 Hz)
and β (18-26 Hz) originating in the sensorimotor cortex
have been postulated to be good signal features for EEG-
based BCI [6, 7]. Ales et al. [8] used both DFT and DWT method for feature extraction, classified the features
extracted from both the methods by neural network
classifier and compared them. They found DWT is proved
the efficient method as compared to DFT for feature
extraction. Hence the wavelet based feature extraction is
considered as the best when compared with any other
transform based feature extraction. Various frequency
domain feature extraction methods show that the wavelet
decomposition method is more powerful for extraction than
of DFT and STFT because of its multi-resolution
characteristics [9]. In wavelet transform we get the localized information for low frequencies (high time
resolution) and in the transient events (high frequency
resolution). Homri et al. applied the Daubechies, Coiflet
and Symmlet wavelet families to a dataset of MI to extract
features and describe right and left hand movement imagery
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
94
[10]. The current work is the extraction of features from
EEG recorded signal and its statistical analysis for the
classification of signal for BCI application.
II. DESCRIPTION OF DATASET
The first step in BCI systems is the data collection and
filtering. The datasets used are the dataset 2-2D motion of
EEG motor activity dataset of the BCI project. The data are
recorded from a 21 year old, right handed male with no known medical conditions. The EEG consists of actual
random movements of left and right hand recorded with
eyes closed. Each row represents one electrode.The order of
the electrodes is FP1 FP2 F3 F4 C3 C4 P3 P4 O1 O2 F7 F8
T3 T4 T5 T6 FZ CZ PZ. The recording was done at 500Hz
using Neurofax EEG System which uses a daisy chain
montage. Two experiments were carried out by taking a
single channel (C3) EEG. First One is the comparison
between left hand backward imagery movement and
backward movement with task. And second one is the
comparison between right hand backward imagery movement and movement with task.
III. METHODS:FEATURE EXTRACTION BY DISCRETE
WAVELET TRANSFORM (DWT)
The key feature of wavelets is the time-frequency
localization. It means that most of the energy of the wavelet
is restricted to a finite time interval. When compared to
STFT, the advantage of time-frequency localization is that
wavelet analysis varies the time-frequency aspect ratio,
producing good frequency localization at low frequencies
(long time windows), and good time localization at high
frequencies (short time windows). This produces a
segmentation, or tiling of the time-frequency plane that is
appropriate for most physical signals, especially those of a transient nature. The wavelet technique applied to the EEG
signal will reveal features related to the transient nature of
the signal, which are not obvious by the Fourier transform.
In numerical analysis and functional analysis, a discrete
wavelet transform (DWT) is any wavelet transform for
which the wavelets are discretely sampled. In general, it
must be said that no time-frequency regions but rather time-
scale regions are defined. The DWT of a signal is
calculated by passing it through a series of filters. First the
samples are passed through a low pass filter with impulse
response g resulting in a convolution of the two. The signal is also decomposed simultaneously using a high-pass filter
h. The output provides the detail coefficients (from the
high-pass filter) and approximation coefficients (from the
low-pass). It is important that the two filters are related to
each other and they are known as a quadrature mirror filter.
All wavelet transforms can be specified in terms of a low-
pass filter g, which satisfies the standard quadrature mirror
filter condition,
G(Z)G(Z-1)+G(-Z)G(-Z-1)=1 ………………(1)
where G(z) denotes the z-transform of the filter g. Its
complementary high-pass filter can be defined as,
H(Z)=ZG(-Z-1) ………………(2)
DWT employs two sets of function, scaling functions and
wavelet functions, which are related to low-pass and high-
pass filters, respectively. The decomposition of the signal
into the different frequency bands is merely obtained by
consecutive high-pass and low-pass filtering of the time
domain signal. The procedure of multi-resolution
decomposition of a signal x[n] is schematically shown in Figure1. Each stage of this scheme consists of two digital
filters and two down-samplers by 2. The first filter, h [.] is
the discrete mother wavelet, high-pass in nature, and the
second, g[.] is its mirror version, low-pass in nature. The
down-sampled outputs of first high-pass and low-pass
filters provide the detail, D1 and the approximation, A1,
respectively. The first approximation, A1 is further
decomposed and this process is continued [11].
Selection of suitable wavelet and the number of
decomposition levels is very important in analysis of
signals using the DWT. The number of decomposition
levels is chosen based on the dominant frequency
components of the signal. The levels are chosen such that
those parts of the signal that correlates well with the
frequencies necessary for classification of the signal are
retained in the wavelet coefficients.
Fig.1. Sub-band decomposition of DWT implementation;
h[n] is the high-pass filter, g[n] the low-pass filter
IV. RESULTS AND DISCUSSION
The recorded single channel EEG (C3 channel) signal is
taken from database, the 50Hz transmission line frequency
is removed by notch filter, bandpass filtered between 0.5 to
30 Hz, and the result is represented in figure 2 and 3. The
pre-processed signal is decomposed using discrete wavelet
transform (db4) which is shown in figure 4 and 5. Table 1
presents the frequencies of different levels of wavelet
decomposition on the EEG signals with a 500 Hz sampling rate. The µ rhythm (8-12 Hz) and β rhythm (18-26 Hz)
* Corresponding author e-mail address
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
95
originating in the sensorimotor cortex were considered to be
good signal features for EEG-based BCIs. The Mean and
standard deviation are calculated and results are compared
for leftbackward imagery Vs leftbackward movement with
task. The result also compared for rightbackward imagery
Vs right backward movement with task, which is given in
table 2, 3 and the peak power values are given in table 4, 5
for backward imagery and backward movement with task
for left and right hand respectively. These feature vectors,
calculated for the sub-bands D5 and D4 reflecting the µ and
β rhythms, respectively, were used for the classification of EEG signals. The tests are done with different types of
wavelets, and the one that gave maximal efficiency was
selected for the particular application.
We have taken Daubechies wavelet of order4 [12].The
level of decomposition is chosen upto 6 for 500Hz
sampling frequency for backward imagery and backward
movement with task. Table 2 shows further reduction of the
feature vector of extracted features. Mean represents Frequency distribution of signal and standard deviation
represents the amount of variability in frequency
distribution. It is found out that the mean ,standard deviation
and peak power of backward movement with task EEG is
high than that of EEG of backward imagery movement in
D4 level (β band).
(a)
(b)
Fig.2. (a) EEG of leftbackward imagery (b) Leftbackward movement with task
(a)
(b)
Fig.3. (a) Rightbackward imagery and (b)Rightbackward movement with task
Fig.4.Wavelet decomposition of leftbackward imagery
Movement
0 1000 2000 3000 4000 5000 6000 7000 8000-50
0
50
100
time in ms
Am
plit
ude
raw EEG
0 1000 2000 3000 4000 5000 6000 7000 8000-2
0
2
4
time in ms
Am
plit
ude
Bandpass filtered EEG
0 500 1000 1500 2000 2500 3000 35000
50
100
time in ms
Am
plit
ude
raw EEG
0 500 1000 1500 2000 2500 3000 3500-4
-2
0
2
time in ms
Am
plit
ude
Bandpass filtered EEG
0 1000 2000 3000 4000 5000 6000 7000 80000
50
100
time in ms
Am
plit
ude
raw EEG
0 1000 2000 3000 4000 5000 6000 7000 8000-4
-2
0
2
4
time in ms
Am
plit
ude
Bandpass filtered EEG
0 500 1000 1500 2000 2500 3000 3500-50
0
50
100
time in ms
Am
plit
ude
raw EEG
0 500 1000 1500 2000 2500 3000 3500-2
-1
0
1
2
time in ms
Am
plit
ude
Bandpass filtered EEG
0 1000 2000 3000 4000 5000 6000 7000 8000-5
0
5Bandpass filtered eeg signal
0 1000 2000 3000 4000 5000 6000 7000 8000-1
0
1x 10
-3 Detail D1
0 1000 2000 3000 4000 5000 6000 7000 8000-0.05
0
0.05Detail D2
0 1000 2000 3000 4000 5000 6000 7000 8000-0.1
0
0.1Detail D3
0 1000 2000 3000 4000 5000 6000 7000 8000-1
0
1Detail D4
0 1000 2000 3000 4000 5000 6000 7000 8000-2
0
2Detail D5
0 1000 2000 3000 4000 5000 6000 7000 8000-1
0
1Detail D6
0 1000 2000 3000 4000 5000 6000 7000 8000-0.5
0
0.5Approximation A6
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
96
Fig.5.Wavelet decomposition of leftbackward movement with task
TABLE.1. Frequencies corresponding to different levels of
decomposition for daubechies wavelet of order 4 with a sampling
rate 500HZ
TABLE.2.Leftbackward imagery Vs Leftbackward
Movement with task using mean and standard deviation
TABLE.3. Rightbackward imagery Vs Rightbackward Movement
with task using mean and standard deviation
TABLE.4. Leftbackward imagery Vs Leftbackward
Movement with task using Peak Power
Leftbackward
imagery
Leftbackward
Movement
Peak power
in D5(µv2)
67 11
Peak power
in D4(µv2)
4.4 4.7
TABLE.5. Rightbackward imagery Vs Rightbackward
Movement with task using Peak Power
Rightbackward imagery
Rightbackward Movement
Peak power in D5(µv2)
62 20.08
Peak power in D4(µv2)
6.37 7.04
V. CONCLUSION
The work has proven that the features based on µ and β
rhythms which are very effective for motor imagery
classification are the reduced feature vector as input to the
classifier. The features are extracted from the EEG data by
DWT. They are good features that can improve the classification of mental tasks. We can say as detection
techniques and experimental designs improve, BCI will
improve as well and provide economic alternatives for
individuals to interact with their environment. EEG based
BCI are safe and low invasive. From the result analysis, we
concluded that the wavelet feature extraction scheme is best
suited for EEG feature extraction in BCI applications.
Further study is to analyze the classification accuracy of the
extracted features.
Acknowledgment
The EEG data were recorded for the initial work in the IMS and SUM Hospital, Bhubaneswar; thanks to Dr. Subhranshu Jena and the EEG record technician for their support to provide the EEG data.
0 500 1000 1500 2000 2500 3000 3500-5
0
5Bandpass filtered eeg signal
0 500 1000 1500 2000 2500 3000 3500-5
0
5x 10
-3 Detail D1
0 500 1000 1500 2000 2500 3000 3500-0.05
0
0.05Detail D2
0 500 1000 1500 2000 2500 3000 3500-0.5
0
0.5Detail D3
0 500 1000 1500 2000 2500 3000 3500-1
0
1Detail D4
0 500 1000 1500 2000 2500 3000 3500-2
0
2Detail D5
0 500 1000 1500 2000 2500 3000 3500-0.2
0
0.2Detail D6
0 500 1000 1500 2000 2500 3000 3500-0.5
0
0.5Approximation A6
Frequency
band(Hz)
Decomposition
levels
EEG Rhythms
125-250 D1 noise
62.5-125 D2 noise
31.25-62.5 D3 noise
15.62-31.25 D4 β
7.81-15.62 D5 µ
3.91-7.81 D6 θ
0-3.9 A6 δ
Leftbackward
imagery
Leftbackward
Movement
Frequency
Bands
Mean ± S.D. Mean ± S.D.
D4 0.1135± 0.1548 0.1619± 0.2174
D5 0.3112± 0.4129 0.2117± 0.2934
D6 0.1220 ± 0.1628 0.0464± 0.0587
A6 0.2239± 0.1079 0.2068± 0.0429
Rightbackward
imagery
Rightbackward
Movement
Frequency
Bands
Mean ± S.D. Mean ± S.D.
D4 0.1298 ± 0.1881 0.2103± 0.3112
D5 0.3573 ± 0.4603 0.2121± 0.2964
D6 0.1386 ± 0.1883 0.0520± 0.0755
A6 0.2858 ± 0.0749 0.0278± 0.0326
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
97
References
[1] L.A. Farwell, E. Donchin, “Talking off the top of your head toward a
mental prosthesis utilizing event-related brain potentials”,
Electroencephalography and Clinical Neurophysiology, vol. 70(6), pp.
510-523, 1988. [2] G. Pfurtscheller, C. Neuper, D. Flotzinger, M. Pregen-zer, “EEG-based
discrimination between imagination of right and left hand movement”,
Electroencephalography and Clinical, Neurophysiology, vol. 103(6),
pp. 642-651, 1997.
[3] J.R. Wolpaw, D.J. McFarland, “Multichannel EEG-based brain-
computer communication”, Electroencephalography and Clinical
Neurophysiology, vol. 90(6), pp. 444-449, 1994.
[4] G.N.Molina, T.Ebrahimi, J.M. Vesin, “Joint Time-Frequency-Space
Classification of EEG in a Brain-Computer Interface Application”,
EURASIP Journal on Applied Signal Processing ,vol.7,pp.713-729, 2003.
[5] G. Pfurtscheller, “Functional brain imaging based on ERD/ERS”,
Vision Research, vol.41 (10-11). pp.1257-1260, 2001.
[6] G. Pfurtscheller, SFH. Lopes, “Event-related EEG/MEG
synchronization and desynchronization: basic principles”, Clinical
Neurophysiology , vol.110(11), pp.1842-1857, 1999.
[7] G. Pfurtscheller, C. Brunner, A. Schlaogl, SFH. Lopes, “ Mu rhythm
(de)synchronization and EEG single-trial classification of different
motor imagery tasks”, NeuroImage , vol 31(1), pp.153-159, 2006.
[8] A. Prochazka, J. Kukal, O. Vysata, “Wavelet Transform Use for
Feature Extraction and EEG Signal Segments Classification”, Int.
Symp on Communication, Control and Signal Processing,vol.5,pp. 719 - 722, 2008.
[9] C.E. Mohan, S.V. Dharani, “Wavelet based feature extraction scheme
of electroencephalogram”, Int Journal of Innovative Research in
Science, Engineering and Technology, vol.3 (1), pp.1-6, 2014.
[10] I. Homri, S. Yacoub, N. Ellouze, “Optimal segments selection for
EEG classification”, 6th Int Conf on Sciences of Electronics,
Technologies of Information and Telecommunications (SETIT),
Sousse, Tunisia, pp.817-821,2013.
[11] P. Jahankhani, V. Kodogiannis, K. Revett, “EEG Signal
Classification Using Wavelet Feature Extraction and Neural
Networks”, IEEE International Symposium on Modern Computing,
pp. 52-57, 2006.
[12] Y. Fang, X. Zheng, “Feature Extraction of Time-Amplitude-
Frequency Analysis for Classifying Single EEG”, Journal of Fiber
Bioengineering and Informatics, vol.7.2, pp. 261-271, 2014.
2nd
International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
98
NON-INVASIVE DEVICE FOR MEASUREMENT OF GLUCOSE AND
HAEMOGLOBIN IN BLOOD
K.M.Nivetha1,K.Pavithra
2,N.Indhujha
3,D.Arulkumar
4
1,2,3 UG students, Electronics and Communication Engineering,Panimalar institute of technology(Anna
University),Chennai,Tamilnadu,India
4 Assistant Professor,Electronics and Communication Engineering,Panimalar institute of technology(Anna
University),Chennai,Tamilnadu,India
ABSTRACT
Diabetes has evolved as a major muscle eating disease which leads to the reduction of hormone erythroprotein responsible
for red blood cell count in our body.This reduction of red blood cell leads to anemia,stroke and other major
disease.Therefore maintaining of glucose and haemoglobin level is mandatory to avoid diabetes,other short term and long
term disease.Present available invasive technologies for measuring glucose and haemoglobin involves frequent pricking of
fingers to take the blood sample which increase patient’s discomfort and loss of blood.In our project the proposed non
invasive method uses NIR-occlusion spectroscopy to measure the glucose and haemoglobin level in blood.Therefore this
method reduces patient’s discomfort and compliance.It also supports effective monitoring of blood and glucose and
haemoglobin level in blood.
Keywords: Diabetes,Anemia, NIR-occlusion spectroscopy
INTRODUCTION
Diabetes is now becoming a fastest growing
disease among old aged and middle aged people. As of 2014, an estimated 387 million people have
diabetes worldwide, with type 2 diabetes making
up about 90% of the cases. This is equal to 8.3% of
the adult population, with equal rates in both
women and men.Diabetes, often referred to by
doctors as diabetes mellitus, describes a group of
metabolic diseases in which the person has high
blood glucose (blood sugar), either because insulin
production is inadequate, or because the body's
cells do not respond properly to insulin, or both. Patients with high blood sugar will typically
experience polyuria (frequent urination), they will
become increasingly thirsty (polydipsia) and hungry (polyphagia).
Diabetes mellitus often described by three types
namely:Diabetes-Type 1, type 2and gestational
diabetes.
Type 1 DM results from the body's
failure to produce enough insulin. This form
was previously referred to as "insulin-
dependent diabetes mellitus" (IDDM) or
"juvenile diabetes". The cause is unknown
Type 2 DM begins with insulin resistance,
a condition in which cells fail to respond to
insulin properly.] As the disease progresses a
lack of insulin may also develop. This form
was previously referred to as "non insulin-
dependent diabetes mellitus" (NIDDM) or
"adult-onset diabetes". The primary cause is
excessive body weight and not enough
exercise.
Gestational diabetes, is the third main
form and occurs when pregnant women
without a previous history of diabetes develop
a high blood glucose level.
A normal blood sugar during fasting(i.e.,8 hours
before eating )should be 70-99mg/dl.Normal blood
sugar two hours after eating should be less than
140mg/dl.[5]
About 25 percent of people with diabetes have
some level of anemia. In anemia, there are fewer
red blood cells than normal, resulting in less oxygen being carried to the body’s cells. People with anemia often feel tired or weak and may have
difficulty getting through activities of daily living.
Other symptoms include paleness, poor appetite,
dizziness, lightheadedness, rapid heartbeat, and
shortness of breath. Because these symptoms can
2nd
International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
99
also be associated with diabetes, they are
sometimes not recognized as evidence of
anemia.Anemia may occur with diabetes because
the hormone that regulates red blood cell
production, erythropoietin (EPO), is produced by
the kidneys. Kidney damage at several levels is a complication of diabetes, and one problem often
leads to the other. Changes in the kidneys that
occur with diabetes range from diabetic
nephropathy all the way to chronic kidney disease.
Early detection and treatment is essential to prevent
or delay disease progression. If anemia is present ,
the blood glucose tests may not be accurate.
Studies that looked at blood glucose monitor
accuracy found that low hematocrit levels can
falsely increase glucose measurements, leading to
monitor test results as much as 20 percent too high. All glucose monitors are designed to measure the
level of glucose in the blood, but, unfortunately,
not all blood is the same. A major difference between blood samples is the percentage of red
blood cells, or hematocrit. The average hematocrit
for men is slightly higher than the average for
women. Young children tend to have a lower
hematocrit than adults. As people age, hematocrit
values usually are lower.Low hematocrit is a
common side effect of many illnesses and of drug
therapies like metformin. Reductions in kidney
function that occur in diabetes can also cause lower
hematocrit values.Normal range for haemoglobin:
In man:13.5-18gm/dl
In woman:11.5-16gm/dl
New born:14-24gm/dl
EXISTING METHOD
Currently, blood glucose and haemoglobin can only be monitored through the use of invasive
techniques. The risk of infection and measurement
inaccuracy are present with all of the invasive
techniques. In addition, discomfort also caused by
the invasive method.
Fig1.1:Haemoglobin
Fig1.2:Glucose
AVAILABLE TECHNOLOGIES There are many non invasive techniques like
near-infrared and Raman spectroscopy,
polarimetry, light scattering, photoacoustic
spectroscopy, polarization technique, mid-
infrared spectroscopy etc are available to measure the glucose level in blood.Pulse oximetry technique
is the present available method to measure
haemoglobin.But there is no non invasive device
for both glucose and haemoglobin measurement.
PROPOSED METHOD The proposed method uses the principle of
NIR-occlusion spectroscopy[6].Due to the use of
this ooclusion principle for glucose and
haemoglobin measurement effective results are
obtained .
Fig.2 Basic Block Diagram
2nd
International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
100
The figure shows the block diagram of the
device. It consists of two NIR reflective sensors[8][2]
correspondingly in the range of 940nm and
870nm.Switches are placed before sensors to select
the reqired sensor When the finger is placed on the
sensor unit, the IR rays emitted by the sensor (LED) passes through the finger and the reflected
waves from the tissue are received by the sensor
(photodetector). Photodetector convert the light
signal into the electrical signal which is given to
the low pass filter. Low pass filter filters the noise
from the detector signal and it is then amplified.
This amplified signal is given into the
microcontroller unit. The microcontroller
PIC16F877A contains an inbuilt Analog to Digital
Converter and it converts the analog signal into
digital values. These digital values are displayed in
the seven segment LCD display. For each parts to operate, 5V power supply is provided.
WORKING PRINCIPLE
The sensing unit consists of two sensors
namely 940nm for glucose measurement and
870nm for haemoglobin measurement.during
measurement the user selects the required switch to
drive the corresponding sensor.The rays from the
sensor pass through the finger. The photodetector
detects the emitted ray from the finger. Occlusion
is the technique of using a pneumatic cuff to resist
blood flow for the minimum amount of time
possible.The flow is resisted with an amount of
pressure that doesn’t exceed the systolic pressure. Occlusion process amplifies the signal amplitude,
decrease the motion noise effect and discriminate
overlapping wavelengths. This enhances the
sensitivity to glucose and Haemoglobin.It helps in
producing the accurate result. NIR spectroscopy
involves sending the infrared rays to a greater depth
in to the skin to measure the required cells in the
blood.The penetration depth varies from 1 to
100mm.to have deeper penetration the emitter is
conneted to a driver circuit.
There are three light matter interactions. They are,
1. Absorption of Infra Red ray
2. Reflectance
3. Scattering
The first interaction absorption is most
preferable[9].The output level depends on the
remaining amount of IR rays that comes out after absorption by the blood cells In contrast to other
configurations, we use transmitted light in addition
to reflected light for glucose and haemoglobin
measurement. In the transmission mode the light
traverses the whole organ (finger), and the photons
typically encounter many more glucose and
haemoglobin molecules along their paths than in
the reflection mode. This enhances the sensitivity
to glucose and haemoglobin and clinically relates
to an average over a whole organ rather than a
highly localized measurement. In addition, the
influence of local factors such as skin morphology and pigmentation is reduced.
. The output radiation is processed by the
microcontroller.The principle measurement of
Haemoglobin is based on the fact of substantial
absorption/transmission of near infrared rays through oxygenated hamoglobin(HbO2).
Fig3:IR interaction with molecules in blood
ADVANTAGE
1. No loss of blood
2. Less discomfort and pain
3. Frequent monitoring of glucose and
haemoglobin levels
4. No pricking of finger
APPLICATION
1. Continuous monitoring of glucose level.
2. Detection of pre diabetes
3. Detection of anemia,hyperglycemia and
hypoglycemia.
4. Used to get faster result
CONCLUSION
The initial problem was the use of invasive and
minimally invasive methods for blood glucose
concentration measurements. To address this problem, this project uses near-infrared as a
possible means to measure blood glucose and Hb
levels. This implementation served to be non-
invasive. To implement the near-infrared
spectroscopy, light emitting diodes that emit light
2nd
International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
101
at 940nm and 870nm were chosen and used. To
detect the reflected light, a photodiode is used. The
peak wavelength of glucose and Hb reflective
sensors are 940nm and 870nm approximately. To
test the device, measurements are taken from
various individuals with low and high blood glucose as well as haemoglobin levels. The voltage reading from the photodiode is amplified and
filtered using a non inverting amplifier and is
converted into values by the microcontroller and
the output is viewed in the display. Using near-
infrared light of wavelength 940nm, minimum
photodiode output voltages were the same for all
individuals (1.5V). Maximum photodiode output
voltages ranges from 3 to 3.8V for the individuals.
Fig 4.1:survey on diabetic patients
Fig 4.2: survey on anemic patients
In conclusion, this project has suggested a means
for non-invasive blood glucose and haemoglobin
levels testing. Although not as accurate as present day invasive or minimally invasive techniques for
measuring blood glucose- Hb concentrations, but
the use of near-infrared light provides a means of
non-invasive measurement with less pain and
discomfort to the diabetic patients
References
1. “Design of simple non-invasive glucose measuring
device”, published in International conference on
computing electrical and electronics engineering
,2013,DOI:10.1109/ICEEE.2013.663395
2. “Near infrared LED based noninvasive blood glucose
sensor”,published in International conference in Signal
Processing & Integrated
Networks(SPIN),2014,DOI:10.1109/SPIN.2014.6777023
3. “Non invasive Sensor Technique for total haemoglobin
measurement in blood”,Journal of Industrial and
intelligent Information vol.1,no.4 dec 2013.
4. Elina A. Genina , (2012) ‘Sensing Glucose and
Other Metabolites in Skin’ Handbook of Biophotonics,
Vol.2, Chapter 55.
5. Murugesh Shivashankar, Dhandayuthapanimani(2011), ‘
A Brief Overview of Diabetes’International Journal of
Pharmacy and Pharmaceutical Sciences, Vol 3, Suppl 4.
6. A.A. Shinde et al., " Non-invasive blood glucose
measurement using NIR technique based on occlusion
spectroscopy ," International Journal of Engineering
Science and Technology (IJEST), vol. 3, No.12
December 2011.
7. O Amir, D Weinstein, M.D. Silviu Zilberman, M. Less,
D. Perl-Treves, H. Primack, A. Weinstein, E. Gabis, B.
Fikhte, and A.Karasik, “Continuous noninvasive glucose
monitoring technology based on occlusion
spectroscopy," Journal of Diabetes Science
andTechnology, vol.1, no.4, pp. 463–469,July 2007.
8. U. Timm, D. McGrath, E. Lewis, J. Kraitl and H.Ewald,
“Sensor system for non-invasive optical heamoglobin
determination,” IEEE Sensors, October, 2009.
9. G. Zonios, J. Bykowski, and N. Kollias, “Skin melanin,
hemoglobin and light scattering properties can be
quantitatively assessed INVIVO using diffuse reflection
spectroscopy,” Journal of Investigative Dermatology,
vol. 117, pp. 1452-1457, 2001.
10. A. Yao, MD and H. Liu, MD “Continuous Noninvasive
Hemoglobin Measurement,” SCA Bulletin, October
2009, vol. 8, no. 5.
11. Rajashree Doshi, Anagha Panditrao(2013),’Noninvasive
optical sensor for haemoglobin determination’
International Journal of Engineering and Research and
Applications,vol.3, Issue 2.
12. O.S. Khalil, “Spectroscopic and clinical aspects of non-
invasive glucose measurements”, Clinical Chemistry, vol.
45, no.2, pp.165-177, 1999.
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
102
PHYSIOTHERAPY FOR REHABILITATION USING MOTION SENSOR Sheeba Abraham 1
1,Vamsi Krishna 2
2
1, 2 Sathyabama University
ABSTRACT
Physiotherapy plays a vital role in motor rehabilitation. People affected with muscular diseases require lifelong exercises to maintain their muscle tone. This requires a lot of training on patients’ part and can be done under the supervision of physiotherapist only. So Motion-based and virtual reality game systems have been used for rehabilitation earlier but require sensors to be attached to the patient and this hinders movement. This is a huge drawback in cases where the patient has limited limb movement and as a result of the wires and sensors attached to his body, feels bound. In order to bring the patients out of the confines of hospital by allowing them to do exercises at home, we have developed a software based application that uses Kinect -the motion sensor for carrying out routine rehabilitation exercise which is easy to use and engage its user so that he can monitor his progress and thus be motivated towards recovery. It also provides feedback to the patient on his performance on the run as well as at the successful completion of the exercise routine. This allows the physiotherapists to efficiently track the performance
of the patient over time which in turn helps him to take better clinical decision.
Keywords: Motor Rehabilitation, Motion based systems, Kinect Sensor
1. INTRODUCTION
People affected by muscular diseases like
cerebral palsy, muscular dystrophy, hemiparesis
should do exercise on a regular basis in order to
maintain a healthy muscle tone. Children affected by such diseases are attended by physiotherapists
regularly to track the condition of these children.
After meeting a physiotherapist we found
that both the children and adults lacks interest in
doing the exercises regularly which leads to further
complications. So we thought of complementing
these exercises to motivate the children to do
exercises with the help of a motion sensor called
Kinect.
1.1 DISEASES AND THEIR
TREATMENTS:
We have given the detailed information
about the diseases on which we have concentrated,
their definition, causes, diagnosis and symptoms and
treatments involved:
1.1.1 Cerebral Palsy:
Cerebral palsy (CP)[7] is a disorder that affects
muscle tone, movement, and motor skills (the ability
to move in a coordinated and purposeful way). CP is
usually caused by brain damage[9] that occurs before
or during a child's birth, or during the first 3 to 5 years of a child's life.
The three types of CP are: Spastic cerebral palsy — causes stiffness and
movement difficulties
Athetoid cerebral palsy — leads to involuntary
and uncontrolled movements
Ataxic cerebral palsy — causes a disturbed sense of balance and depth perception
1.1.1.1 Causes and treatment of Cerebral Palsy
The exact causes of most cases of CP are unknown,
but many are the result of problems during pregnancy
in which the brain is either damaged or doesn't
develop normally.Problems during labour and
delivery can cause CP[8] in some cases but this is the
exception Orthopaedic surgery can help repair
dislocated hips and scoliosis (curvature of the spine),
which are common problems associated with CP.
1.1.2 Hemiparesis:
Hemiparesis is a condition that is commonly
caused by either stroke or cerebral palsy, although
it can also be caused by multiple sclerosis, brain
tumours, and other diseases of the nervous system or
brain.
1.1.2.1 Causes and treatment of Hemiparesis
Brain damage caused by head injuries, cancerous
growths in a person's brain, or disease may also lead
to the development of muscle weakness. Damage to the person's brain can lead to muscle weakness as
well. Stroke; however, is the most common reason
people develop hemiparesis. Physiatrists, Physical
Therapists, Occupational Therapists, Electrical
Stimulation, Cortical Physiatrists, Physical
Therapists, Occupational Therapists, Electrical
Stimulation, Cortical Stimulation, Botox/Baclofen,
Motor Imagery, Modified Constraint-induced
Therapy, Medical science has created, or is looking
into, some promising new treatments for people with
hemiparesis that can also help people who have
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
103
experienced a stroke to improve movement in their
legs and arms after the initial stroke.
1.1.2.2 Forms of Hemiparesis
• Right-sided Hemiparesis: Involves injury to
the left side of the person's brain. The left side of a person's brain controls speaking
and language.
• Left-sided Hemiparesis: Involves injury to
the right side of the person's brain, which
controls learning processes, certain types of
behaviour, and non-verbal communication.
• Ataxia: Injury to the lower portion of a
person's brain may affect their body's ability
to coordinate movement.
• Pure Motor Hemiparesis: Pure motor
hemiparesis is the most common type of
hemiparesis. • Ataxic Hemiparesis Syndrome: Ataxic
hemiparesis syndrome involves weakness or
clumsiness on one side of a person's body.
2. METHODOLOGY
2.1 KINECT- THE MOTION SENSING
DEVICE
We have used Kinect as a tool to implement
our ideas about complementing rehabilitation. Kinect
is a motion sensor which tracks the full skeleton and
provides feedback about the movement of the person
with accuracy.
2.2 THE MOTION SENSOR- KINECT:
We used Kinect as the tool to achieve our
goal and developed programs that are compatible
with the Kinect and provided the detailed
information about the working of Kinect.
Fig 2.1 Kinect and its parts
Fig 2.2. Kinect Tracks Wholebody Skeleton
2.2.1 Introduction:
The device provides a natural user interface (NUI) [1] that allows users to interact intuitively and
without any intermediary device, such as a controller.
The Kinect system [2] identifies individual players
through face recognition and voice recognition. A
depth camera, which “sees” in 3-D, creates a skeleton
image of a player and a motion sensor detects their
movements
2.2.2 Construction of Kinect :
The Kinect sensor is a horizontal bar
connected to a small base with a motorized pivot and
is designed to be positioned lengthwise above or below the video display. The device features an
"RGB camera, depth sensor and multi-array
microphone running proprietary software", which
provide full-body 3D motion capture, facial
recognition and voice recognition capabilities.
The depth sensor consists of an infrared
laser projector [2] combined with a monochrome
CMOS sensor, which captures video data in 3D under
any ambient light conditions.
2.2.3 Working of Kinect:
The Kinect uses the structured light and machine learning[1] to infer the body position.
Inferring body position is a two stage process:
• It computes the depth map using structured
light.
• It uses machine learning for inferring the
body position.
Stage 1:
• The depth map is constructed by analysing a
speckle pattern of infrared laser light
• The Kinect uses an infrared projector and
sensor; it does not use its RGB camera for depth computation
The technique of analysing a known pattern
is called structured light.
• Structured light general principle: project a
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
104
known pattern onto the scene and infer
depth from the deformation of that pattern
on to the scene and inferdepth from focus
uses the principle that stuff that is more
blurry is further away.
• The Kinect dramatically improves the accuracy of traditional depth from focus
• The Kinect uses a special (“astigmatic”) lens
[2]with different focal length in x- and y-
directions
• A projected circle then becomes an ellipse
whose orientation depends on depth
• Depth from stereo uses parallax
• Kinect analyses the shift of the speckle
pattern by projecting from one location and
observing from another.
Stage 2:
Stage 2 has 2 sub stages (use intermediate “body parts” representation)
Stage 2.1: Start with 100,000 depth images with known
skeletons (from a motion capture system)
For each real image, render dozens more
using computer graphics techniques
Use computer graphics to render all
sequences for 15 different body types, and
vary several other parameters
Thus obtain over a million training
examples.
Stage 2.2
Transforms the body part image into a
skeleton The mean shift algorithm is used to robustly
compute modes of probability distributions
Mean shift is simple, fast, and effective.
2.3 THE SOFTWARE LANGUAGES AND
ALGORITHMS USED:
The languages used for programming are C ,
XML. The software packages used for hardware
aspects are Unity 3d and Blender [3].
2.3.1 Unity 3d:
Floor- 3D area on which skeleton moves
Image viewer- shows image of patient
Main camera- window that displays the skeleton and image viewer
User Interface
Zig skeleton- represents head, neck, torso,
waist, collar, shoulder, elbow, wrist, hand,
fingertip, hip, knee, ankle, foot.
2.3.2 User interface and algorithms for camera views
and skeletal pose:
The camera is shown as a label under which
buttons for different views of the camera are given.
Right side camera,
Front camera
Top camera.
2.3.3 Algorithms for camera views and skeletal pose:
Right side camera:
Enter camera option
If right side camera is selected, main camera
position is transformed as (10,0,0) and Euler’s angle as (0,270,0)
The image viewer position is transformed as
(-10,5,10) and Euler angles as (0,90,0) and
image viewer scale as (10,10,0.1f)
Front camera:
Enter camera option
If front camera is selected the main camera’s
position is transformed as (0,0,10) and Euler
angles as (0,180,0)
The image viewer position is transformed as (-10,5,-10) and Euler angles as (0, 0,0) and
image viewer scale as (10,10,0.1f).
Top camera:
• Enter the camera option
• top camera is selected the main camera’s
position is transformed as (0,10,0) and Euler
angles as (90,0,0)
• The image viewer position is transformed as
(5,0,2.5f) and Euler angles as (270,0,180)
and image viewer scale as (5,5,0.1f).
Skeletal pose algorithm:
• Enter the skeletal pose
• If mirroring off button is clicked zig
skeleton mirror is false and the skeleton
does not mirror the image of the person
• Else if mirroring on button is clicked zig
skeleton mirror is true and the skeleton
mirrors the image of the person.
Algorithm for setting elevation in Kinect:
• Enter the angle to which Kinect must be adjusted
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
105
• Select the ‘Set Elevation’ button
• The Kinect adjusts to specified angle
User interface, algorithm for therapy and type of
exercise selection:
Cerebral palsy:
Algorithm:
• Select the cerebral palsy button under the
label therapy
• Select the required exercise
1.shoulder abduction
2.horizontal adduction
3.shoulder flexion and extension
4.elbow flexion and extension.
Algorithms for exercises: Cerebral Palsy:
a b
c d
a)Shoulder abduction b)Horizontal adduction
c)Shoulder flexion and extension d)Elbow flexion and extension
Fig. 2.3 Exercise for cerebral Palsy
3. RESULTS AND DISCUSSION
We have developed the programs and were
able to execute them with kinect. We got the
output and attached the output screen shots for all
8 exercises.
3.1 Shoulder Abduction:
• Stand 6 feet away from the sensor • Raise the arm to 90 degrees and bring back
to rest position (0 degrees)
• The counter is triggered and increases by 1
• The exercise is repeated till counter reaches
10 after which it resets to 0.
Fig 3.1 Output for Shoulder Abduction
3.2 Horizontal Adduction:
• Stand 6 feet away from the sensor
• Raise the arm to 90 degrees and bring back
to rest position (0 degrees)
• The counter is triggered and increases by 1 • The exercise is repeated till counter reaches
10 after which it resets to 0.
Fig 3.2 Output for Horizontal Adduction
3.3 Shoulder Flexion And Extension
• Stand 6 feet away from the sensor
• Raise the hand to 90 degrees and bring back to rest position (0 degrees)
• The counter is triggered and increases by 1
• The exercise is repeated till counter reaches
10 after which it resets to 0.
n
Fig 3.2 Output for Shoulder Flexion and extension
3.5 Elbow Flexion and Extension:
• Stand 6 feet away from the sensor
• Flex the hand to 90 degrees and bring back
to rest position (0 degrees) • The counter is triggered and increases by 1
• The exercise is repeated till counter reach10
after which it resets to 0.
•
Fig 3.4 Output for Elbow Flexion and Extension
on
4. SUMMARY
The main objective of our paper is to bring a
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
106
change in the exercise routine of the physically
challenged children and develop software that can
present the usual exercises in an unusual form which
motivates and helps the children towards better
recovery. The people affected with muscular diseases
need therapeutic attention regularly for avoiding worsening of disease.
The main tool that we use to implement the
software is Kinect. The Kinect is a motion sensing
device that tracks the full body skeleton and moves
according to the user. The patient gets a feedback
about their performance in points after completing
each exercise. And every exercise has repetitions too.
As the patient completes one level they are
challenged with progressive difficult levels to test
their capability which helps them remain active. If
the therapist is not satisfactory with the results the patients can repeat their exercise.
The software languages that we used to
program are: c# and the game engines which we used
to design the character for Kinect and implement the
exercises are unity 3d and blender. The Zigfu pluggin
is used to integrate the entire exercises into Kinect.
5. CONCLUSION The exercises for children affected with
cerebral palsy and were collected after interviewing
the therapists and observing the affected children
directly. We have used c# for developing the exercise
programs and integrated it with unity –game engine
using zigfu pluggin.
The software is user friendly and provides
points and counts after completion of each exercise
and we have included light background music throughout exercise and a cheering sound while
displaying points to motivate the user to do the
exercises better.. This can be used for assisting
therapists and can be implemented in hospitals,
homes, rehabilitation centres and schools were many
patients need therapeutic attention throughout their
life. The efforts were really worth and the results
were highly satisfactory and beneficial for both
patients and the therapists as well.
6. REFERENCES
[1]. Jarret webb, James ashly, “Beginning Kinect Programming
with the Microsoft Kinect SDK”,Apress, pages 310,2012.
[2]. Sean kean, Jonathan hall, Phoenix perry, “Meet the Kinect:
An Introduction wto Programming Natural User Interfaces
(Technology in Action)”, publisher: wPaul manning, editor:
Jonathan Gennick. Pages 210.2011 20112
[3]. Hamilton, Naomi "The A-Z of Programming Languages:
C#", Computerworld,October 1, 2008.
[4]. O'Sullivan, S."Ch. 12: Stroke". In O'Sullivan, S.; Schmitz, T.
Physical Rehabilitation (5th ed) Philadelphia: F.A. Davis. pp.
705–769, (2007).
[5]. Patricia A. Downie,”Cash's Textbook of neurology for
physiotherapists”, wPublisher:Lippincott, Editor:Patricia A.
Downie, 653 pages,5 Aug 2008.
[6]. Lagerqvist, J. & Skargren, E. "Pusher syndrome: reliability,
validity, and sensitivity to change of a classification
instrument".Advances in Physiotherapy 8154–160, 2006.
[7]. Joan ross,”I Can't Walk but I Can Crawl: A Long Life with
Cerebral Palsy “(Lucky Duck Books), Sage Publications
(CA) 232 pages, October 18th 2005.
[8]. Sophie Levitt, “Treatment of Cerebral alsy and Motor
Delay”, Published by wWiley-Blackwell Paperback, 360
pages,2003.
[9]. Elaine Geralis ,”Children with Cerebral Palsy: A Parents'
Guide”, Woodbine House,Paperback, 481 pages. (first
published December 1991), January 1st 1998.
[10]. Grace E. Woods, “Infantile Cerebral Palsy”, Clinical Press
Ltd, Hardcover, pages 120. February 2nd
1995.
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
107
PALM PRINT BASED SECURITY SYSTEM FOR LOCKERS
Angelena Mathias 1 1, S. Caroline 2
2
St. Xavier’s Catholic College of Engineering, Nagercoil, India
ABSTRACT
According to ancient Greek scripts Biometrics means study of life. Biometric studies commonly include palm print, face, iris, voice, signature, hand geometry recognition and verification. Among the available biometric traits Palm Print proves to be one of the best traits providing good mismatch ratio and also reliable. The present scenario to operate a bank locker is with locks which are having keys. By this we can’t say that we are going to provide good security to our lockers. To provide perfect security and work easier, two different technologies Embedded Systems and Biometrics are used. Data acquisition is done by a scanner and images are stored in database. It has in-built ROM, DSP
and RAM where we can store up to 100 users palm prints. This module operates in 2 modes they are Master and User mode. In User mode, the scanned images are verified with the stored images. The images of the persons who are authorized to enter into the room will be stored in the module with a unique id. To prove that the persons are authorized to enter that area they need to scan their images. After the scanning has been completed the door automatically unlocks with the help of electric lock. If an unauthorized person tries to scan his image then the door doesn’t open for him.
Keywords— SVM, Euclidean distance, SIFT, Median Filter.
I. INTRODUCTION
Biometrics refers to the automatic identification
or identity verification of living persons using their
enduring physical or behavioral characteristics.
Many body parts, personal characteristics and
imaging methods have been suggested and used for
biometric systems such as fingers, hands, feet,
faces, eyes, ears, teeth, veins, voices, signatures,
typing styles, gaits and odors. As physical
characteristics are constant they do not change over
time and are also difficult to fake or change on
purpose. Physiological approaches include
fingerprints, iris, retinal scans, hand, finger, face,
ear geometry, hand vein, nail bed recognition, DNA
and palm prints. It is also a standardised model with
high accuracy. The cost of a palm print system is
lower than a finger print system. This is because in
palm print technique we use low resolution images
hence the cost of the capturing device and also the
memory requirement is less. It contains more
information and also hardly affected by age and
accessories. It is used in forensics, offices, and
departmental stores for security purpose.
Depending on where the biometrics is deployed,
the applications can be categorized in the following
five main groups: forensic, government,
commercial, health-care and traveling and
immigration. However, some applications are
common to these groups such as physical access,
PC/network access, time and attendance etc.
II. MATERIALS AND METHODS
The block diagram of the palm print security
system is shown in Fig. 1. The images are collected,
preprocessed and feature extracted and stored in the
database. The sample image is then preprocessed,
feature extracted and it is then matched with the
database image. If the sample matches the image in
database then the access is granted.
Fig. 1 Block diagram of Palm print security system
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
108
A. Palm print Image Acquisition
It is an action of retrieving an image from some
source, usually hardware based. Performing image
acquisition is always the first step in image
processing. The image that is acquired is completely
unprocessed and is the result of whatever hardware
was used to generate it.
Fig. 2 shows a flatbed scanner which is a
commonly used image capture device, the price of
the capture device is very low and the reliability and
facility of the device could also be ensured. The
flatbed scanner has its own lighting system and is
not affected by the surrounding light, which saved
design time for establishing a stable lighting system.
Fig. 2 Palm print Image Acquisition System
There are various kinds of flatbed scanners on
the market and we could easily choose a desired
flatbed scanner that is small and low in price, which
could capture a high-quality image from a user’s
palm. In general, we could design the capture
device with a good balance of cost and quality to
meet the above requirements.
B. Pre-Processing
Preprocessing is done to enhance the visual
appearance of the image. The datasets manipulation
is improved. The palm print image is formed of
principal lines, wrinkles, minutiae points, singular
points and texture. These palm print images, stored
in database, are not of quality, therefore these
images must undergo some pre-processing It
involves the following steps
Resizing
The input image may be of any size. It is
first resized to a size of (256 × 256 pixels). This is
to make all the palm prints of the same size and to
reach the result with fast response.
Median Filtering
Palm print image processing can be
implemented by using some filters and
enhancement in order to prepare it to the feature
extraction steps. Median filter is a nonlinear digital
filtering technique often used to remove noise. It is
used to remove speckle noise and salt and pepper
noise.
Histogram Equalization
It is a technique for adjusting image
intensities to enhance contrast. This method usually
increases the global contrast of many images,
especially when the usable data of the image is
represented by close contrast values.
Image Thresholding
It is simplest method of image segmentation.
It is a nonlinear operation that converts a gray scale
image into binary images. It is an effective way of
partitioning an image into a foreground and
background.
Detecting key points (SIFT)
SIFT key points of objects are first extracted
from a set of reference images and stored in a
database. An object is recognized in a new image by
individually comparing each feature from the new
image to this database and finding candidate
matching features based on Euclidean distance of
their feature vectors. From the full set of matches,
subsets of key points that agree on the object and its
location, scale and orientation in the new image are
identified to filter out good matches. The
determination of consistent clusters is performed
rapidly by using an efficient hash
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
109
table implementation of the generalized Hough
transform. Each cluster of 3 or more features that
agree on an object and its pose is then subject to
further detailed model verification and subsequently
outliers are discarded. Finally the probability that a
particular set of features indicates the presence of an
object is computed, given the accuracy of fit and
number of probable false matches. Object matches
that pass all these tests can be identified as correct
with high confidence.
C. Sharpening
The high pass filtered version of the Pre-
Processed image is added to the Pre-Processed
image in Image Sharpening. Sharpened image
highlights the edges and details of the Pre-processed
image.
D. Feature Extraction
Feature selection is one of the most
important tasks in data mining area, with methods
which allows determining the most relevant features
for pattern recognition. A suitable subset of features
is found when it permits synthesizing the similarity
of the pattern within its class and dissimilarity
amongst other different classes. The features which
give predominant difference between normal and
infected cells are identified and used for training
purpose. The selected features are geometrical,
color and statistical based. The mathematical
morphology provides an approach to the processing
of image based on shape. The statistical features use
gray level histogram and saturation histogram of the
pixels in the image and based on such analysis, the
mean value and variance are treated as the features
and are calculated. Texture is a property of natural
image like smoothness, coarseness and regularity of
a region. Though texture is an intuitive concept,
there is no commonly accepted definition for it.
Texture classification is one of the four
problem domains in the field of texture analysis
such as synthesis, classification, segmentation and
shape from texture. An image texture is a set of
metrics calculated in image processing designed to
quantify the perceived texture of an image. Image
Texture gives us information about the spatial
arrangement of color or intensities in an image or
selected region of an image. Image textures can be
artificially created or found in natural scenes
captured in an image. Image textures are one way
that can be used to help in classification of images.
To analyze an image texture in computer graphics,
there are two ways to approach the issue: Structured
Approach and Statistical Approach. Texture is
generated from the gray scale image matrices of the
red, green and blue components, as well as the
intensity component from the hue-saturation-
intensity image space. First order features, based on
the image histograms are used.
Average gray level or mean
The mean of the set of x values can be
denoted by , Mean of an image can be defined as
the sum of all pixel values to the total number of
pixels.
Where
- Sum of all x values
- Number of x values
Variance
The variance of a random variable X is its
second central moment, the expected value of the
squared deviation from the mean.
Where
– Mean of the image
SVM
SVM is one of the methods which the
statistical learning theory can be introduced to
practical application. It has its own advantages in
solving the pattern recognition problem with small
samples, nonlinearity, and higher dimension. SVM
can be easily introduced into learning problem such
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
110
as palm classification. As for the multiple
classification case, the basic idea is to transform the
problem into a 2-class problem, which is usually
implemented using a combined classifier. A poll is
carried out for all these 2 class classifiers and the
class with the most polls is determined as the class
that the point being analysed belongs to. When the
sample x is fed into the 2-class classifier
constructed from the class sample and the
class sample, the classifying function is then given
as
Where
s - Support vector
- Location of the hyper plane
-Vertical vector on the classifying hyper plane
Sign - is the signed function
- Kernel function
The central part of palm is consisting of three
ridges, flexion and secondary creases. The flexion
creases are otherwise called principal lines and the
secondary creases as wrinkles. These flexion and
the major secondary creases are formed between the
3rd and 5th months of pregnancy and superficial
lines appear after we are born. Even though flexion
is genetically dependent, most of other creases are
not. Even identical twins are having different palm
prints. These non-genetically deterministic and
complex patterns are very much useful in personal
identification.
E. Matching
The palm print features extracted are
compared with the sample images by the calculation
of Euclidean distance between the values. When
there is minimum Euclidean distance between the
features of the images, the palm prints are classified
as matching or mismatching with the database and
hence used for personal authentication purpose.
III. RESULTS
The simulation of the palm feature extraction and
matching is done using MATLAB. The
performance metrics such as mean, variance and
Euclidean distance is calculated. The Palm images
are collected and stored in the Database and their
corresponding values are also calculated. There are
five major steps carried out in this proposed system.
They are image acquisition, preprocessing,
Sharpening, Feature extraction and matching. Palm
image of any size is captured and preprocessing is
done. The results of each process are shown
separately and matching is done finally.
Fig. 3 Sample image matches the database image
Fig. 3, if the sample image matches the
image in the database then the access is granted and
the locker can be opened. Fig. 4 if the sample image
does not match the images in the database then the
system cannot be accessed and we get an Access
Denied message and the locker cannot be opened.
2nd International Conference on Bio Signals, Images and Instrumentation, (ICBSII - 2015).
111
Fig. 4.14 Sample image do not match the database
image
IV. DISCUSSION
The experimental result for image recognition
provides more accurate results and they are used for
the further development of a fully automated palm
print-based security system with high performance
in terms of effectiveness, accuracy, robustness, and
efficiency.
V. CONCLUSION
The proposed method gave an excellent
performance even if the palm print image is
contaminated with noise, contrast quality, size and
rotation. The feature extraction of palm prints based
on the textural analysis using MATLAB is an
efficient technique when compared to other
methods of palm print classification.
References
[1] A.C.M. Fong et al, (2007) “Palm print Verification for Controlling
Access to Shared Computing Resources”, IEEE Computer Society,
Vol. 6, No. 4 pp 1536-1268.
[2] Alfred C. Weaver, (2006) “Biometric Authentication”, IEEE Computer
Society, Vol. 39, No. 2 pp 96-97.
[3] Anil K. Jain et al, (2004) “An Introduction to Biometric Recognition”,
IEEE transactions on circuits and systems for video technology, Vol.
14, No. 1 pp 1051-8215.
[4] Anil K. Jain et al, (2010) “A Hybrid Approach for Generating Secure
and Discriminating Face Template”, IEEE transactions on information
forensics and security, Vol. 5, No. 1 pp 340-351.
[5] Anil K. Jain et al, (2013) “A Coarse to Fine Minutiae-Based Latent
Palm print Matching”, IEEE transactions on pattern analysis and
machine intelligence, Vol. 35, No. 10 pp 0162-8828.
[6] Anil K. Jain and Karthik Nandakumar, (2012) “Biometric
Authentication System Security and User Privacy”, IEEE computer
society, Vol.45, No. 11 pp 0018-9162.
[7] Bin Li et al, (2013) “Hashing Based Fast Palm print Identification for
Large-Scale Databases”, IEEE Transactions on systems information
forensics and security, Vol. 8, No. 5 pp 1556-6013.
[8] Bin-Yih Liao et al, (2014) “Interactive Artificial Bee Colony Supported
Passive Continuous Authentication System”, IEEE systems journal,
Vol. 8, No. 2 pp 1932-8184.
[9] Bob Zhang et al, (2013) “Palm-Print Classification by Global
Features”, IEEE transactions on systems, man and cybernetics systems,
Vol. 43, No. 2 pp 2168-2216.
[10] B. V. K. Vijaya Kumar and Vishnu Naresh Boddeti, (2013) “A
Framework for Binding and Retrieving Class-Specific Information to
and from Image Patterns Using Correlation Filters”, IEEE transactions
on pattern analysis and machine intelligence, Vol. 35, No. 9 pp 0162-
8828.
[11] Carlos M. Travieso et al, (2010) “Derivative Method for Hand Palm
Texture Biometric Verification”, IEEE Conference Publications
CICYT pp 1314-1333.
[12] Carlos M. Travieso et al, (2010) “Intra-Modal Biometric System Using
Hand-Geometry and Palmprint Texture”, IEEE Conference
Publications pp 978-1-4244-7402-8.
[13] Choonseog Park et al, (2009) “Fast and Accurate Retinal Vasculature
Tracing and Kernel Isomap based Feature Selection”, IEEE Conference
Publications pp 1474-1484.
[14] David Zhang, (2009) “Palm print Identification”, IEEE conference
publications, pp 22-28.
[15] David Zhang et al, (2006) “Palm Line Extraction and Matching for
Personal Authentication”, IEEE transactions on systems, man,
cybernetics, Vol. 36, No. 5 pp 1083-4427.
[16] Emad Fatemizadeh et al, (2011) “Multispectral Palm print Recognition
Using a Hybrid Feature”, IEEE Conference Publications pp 1112-5997.
[17] G. S. Lipane and S. B. Gundre, (2005) “Palm Print Recognition Review
Paper”, International Journal, Vol. 4, No 2 pp 2231-5381.
[18] Hiroshi Nakajima et al, (2006) “A Phase-Based Palm print Recognition
Algorithm and Its Experimental Evaluation”, IEEE Conference
Publications pp 0-7803-9733-9.
[19] Ivan Fratric, (2008) “Techniques and Recent Directions in Palm print
and Face Recognition”, IEEE Conference Publications pp 123-132.
[20] Jaishanker K. Pillai et al, (2011) “Secure and Robust Iris Recognition
Using Random Projections and Sparse Representations”, IEEE
transactions on pattern analysis and machine intelligence, Vol. 33, No.
9 pp 0162-8828.
[21] Jane You et al, (2004) “On Hierarchical Palm print Coding With
Multiple Features for Personal Identification in Large Databases”,
IEEE transactions on circuits and systems for video technology, Vol.
14, No.2 pp 1051-8215.
[22] Javier Galbally et al, (2014) “Image Quality Assessment for Fake
Biometric Detection: Application to Iris, Fingerprint, and Face
Recognition”, IEEE transactions on image processing, Vol. 23, No. 2
pp 1057-7149.
[23] Kasturika B. Ray and Rachita Misra, (2012) “Textural Features of Palm
Print”, International Journal of Computer Science and
Telecommunications, Volume 3, Issue 4 pp 78-81.
[24] K P Tripathi, (2011) “A Comparative Study of Biometric Technologies
with Reference to Human Interface”, International journal of computer
applications, Vol. 14, No. 5 pp 0975-8887.
[25] Shriram D. Raut and Vikas T. Humbe, (2012) “Biometric Palm Prints
Feature Matching for Person Identification”, IEEE conference
publications pp 61-69.
Proceedings of the
International Conference on Bio Signals, Images and Instrumentation
ICBSII 2015
Biomedical Engineering is a field of study that integrates two dynamic professions, Medicine
and Engineering. It has recently established itself as an independent field with the objective
of assisting medicine towards the betterment of society, through research.
Being an interdisciplinary science, it has associations with various other subjects such as
Electrical Engineering, Mechanical Engineering, Chemical Engineering and Biotechnology.
The spectrum of Bio-medical research aims to unite these disciplines in synergy, leading to
new possibilities thus enabling the development of technology that could save lives.
The International Conference on Bio Signals, Images and Instrumentation (ICBSII) was
conceived with the thought of bringing together scientists, engineers and researchers from
various domains all over the world.it has been a platform where some of the greatest minds of
the country and abroad could interact, exchange ideas and work together towards a common
goal.
Research papers were received from diverse areas such as Physiological Modelling, Medical
Imaging, Medical Robotics, Biomechanics, Biomedical Instrumentation and Nano-materials
amounting to a total of 152 papers. After a rigorous review process by an expert review
committee, 40 papers that displayed quality in idea and work were selected for final
presentation at the conference.
This conference is the fruit of a vision of the Management, faculty and students of the
Department of Biomedical Engineering, SSN College of Engineering who worked
unanimously towards materializing it and were instrumental in its success. Dr. A. Kavitha,
Associate Professor and Head of Bio-Medical Engineering has over 17 years of teaching
experience. She has 13 years of research experience in the field of Biomedical
Instrumentation, Medical Image Processing and Respiratory mechanics. Her current research
interest include Machine learning techniques, Bio-Medical Signal and Image Processing,
Statistical learning theory and Mathematical modelling in cognitive neuroscience.
The department of Biomedical Engineering, since its inception in 2005, has been a pioneer in
the field of biomedical technology, instrumentation, and administration. The department has
excellent infrastructure, experienced faculty members and motivated students. It also has
collaborations with several industries such as L&T Medical System, Texas Instruments,
National Instruments, Materialize, Chettinad Hospitals and Sri Ramachandra Medical
College.