tumor detection
TRANSCRIPT
AUTOMATIC TUMOR DETECTION ANDCLASSIFICATION OF BRAIN IMAGE
A PROJECT REPORT
Submitted by
KALIDAS.U 72306106019
KANIMOZHI.K 72306106020
KANIMOZHI.T 72306106021
RAJESH.V 72306106046
In partial fulfillment for the award of the degreeof
BACHELOR OF ENGINEERING
in
ELECTRONICS AND COMMUNICATION
ENGINEERING
of
ANNA UNIVERSITY, CHENNAI – 600 025
DEPARTMENT OF ELECTRONICS AND COMMUNICATION
ENGINEERING
VELALAR COLLEGE OF ENGINEERING AND TECHNOLOGY
ERODE-638 012.
APRIL 2010
VELALAR COLLEGE OF ENGINEERING ANDTECHNOLOGY, ERODE- 9.
DEPARTMENT OF ELECTRONICS ANDCOMMUNICATION ENGINEERING
1
Certificate
2
BONAFIDE CERTIFICATE
This is to certify that, the project report titled “AUTOMATIC TUMOR
DETECTION AND CLASSIFICATION OF BRAIN IMAGE” is the
bonafide work of
KALIDAS.U 72306106019
KANIMOZHI.K 72306106020
KANIMOZHI.T 72306106021
RAJESH.V 72306106046
Submitted in partial fulfillment of the requirements for the degree of
BACHELOR OF ENGINEERING during the year 2006-2010.
Dr.K.VENKATACHALAM, M.Tech., Ph.D., Mrs. J.NANDHINI B.E.,
HEAD OF THE DEPARTMENT SUPERVISOR & LECTURER
DEPARTMENT OF ECE DEPARTMENT OF ECE
Submitted for the university examination held on 08.04.2010 & 09.04.2010
INTERNAL EXAMINER EXTERNAL EXAMINER
3
Acknowledgement
4
ACKNOWLEDGEMENT
We are privileged to express our heartfelt thanks to our honorable
secretary Mr.S.D.CHANDRASEKAR B.A., who provided all the
facilities to build our project.
We hereby thank our former Principal and Administrative Director
Dr. P. SABAPATHI B.E. (Hons.), M.Sc., (Engg.), Ph.D., and our
Principal Dr. K.PALANISWAMY, M.E., Ph.D., who have been a great
inspiration not only for this project but also throughout this course of
study.
We express our profound gratitude to our beloved Head of the
Department Dr. K. VENKATACHALAM, M.Tech., (PhD) who
laconically brought us to the processor world.
We are highly indebted to our gregarious guide
Mrs.J.NANDHINI for her valuable guidance, advice and helps rendered
whenever we approached her in times of need.
We express our sincere thanks to our project coordinator
Dr.T.BALAKUMARAN,M.E., Phd., for their guidance to complete our
project successfully.
We are also highly thankful to all our indefatigable staff members
and non teaching staffs for helping us throughout the completion of the
project.
5
Abstract
6
AUTOMATIC TUMOR DETECTION AND CLASSIFICATION OF BRAIN IMAGE
ABSTRACT:
Segmentation of anatomical regions of the brain is the fundamental
problem in medical image analysis. In this paper, a brain tumor
segmentation method has been developed and validated segmentation on
2D MRI Data. This method can segment a tumor provided that the desired
parameters are set properly. This method does not require any
initialization while the others require an initialization inside the tumor. In
our segmentation approach watershed segmentation algorithm is used.
Watershed uses the intensity as a parameter to segment the whole image
data set. The input MRI image is preprocessed and loaded into matlab
workspace. In the segmentation process the image is divided into blocks
depending on the edge, gray and threshold parameter. The blocks are
divided by comparing the intensity value of the image with the parameters
as the intensity of the tumor affected area will be higher. Likewise the
tumor surface from the MRI image is segmented out. After the detection
of the tumor it is then classified using ICA algorithm which gives the type
of the tumor for the doctor’s convenience. Here the threshold limit is
applied to each image and the limit is tested on the ICA applied
algorithm. According to the intensity, tumor is classified into
ASTROCYTOMA, GLIOBLASTOMA, LYMPHOMA,
MENINGLOMA
7
i
8
Table of contents
9
TABLE OF CONTENT
CHAPTE
R
NO
TITLE
PAG
E NO
ABSTRACT i
LIST OF FIGURES v
LIST OF ABBREVATION vi
1. INTRODUCTION 1
2. LITERATURE REVIEW 4
2.1 IMAGING TECHNIQUES 4
2.1.1 Electron microscopy 4
2.1.2 Fluoroscopy 5
2.1.3 X Rays 5
2.1.3.1 Projection radiography 5
2.1.3.2 Computer tomography 6
2.1.3.3 Angiogram 8
2.1.4 Mammography 9
2.1.5 Magnetic Resonance Imaging 11
2.1.6 Ultrasonography 12
2.1.7 Thermography 13
2.1.8 Positron Emission Tomography 14
2.1.9 Photo Acoustic Imaging 14
2.1.10 Endoscopic Imaging 15
2.2 BRAIN TUMOR AND STAGES 15
2.2.1 Introduction 16
2.2.2 Stages of tumor 17
2.3 CAUSES OF BRAIN TUMOR 17
10
2.3.1 Race 17
2.3.2 Age 17
2.3.3 Family history 18
2.4 SYMPTOMPS OF BRAIN TUMOR 18
2.4.1 Head ache 18
2.4.2Seizures 18
2.4.3 Nausea and vomiting 19
2.4.4 Behavioural and cognitive Problems 19
2.5 TESTS AND DIAGNOSIS 19
2.5.1 A Neurological exam 19
2.5.2 Imaging test 19
2.5.3 Biopsy 20
2.6 TYPES OF TUMOR 20
2.6.1 Acoustic Neuroma 20
2.6.2 Astrocytom 21
2.6.2.1 Pilocytic Astrocytoma 21
2.6.2.2 Low-grade Astrocytoma 21
2.6.2.3 Anaplastic Astrocytoma 22
2.6.2.4 Anaplastic Astrocytoma 22
2.6.3 Glioblastoma multiframe 23
2.6.4 Chordoma 23
2.6.5 CNS Lymphoma 24
2.6.6 Craniopharyngioma 25
2.6.7 Brain stem Glioma 26
2.6.8 Meningioma 27
2.6.9 Schwannoma 29
2.6.10 Ependymoma 29
2.6.11 Rhabdoid tumor 31
11
ii
iii
3. SEGMENTATION ALGORITHMS 32
3.1 EDGE DETECTION 32
3.1.1 Sobel operator 32
3.1.2 Canny operator 35
3.1.3 Prewitt’s operator 36
3.1.4 Robertt’s cross operator 38
3.2 HISTOGRAM EQUALIZATION 40
3.3 THRESHOLDING TECHNIQUES 42
3.4 REGION BASD SEGMENTATION 44
3.5 FUZZY C-MEANS ALGORITHM 45
4. PROJECT DESCRIPTION 50
4.1 BLOCK DIAGRAM 50
4.2 WATERSHED SEGMENTATION 50
4.3 INDEPENDENT COMPONENT
ANALYSIS
55
4.3.1 Introduction 55
4.3.1.1 Linear noiseless ICA 56
4.3.2 Need for classification 57
4.3.3 Preprocessing steps in ICA 60
4.3.3.1 Centering 60
4.3.3.2 Whitening 60
4.4 COMPARISION OF PCA AND ICA 62
5. RESULT 63
6. CONCLUSION 65
APPENDIX 66
7. REFERENCE 71
12iv
List of figures
13
LIST OF FIGURES
FIGURE
NOTITLE PAGENO
2.1 Computer Tomography 7
3.1 Original Brain MR Image 34
3.2 Output of Edge Detection by Sobel Operator 34
3.3 Output of Edge Detection by Canny Operator 36
3.4 Output of Edge Detection by Prewitt Operator 38
3.5 Output of Edge Detection by Roberts Operator
40
3.6 Histogram 41
3.7 Output of Histogram Equalized Image 41
3.8 Output for Various Threshold Values 43
3.9 Output of Region Based Segmentation 45
3.10 Output of FCM Algorithm 49
4.1 Block Diagram 50
4.2 Segmentation using Watershed Algorithm 51
4.3 Original MR Image 53
4.4 Enhanced Image 53
4.5 Boundary Extraction of Reconstructed Image 54
4.6 Boundary Super Imposed on Original Image 54
4.7 Block Diagram of Spatial and Temporal ICA 58
4.8 Plot of ICA and PCA 62
14v
LIST OF ABBREVATIONS
15
LIST OF ABBREVIATIONS
MRI MAGNETIC RESONANCE IMAGING
CT COMPUTER TOMOGRAPHY
OPT ORTHOPANTOMOGRAPHY
NDE NONDESTRUCTIVE EVALUATION
DSA DIGITAL SUBTRACTION ANGIOGRAPHY
BSE BREAST SELF-EXAMINATION
PEM POSITRON EMISSION MAMMOGRAPHY
NMRI NUCLEAR MAGNETIC RESONANCE IMAGING
OCT OPTICAL COHERENCE TOMOGRAPHY
PNET PRIMITIVE NEUROECTODERMAL TUMOR
FCM FUZZY C-MEANS
SICA SPATIAL INDEPENDENT COMPONENT ANALYSIS
TICA TEMPORAL INDEPENDENT COMPONENT ANALYSIS
16vi
Chapter-1Introduction
17
CHAPTER 1
INTRODUCTION
The body is made up of many types of cells. Each
type of cell has special functions. Most cells in the body
grow and then divide in an orderly way to form new cells
as they are needed to keep the body healthy and working
properly. When cells lose the ability to control their
growth, they divide too often and without any order. The
extra cells form a mass of tissue called a tumor. Tumors
are benign or malignant. The aim of this work is to design
an automated tool for brain tumor quantification using
MRI image data sets. Magnetic Resonance Imaging (MRI) is the
state of the art medical imaging technology which allows cross sectional
view of the body with unprecedented tissue contrast. MRI plays an
important role in assessing pathological conditions of the ankle, foot and
brain. It has rapidly evolved into an accepted modality for medical
imaging of disease processes in the musculoskeletal system, especially
the foot and brain due to the use of non-ionizing radiation.
MRI provides a digital representation of tissue characteristic that
can be obtained in any tissue plane. The images produced by an MRI
scanner are best described as slices through the brain. MRI has the added
advantage of being able to produce images which slice through the brain
in both horizontal and vertical planes. This work is a small and
modest part of a quite complex system. The whole system
when completed visualizing the inside of the human body,
it makes surgeons able to perform operations inside a
18
patient without open surgery. More specifically the aim for
this work is to segment a tumor in a brain. This will make
the surgeon able to see the tumor and then ease the
treatment. The instruments needed for this could be
ultrasound, Computer Tomography (CT scan) and
Magnetic Resonance Imaging (MRI). In this Paper, the
technique used is Magnetic Resonance Imaging (MRI). The
segmentation of brain tumors in magnetic resonance images (MRI) is a
challenging and difficult task because of the variety of their possible
shapes, locations, image intensities.
Segmentation is an important process to extract information from
complex medical images. Segmentation has wide application in medical
field. The main objective of the image segmentation is to partition an
image into mutually exclusive and exhausted regions such that each
region of interest is spatially contiguous and the pixels within the region
are homogeneous with respect to a predefined criterion. Widely used
homogeneity criteria include values of intensity, texture, color, range,
surface normal and surface curvatures. Here Watershed
segmentation based algorithm has been used for
detection of tumor. Watershed segmentation uses the
intensity as a parameter to segment the whole image
data set. Moreover, the additional complexity of
estimation imposed to other algorithms causes a tendency
towards density dependent approaches. Among all
possible methods for this purpose, watershed can be used
as a powerful tool which implicitly extracts the tumor
surface. For detection of tumor and its classification in 2D
the software used is MATLAB.
19
After the segmentation of the detected tumor, the
classification is applied to the segmented surface. The
algorithm used here for the classification is ICA.
Independent component analysis (ICA) which has recently been
developed in the area of image processing. ICA is a variant of principal
component analysis (PCA) in which the components are assumed to be
mutually statistically independent instead of merely uncorrelated. The
stronger condition allows one to remove the rotational invariance of
PCA, i.e. ICA provides a meaningful unique bilinear decomposition of
two-way data that can be considered as a linear mixture of a number of
independent source signals. On applying the ICA algorithm to the
segmented tumor it is classified that, if the intensity found is between 248
and 256 , it is found to be ASTROCYTOMA and for the values between
224 and 228 it is found to be GLIOBLASTOMA. For the values between
238 and 240 it is found to be LYMPHOMA and for values between 263
and 290 it is found to be MENINGLOMA.
This report consists of six chapters. The second chapter provides a
brief insight about the medical imaging techniques commercially
available. The third chapter explains about the development of brain
tumor and its types. The fourth chapter gives a literature survey of
various segmentation algorithms available for brain MRI image. The fifth
chapter gives a brief description about this project and its corresponding
results and the sixth chapter leads to the conclusion.
20
Chapter-2Literature review
21
CHAPTER 2
MEDICAL IMAGING TECHNIQUES
2.1.1 ELECTRON MICROSCOPY
An Electron Microscope is a type of microscope that uses a
particle beam of electrons to illuminate a specimen and create a highly-
magnified image. Electron microscopes have much greater resolving
power than light microscopes that use electromagnetic radiation and can
obtain much higher magnifications of up to 2 million times, while the
best light microscopes are limited to magnifications of 2000 times. Both
electron and light microscopes have resolution limitations, imposed by
the wavelength of the radiation they use. The greater resolution and
magnification of the electron microscope is because the wavelength of an
electron; its de Broglie wavelength is much smaller than that of a photon
of visible light.
The electron microscope uses electrostatic and electromagnetic
lenses in forming the image by controlling the electron beam to focus it
22
at a specific plane relative to the specimen. This manner is similar to how
a light microscope uses glass lenses to focus light on or through a
specimen to form an image.
Types of Electron
a) Transmission Electron Microscope (TEM)
b) Scanning Electron Microscope (SEM)
c) Reflection Electron Microscope (REM)
d) Scanning Transmission Electron Microscope (STEM)
2.1.2 FLUOROSCOPY
Fluoroscopy is an imaging technique commonly used by
physicians to obtain real-time moving images of the internal structures of
a patient through the use of a fluoroscope. In its simplest form, a
fluoroscope consists of an x-ray source and fluorescent screen between
which a patient is placed. However, modern fluoroscopes couple the
screen to an x-ray image intensifier and CCD video camera allowing the
images to be recorded and played on a monitor.
The first fluoroscopes consisted of an x-ray source and fluorescent
screen between which the patient would be placed. As the x rays pass
through the patient, they are attenuated by varying amounts as they
interact with the different internal structures of the body, casting a
shadow of the structures on the fluorescent screen. Images on the screen
are produced as the untenanted X rays interact with atoms in the screen
through the photoelectric effect, giving their energy to the electrons.
While much of the energy given to the electrons is dissipated as heat, a
fraction of it is given off as visible light, producing the images. Early
23
radiologists would adapt their eyes to view the dim fluoroscopic images
by sitting in darkened rooms, or by wearing red adaptation goggles.
2.1.3 X- RAYS
2.1.3.1 PROJECTION RADIOGRAPHY
Radiographs, more commonly known as x-rays, are often used to
determine the type and extent of a fracture as well as for detecting
pathological changes in the lungs. With the use of radio-opaque contrast
media, such as barium, they can also be used to visualize the structure of
the stomach and intestines - this can help diagnose ulcers or certain types
of colon cancer.
2.1.3.2 COMPUTED TOMOGRAPHY
Tomography is the method of imaging a single plane, or slice, of
an object resulting in a tomogram. There are several forms of
tomography:
Linear tomography
Poly tomography
Zonography
Orthopantomography (OPT or OPG)
Computed Tomography (CT), or Computed Axial Tomography
A basic problem in imaging with x-rays (or other penetrating
radiation) is that a two-dimensional image is obtained of a three-
dimensional object. This means that structures can overlap in the final
image, even though they are completely separate in the object. This is
24
particularly troublesome in medical diagnosis where there are many
anatomic structures that can interfere with what the physician is trying to
see. During the 1930's, this problem was attacked by moving the x-ray
source and detector in a coordinated motion during image formation.
From the geometry of this motion, a single plane within the patient
remains in focus, while structures outside this plane become blurred. This
is analogous to a camera being focused on an object at 5 feet, while
objects at a distance of 1 and 50 feet are blurry. These related techniques
based on motion blurring are now collectively called classical
tomography. The word tomography means "a picture of a plane". In spite
of being well developed for more than 50 years, classical tomography is
rarely used. This is because it has a significant limitation: the interfering
objects are not removed from the image, only blurred. The resulting
image quality is usually too poor to be of practical use. The long sought
solution was a system that could create an image representing a 2D slice
through a 3D object with no interference from other structures in the 3D
object. This problem was solved in the early 1970s with the introduction
of a technique called computed tomography (CT). Computed
Tomography (CT) is a powerful nondestructive evaluation (NDE)
technique for producing 2-D and 3-D cross-sectional images of an object
from flat X-ray images. Figure 2.1 shown below is a schematic of a CT
system.
25
Figure 2.1 Computed Tomography
26
Characteristics of the internal structure of an object such as
dimensions, shape, internal defects, and density are readily available
from CT images. The test component is placed on a turntable stage that is
between a radiation source and an imaging system. The turntable and the
imaging system are connected to a computer so that x-ray images
collected can be correlated to the position of the test component. The
imaging system produces a 2-dimensional shadowgraph image of the
specimen just like a film radiograph.
2.1.3.3 ANGIOGRAPHY
Angiography or Arteriography is a medical imaging technique
used to visualize the inside, or lumen, of blood vessels and organs of the
body, with particular interest in the arteries, veins and the heart
chambers. This is traditionally done by injecting a radio-opaque contrast
agent into the blood vessel and imaging using X-ray based techniques
such as fluoroscopy. The word itself comes from the Greek words
angeion, "vessel", and graphein, "to write or record". The film or image
of the blood vessels is called an angiograph, or more commonly, an
angiogram.
Although the term angiography is strictly defined as based on
projectional radiography, the term has been applied to newer vascular
imaging techniques such as CT angiography and MR angiography.
Depending on the type of angiogram, access to the blood vessels is
gained most commonly through the femoral artery, to look at the left side
of the heart and the arterial system or the jugular or femoral vein, to look
at the right side of the heart and the venous system. Using a system of
guide wires and catheters, a type of contrast agent (which shows up by
27
absorbing the x-rays), is added to the blood to make it visible on the x-
ray images.
The X-ray images taken may either be still images, displayed on a
image intensifier or film, or motion images. For all structures except the
heart, the images are usually taken using a technique called digital
subtraction angiography (DSA). Images in this case are usually taken at 2
- 3 frames per second, which allows the radiologist to evaluate the flow
of the blood through a vessel or vessels. This technique "subtracts" the
bones and other organs so only the vessels filled with contrast agent can
be seen. The heart images are taken at 15-30 frames per second, not using
a subtraction technique. Because DSA requires the patient to remain
motionless, it cannot be used on the heart. Both these techniques enable
the radiologist or cardiologist to see stenosis (blockages or narrowings)
inside the vessel which may be inhibiting the flow of blood and causing
pain.
2.1.4 MAMMOGRAPHY
Mammography is the process of using low-dose amplitude-X-rays
(usually around 0.7 mSv) to examine the human breast and is used as a
diagnostic as well as a screening tool. The goal of mammography is the
early detection of breast cancer, typically through detection of
characteristic masses and/or microcalcifications. Mammography is
believed to reduce mortality from breast cancer. No other imaging
technique has been shown to reduce risk, but breast self-examination
(BSE) and physician examination are considered essential parts of
regular breast care.
In many countries routine mammography of older women is
encouraged as a screening method to diagnose early breast cancer. The
28
United States Preventive Services Task Force recommends screening
mammography, with or without clinical breast examination, every 1-2
years for women aged 40 and older. Altogether clinical trials have found
a relative reduction in breast cancer mortality of 20%, but the two
highest-quality trials found no reduction in mortality. Mammograms have
been controversial since 2000, when a paper highlighting the results of
the two highest-quality studies was published. Normally longer
wavelength X-rays are used for taking mammograms. Radiologists then
analyze the image for any abnormal findings.
At this time, mammography along with physical breast
examination is the modality of choice for screening for early breast
cancer. Ultrasound, ductography, positron emission mammography
(PEM), and magnetic resonance imaging are adjuncts to mammography.
Ultrasound is typically used for further evaluation of masses found on
mammography or palpable masses not seen on mammograms.
Ductograms are still used in some institutions for evaluation of bloody
nipple discharge when the mammogram is non-diagnostic. MRI can be
useful for further evaluation of questionable findings as well as for
screening pre-surgical evaluation in patients with known breast cancer to
detect any additional lesions that might change the surgical approach, for
instance from breast-conserving lumpectomy to mastectomy. New
procedures, not yet approved for use in the general public, including
breast tomosynthesis may offer benefits in years to come.
Mammography has a false-negative (missed cancer) rate of at least
10 percent. This is partly due to dense tissues obscuring the cancer and
the fact that the appearance of cancer on mammograms has a large
overlap with the appearance of normal tissues.
29
2.1.5 MAGNETIC RESONANCE IMAGING (MRI)
MRI or nuclear magnetic resonance imaging (NMRI), is primarily
a medical imaging technique most commonly used in radiology to
visualize the internal structure and function of the body. MRI provides
much greater contrast between the different soft tissues of the body than
computed tomography (CT) does, making it especially useful in
neurological (brain), musculoskeletal, cardiovascular, and oncological
(cancer) imaging. Unlike CT, it uses no ionizing radiation, but uses a
powerful magnetic field to align the nuclear magnetization of (usually)
hydrogen atoms in water in the body. Radio frequency (RF) fields are
used to systematically alter the alignment of this magnetization, causing
the hydrogen nuclei to produce a rotating magnetic field detectable by the
scanner. This signal can be manipulated by additional magnetic fields to
build up enough information to construct an image of the body.
How MRI works
The body is largely composed of water molecules which each
contain two hydrogen nuclei or protons. When a person goes inside the
powerful magnetic field of the scanner, these protons align with the
direction of the field.
A radio frequency electromagnetic field is then briefly turned on,
causing the protons to absorb some of its energy. When this field is
turned off the protons release this energy at a resonance radio frequency
which can be detected by the scanner. The frequency of the emitted
signal depends on the strength of the magnetic field. The position of
protons in the body can be determined by applying additional magnetic
fields during the scan which allows an image of the body to be built up.
30
These are created by turning gradients coils on and off which creates the
knocking sounds heard during an MR scan.
Diseased tissue, such as tumors, can be detected because the
protons in different tissues return to their equilibrium state at different
rates. By changing the parameters on the scanner this effect is used to
create contrast between different types of body tissue. MRI is used to
image every part of the body, and is particularly useful for neurological
conditions, for disorders of the muscles and joints, for evaluating tumors,
and for showing abnormalities in the heart and blood vessels.
2.1.6 ULTRASONOGRAPHY
Medical ultrasonography uses high frequency broadband sound
waves in the megahertz range that are reflected by tissue to varying
degrees to produce (up to 3D) images. This is commonly associated with
imaging the fetus in pregnant women. Uses of ultrasound are much
broader, however. Other important uses include imaging the abdominal
organs, heart, breast, muscles, tendons, arteries and veins. While it may
provide less anatomical detail than techniques such as CT or MRI, it has
several advantages which make it ideal in numerous situations, in
particular that it studies the function of moving structures in real-time,
emits no ionizing radiation, and contains speckle that can be used in
electrograph. It is very safe to use and does not appear to cause any
adverse effects, although information on this is not well documented. It is
also relatively inexpensive and quick to perform. The real time moving
image obtained can be used to guide drainage and biopsy procedures.
Doppler capabilities on modern scanners allow the blood flow in arteries
and veins to be assessed.
31
2.1.7 THERMOGRAPHY
Thermal imaging, thermographic imaging, or thermal video, is a
type of infrared imaging science. Thermographic cameras detect
radiation in the infrared range of the electromagnetic spectrum (roughly
900–14,000 nanometers or 0.9–14 µm) and produce images of that
radiation, called thermograms. Since infrared radiation is emitted by all
objects based on their temperatures, according to the black body radiation
law, thermography makes it possible to "see" one's environment with or
without visible illumination. The amount of radiation emitted by an
object increases with temperature, therefore thermography allows one to
see variations in temperature (hence the name). When viewed by
thermographic camera, warm objects stand out well against cooler
backgrounds; humans and other warm-blooded animals become easily
visible against the environment, day or night.
The use of thermal imaging has increased dramatically with
governments and airports staff using the technology to detect suspected
swine flu cases during the 2009 pandemic. Other uses include,
firefighters use it to see through smoke, find persons, and localize the
base of a fire. With thermal imaging, power lines maintenance
technicians locate overheating joints and parts, a tell-tale sign of their
failure, to eliminate potential hazards. Some physiological activities,
particularly responses, in human beings and other warm-blooded animals
can also be monitored with thermo graphic imaging.
2.1.8 POSITRON EMISSION TOMOGRAPHY
Positron emission tomography (PET) is a nuclear medicine
imaging technique which produces a three-dimensional image or picture
32
of functional processes in the body. The system detects pairs of gamma
rays emitted indirectly by a positron-emitting radionuclide (tracer), which
is introduced into the body on a biologically active molecule. Images of
tracer concentration in 3-dimensional space within the body are then
reconstructed by computer analysis. In modern scanners, this
reconstruction is often accomplished with the aid of a CT X-ray scan
performed on the patient during the same session, in the same machine.
If the biologically active molecule chosen for PET is FDG, an
analogue of glucose, the concentrations of tracer imaged then give tissue
metabolic activity, in terms of regional glucose uptake. Although use of
this tracer results in the most common type of PET scan, other tracer
molecules are used in PET to image the tissue concentration of many
other types of molecules of interest.
2.1.9 PHOTO ACOUSTIC IMAGING
Photo acoustic imaging is a recently developed hybrid biomedical
imaging modality based on the photo acoustic effect. It combines the
advantages of optical absorption contrast with ultrasonic spatial
resolution for deep imaging in (optical) diffusive or quasi-diffusive
regime. Recent studies have shown that photo acoustic imaging can be
used in vivo for tumor angiogenesis monitoring, blood oxygenation
mapping, functional brain imaging, and skin melanoma detection etc.
2.1.10 ENDOSCOPIC IMAGING
Endoscopy is a medical tool at the forefront in the diagnosis and
treatment of human diseases. Endoscopes, inserted through orifices such
as the mouth, nose, anus, and urethra, play an instrumental role in the
management of diseases of the pharynx, esophagus, stomach, small
33
intestine, colon, larynx, bronchial tree, and urinary system. The
fundamental concepts of endoscopy, including the optical and
mechanical components of typical endoscopes are given. The field of
endoscopy continues to evolve with the aid of technological innovations
and development. Early endoscopes consisted of simple rigid tubes that
provided limited views of a few easily accessed organs. Recent
developments have substantially enhanced the capabilities of endoscopes.
For example, fiber optic imaging bundles have allowed for the
development of flexible instruments that may be guided through tortuous
organs to visualize deeply into the body.
Conventional endoscopy is based on the detection of diffusely
reflected white light from tissue surfaces to reveal neoplasm.
Advancements in optical imaging take full advantage of light's properties
such as its spectral content and coherence to improve contrast and
resolution. Ultrasound imaging has been combined with endoscopy to
enable visualization beyond the tissue surface. Novel imaging modalities
such as fluorescence imaging and optical coherence tomography (OCT)
provide even more informative images. Technological improvements that
are enhancing the capability of endoscopes to visualize, diagnose, and
treat human diseases are revolutionizing the practice of medical
endoscopy.
BRAIN TUMOR AND ITS STAGES
2.2.1 INTRODUCTION
The brain tumor is an abnormal growth of cells within the brain. Brain
tumors can be
1. Benign (Non-cancerous)
34
2. Malignant(Cancerous)
Benign brain tumors do not contain cancer cells. Usually, benign
tumors can be removed, and they seldom grow back. The border or edge
of a benign brain tumor can be clearly seen. Cells from benign tumors do
not invade tissues around them or spread to other parts of the body.
However, benign tumors can press on sensitive areas of the brain and
cause serious health problems. Unlike benign tumors in most other parts
of the body, benign brain tumors are sometimes life threatening. Very
rarely, a benign brain tumor may become malignant.
Malignant brain tumors contain cancer cells. Malignant brain
tumors are generally more serious and often are lives threatening. They
are likely to grow rapidly and crowd or invade the surrounding healthy
brain tissue. Very rarely; cancer cells may break away from a malignant
brain tumor and spread to other parts of the brain, to the spinal cord, or
even to other parts of the body. The spread of cancer is called metastasis.
Sometimes, a malignant tumor does not extend into healthy tissue. The
tumor may be contained within a layer of tissue or the bones of the skull
or another structure in the head may confine it. This kind of tumor is
called encapsulated.
2.2.2 STAGES OF TUMOR
Different stages of tumors are given as follows:
Stage 0 - A typical cells in a normal anatomical configuration
Stage 1 - Tumor limited to the local anatomical site
Stage 2 - Involvement of ipsilateral regional lymph nodes
Stage 3 - Involvement of contra lateral lymph nodes
35
Stage 4 - Involvement of a distant site
The stage together with an assessment of the degree of
differentiation is very important for treatment planning and for
determining cancer prognosis.
2.3 CAUSES OF BRAIN TUMOR
2.3.1 RACE
Brain tumors occur more often among white people than among
people of other races.
2.3.2 AGE
Most brain tumors are detected in people who are 70 years old or
older. However, brain tumors are the second most common cancer in
children. (Leukemia is the most common childhood cancer.) Brain
tumors are more common in children younger than 8 years old than in
older children.
2.3.3 FAMILY HISTORY
People with family members who have gliomas may be more
likely to develop this disease. Being exposed to radiation or certain
chemicals at work:
Radiation - Workers in the nuclear industry have an increased risk
of developing a brain tumor.
Formaldehyde - Pathologists and embalmers who work with
formaldehyde have an increased risk of developing brain cancer.
Scientists have not found an increased risk of brain cancer among
other types of workers exposed to formaldehyde.
36
Vinyl chloride - Workers who make plastics may be exposed to
vinyl chloride. This chemical may increase the risk of brain
tumors.
Acrylonitrile - People who make textiles and plastics may be
exposed to acrylonitrile. This exposure may increase the risk of
brain cancer.
2.4 SYMPTOMS OF BRAIN TUMOR
2.4.1 HEADACHES:
This was the most common symptom, with 46% of the patients
reporting having headaches. They described the headaches in many
different ways, with no one pattern being a sure sign of brain tumor.
Many - perhaps most - people get headaches at some point in their life, so
this is not a definite sign of brain tumors.
2.4.2 SEIZURES:
This was the second most common symptom reported, with 33%
of the patients reporting a seizure before the diagnosis was made.
Seizures can also be caused by other things, like epilepsy, high fevers,
stroke, trauma, and other disorders. This is a symptom that should never
be ignored, whatever the cause. In a person who never had a seizure
before, it usually indicates something serious and you must get a brain
scan. A seizure is a sudden, involuntary change in behavior, muscle
control, consciousness, and/or sensation. Symptoms of a seizure can
range from sudden, violent shaking and total loss of consciousness to
muscle twitching or slight shaking of a limb. Staring into space, altered
vision, and difficulty in speaking are some of the other behaviors that a
37
person may exhibit while having a seizure. Approximately 10% of the
population will experience a single seizure in their lifetime.
2.4.3 NAUSEA AND VOMITING:
As with headaches, these are non-specific - which means that most
people who have nausea and vomiting do not have a brain tumor.
Twenty-two percent of the people in our survey reported that they had
nausea and /or vomiting as a symptom.
2.4.4 BEHAVIORAL AND COGNITIVE PROBLEMS:
Many reported behavioral and cognitive changes, such as:
problems with recent memory, inability to concentrate or finding the
right words, acting out - no patience or tolerance, and loss of inhibitions -
saying or doing things that are not appropriate for the situation.
2.5 TESTS AND DIAGNOSIS
2.5.1 A NEUROLOGICAL EXAM.
A neurological exam may include, among other things, checking
your vision, hearing, balance, coordination and reflexes. Difficulty in one
or more areas may provide clues about the part of your brain that could
be affected by a brain tumor.
2.5.2 IMAGING TESTS.
Magnetic resonance imaging (MRI) is commonly used to help
diagnose brain tumors. MRI uses magnetic fields and radio waves to
generate images of the brain. In some cases a dye may be injected
through a vein in your arm before your MRI. A number of specialized
MRI scans may help your doctor in evaluation and treatment planning,
38
including functional MRI, perfusion MRI and magnetic resonance
spectroscopy.
2.5.3 BIOPSY.
A biopsy can be performed as part of an operation to remove the
brain tumor, or a biopsy can be performed using a needle. A stereo tactic
needle biopsy may be done for brain tumors in hard to reach areas or very
sensitive areas within your brain that might be damaged by a more
extensive operation. A neurosurgeon drills a small hole, called a burr
hole, into the skull. A narrow, thin needle is then inserted through the
hole. Tissue is removed using the needle, which is frequently guided by
computerized tomography (CT) or MRI scanning. The biopsy sample is
then viewed under a microscope to determine if it is cancerous or benign.
This information is helpful in guiding treatment.
2.6 TYPES OF TUMOR
2.6.1 Acoustic Neuroma
An acoustic neuroma is also known as a vestibular schwannoma or
neurilemmoma.
Characteristics
Grows on the sheath surrounding the eighth cranial nerve in the
inner ear.
More common in women than men.
Symptoms
Hearing loss in one ear
Dizziness or vertigo
39
Tinnitus (ringing in the ear)
Tingling or numbness in the face
Walking and balance problems
Lack of coordination
2.6.2 Astrocytom
2.6.2.1 Pilocytic Astrocytoma
Also called: Juvenile Pilocytic Astrocytoma (JPA).
Characteristics
Slow growing, with relatively well-defined borders
Grows in the cerebrum, optic nerve pathways, brain stem and
cerebellum
Occurs most often in children and teens
Accounts for two percent of all brain tumors
2.6.2.2 Low-Grade Astrocytoma
An astrocytoma is a type of glioma that develops from star-shaped
cells (astrocytes) that support nerve cells. The WHO classifies a low-
grade astrocytoma as a grade II tumor.
Characteristics
Slow growing
Rarely spreads to other parts of the CNS
Borders not well defined
Common among men and women in their 20s-50s
40
2.6.2.3 Anaplastic Astrocytoma
An astrocytoma is a glioma that develops from star-shaped glial
cells (astrocytes) that support nerve cells. An anaplastic astrocytoma is
classified as a grade III tumor.
Characteristics
Grows faster and more aggressively than grade II astrocytomas
Tumor cells are not uniform in appearance
Invades neighboring tissue
Common among men and women in their 30s-50s
More common in men than women
Accounts for four percent of all brain tumors
2.6.2.4 Anaplastic Astrocytoma
An astrocytoma is a glioma that develops from star-shaped glial
cells (astrocytes) that support nerve cells. An anaplastic astrocytoma is
classified as a grade III tumor.
Characteristics
Grows faster and more aggressively than grade II astrocytomas
Tumor cells are not uniform in appearance
Invades neighboring tissue
Common among men and women in their 30s-50s
More common in men than women
Accounts for four percent of all brain tumors
41
2.6.3 Glioblastoma Multiforme (GBM)
An astrocytoma is a glioma that develops from star-shaped glial
cells (astrocytes) that support nerve cells. A glioblastoma multiforme is
classified as a grade IV astrocytoma. It is also referred to as a
glioblastoma or GBM.
Characteristics
Most invasive type of glial tumor
Commonly spreads to nearby tissue
Grows rapidly
May be composed of several different kinds of cells (i.e.,
astrocytes, oligodendrocytes)
May have evolved from a low-grade astrocytoma or an
oligodendroglioma
Common among men and women in their 50s-70s
More common in men than women
Accounts for 23 percent of all primary brain tumors
2.6.4 Chordoma
Characteristics
Rare and low grade
Occurs at the sacrum, near the lower tip of the spine, or at the base
of the skull
Originates from cells left over from early fetal development
42
Invades the bone and soft tissues but rarely the brain tissue
Can block the ventricles, causing hydrocephalus
Can metastasize (spread) or recur
Symptoms
Double vision
Headaches
2.6.5 CNS Lymphoma
CNS Lymphoma is a type of cancer that develops in the lymphatic
system. The lymphatic system is a network of small organs called lymph
nodes and vessels (similar to blood vessels) that carry a clear, watery
fluid called lymph throughout the body. This fluid supplies cells called
lymphocytes that fight disease and infection. To correctly diagnose
primary CNS Lymphoma, staging must be done. Staging is the process of
using CT scanning to examine many parts of the body. Staging helps to
confirm where the cancer originated and how far it has spread.
Characteristics
Very aggressive
Usually involves multiple tumors throughout the central nervous
system (CNS)
More common in people whose immune systems are compromised
Often develops in the brain, commonly in the areas adjacent to the
ventricles
Can be primary (originating in the brain) or secondary
43
Most common among men and women in their 60s-80s, but
incidence is increasing in young adults
Twice as common in men as in women
Accounts for three percent of all brain tumors
Symptoms
Headaches
Partial paralysis on one side of the body
Seizures
Cognitive or speech disorders
Vision problems
2.6.6 Craniopharyngioma
Characteristics
Most common in the parasellar region, an area at the base of the
brain and near the optic nerves
Also grows in the regions of the optic nerves and the
hypothalamus, near the pituitary gland
Tends to be low grade
Often accompanied by a cyst
Originates in cells left over from early fetal development
Occurs in children and men and women in their 50s and 60s
44
Symptoms
Headaches
Visual changes
Weight gain
Delayed development in children
2.6.7 Brain Stem Glioma
Characteristics
Named for its location at the base of the brain
Can range from low grade to high grade
Occurs most often in children between three and ten years of age,
but can occur in adults
Symptoms
Headaches
Nausea
Speech or balance abnormalities
Difficulty swallowing
Weakness or numbness of the arms and/or legs
Facial weakness
Double vision
45
Symptoms can develop slowly and subtly and may go unnoticed for
months. In other cases, the symptoms may arise abruptly. A sudden onset
of symptoms tends to occur with rapidly growing, high-grade tumors.
2.6.8 Meningioma
These tumors grow from the meninges, the layers of tissue
covering the brain and spinal cord. As they grow, meningiomas compress
adjacent brain tissue. Symptoms are often related to this compression of
brain tissue, which can also affect cranial nerves and blood vessels. In
some cases, meningioma growth can also extend into the bones of the
head and face, which may produce visible changes. Most meningiomas
are considered nonmalignant or low grade tumors. However, unlike
nonmalignant tumors elsewhere in the body, some of these brain tumors
can cause disability and may sometimes be life threatening. In many
cases, meningiomas grow slowly. Other meningiomas grow more rapidly
or have sudden growth spurts. There is no way to predict the rate of
growth of a meningioma or to know for certain how long a specific tumor
was growing before diagnosis. Meningiomas are graded from low to
high. The lower the grade, the lower the risk of recurrence and aggressive
growth.
The WHO classification divides meningiomas into three grades:
Grade I: Benign Meningioma
Grade II: Atypical Meningioma
Grade III: Malignant (Anaplastic) Meningioma
46
Characteristics
May arise after previous treatment from ionizing radiation or
excessive x-ray exposure
Common among women and men in their 40s-50s, but can occur at
any age
Twice as common in women as in men
Accounts for over 30 percent of all primary brain tumors
In very rare cases, can invade the skull or metastasize to the skin or
lungs
Women with meningiomas can experience tumor growth during
pregnancy
In rare cases, multiple meningiomas can develop at the same time
in different parts of the brain and/or spinal cord
Symptoms
Seizures
Headaches
Nausea and vomiting
Vision changes
Behavioral and cognitive changes
Sometimes no symptoms occur and tumor is detected incidentally
47
2.6.9 Schwannoma
Also known as vestibular schwannoma and acoustic neuroma (see
acoustic neuroma).
Characteristics
Arises from cells that form a protective sheath around nerve fibers
Typically grows around the eighth cranial nerve, but can be found
around other cranial or spinal nerves
Symptoms
Reduced hearing in the ear on the side of the tumor when eighth
cranial nerve is involved Tinnitus (ringing in the ear)
Balance problems
Deficits depend on the nerve that is affected
2.6.10 Ependymoma
Ependymal tumors begin in the ependyma, cells that line the
passageways in the brain where cerebrospinal fluid (CSF) is produced
and stored. Ependymomas are classified as either supratentorial (in the
cerebral hemispheres) or infratentorial (in the back of the brain).
Variations of this tumor type include subependymoma, subependymal
giant-cell astrocytoma, and malignant ependymoma. Ependymoblastoma,
which occurs in infants and children under three years, is no longer
considered a subtype of ependymoma. For ependymoblastoma, see
primitive neuroectodermal tumor (PNET) in the Non-glial Tumors
section.
48
Characteristics
Usually localized to one area of the brain
Develops from cells that line the hollow cavities at the bottom of
the brain and the canal containing the spinal cord
Can be slow growing or fast growing
May be located in the ventricles (cavities in the center of the brain)
May block the ventricles, causing hydrocephalus (water on the
brain)
Sometimes extends to the spinal cord
Common in children, and among men and women in their 40s and
50s
Occurrence peaks at age five and again at age 34
Accounts for two percent of all brain tumors
Symptoms
Severe headaches
Nausea and vomiting
Difficulty walking
Fatigue and sleepiness
Problems with coordination
Neck pain or stiffness
Visual problems
49
2.6.11 Rhabdoid Tumor
Characteristics
Rare
Highly aggressive and tends to spread throughout the CNS
Often appears in multiple sites in the body, especially the kidneys
Difficult to classify; may be confused with medulloblastoma or
PNETs
Occurs most often in young children but can also occur in adults
Symptoms
Vary depending on location of tumor in the brain or body
An orbital tumor may cause the eye to protrude
Balance problems may occur
50
Chapter-3Segmentation algorithms
51
CHAPTER 3
SEGMENTATION ALGORITHMS
Segmentation refers to the process of partitioning a digital image
into multiple segments (sets of pixels) (Also known as super pixels). The
goal of segmentation is to simplify and/or change the representation of an
image into something that is more meaningful and easier to analyze.
Image segmentation is typically used to locate objects and boundaries
(lines, curves, etc.) in images. More precisely, image segmentation is the
process of assigning a label to every pixel in an image such that pixels
with the same label share certain visual characteristics.
3.1 EDGE DETECTION
An edge is the boundary between two regions with relatively
distinct gray-level properties. Edge detection is a terminology in image
processing and computer vision, particularly in the areas of feature
detection and feature extraction, to refer to algorithms which aim at
identifying points in a digital image at which the image brightness
changes sharply or more formally has discontinuities.
3.1.1 SOBEL OPERATOR
The Sobel operator is used in image processing, particularly within
edge detection algorithms. Technically, it is a discrete differentiation
operator, computing an approximation of the gradient of the image
intensity function. At each point in the image, the result of the Sobel
operator is either the corresponding gradient vector or the norm of this
vector. The Sobel operator is based on convolving the image with a
small, separable, and integer valued filter in horizontal and vertical
direction and is therefore relatively inexpensive in terms of computations.
52
On the other hand, the gradient approximation which it produces is
relatively crude, in particular for high frequency variations in the image.
The operator consists of a pair of 3×3 convolution kernels as shown in
Figure. One kernel is simply the other rotated by 90°.
These kernels are designed to respond maximally to edges running
vertically and horizontally relative to the pixel grid, one kernel for each
of the two perpendicular orientations. The kernels can be applied
separately to the input image, to produce separate measurements of the
gradient component in each orientation (call these Gx and Gy). These can
then be combined together to find the absolute magnitude of the gradient
at each point and the orientation of that gradient. The gradient magnitude
is given by equation 3.1,
(3.1)
Typically, an approximate magnitude is computed using equation 3.2,
(3.2)
which is much faster to compute.
53
The angle of orientation of the edge (relative to the pixel grid) giving rise
to the spatial gradient is given by equation 3.3,
(3.3)
Figure 3.1 Original Brain MR Image
Figure 3.2 Output of Edge Detection by Sobel Operator
54
3.1.2 CANNY OPERATOR
Canny (1986) considered the mathematical problem of deriving an
optimal smoothing filter given the criteria of detection, localization and
minimizing multiple responses to a single edge. He showed that the
optimal filter given these assumptions is a sum of four exponential terms.
He also showed that this filter can be well approximated by first-order
derivatives of Gaussians. Canny also introduced the notion of non-
maximum suppression, which means that given the presmoothing filters,
edge points are defined as points where the gradient magnitude assumes a
local maximum in the gradient direction.
Although his work was done in the early days of computer vision,
the Canny edge detector (including its variations) is still a state-of-the-art
edge detector. Unless the preconditions are particularly suitable, it is hard
to find an edge detector that performs significantly better than the Canny
edge detector.
The Canny-Deriche detector (Deriche 1987) was derived from
similar mathematical criteria as the Canny edge detector, although
starting from a discrete viewpoint and then leading to a set of recursive
filters for image smoothing instead of exponential filters or Gaussian
filters.
55
Figure 3.3 Output of Edge Detection by Canny Operator
Fig 3.3 shows the edge detection output by applying the Canny operator.
Canny operator has detected not only the tumor region also detects the
unwanted artifacts.
3.1.3 PREWITT’S OPERATOR
Prewitt is a method of edge detection in image processing which
calculates the maximum response of a set of convolution kernels to find
the local edge orientation for each pixel.Prewitt operator is similar to the
Sobel operator and is used for detecting vertical and horizontal edges in
images.
56
Various kernels can be used for this operation. The whole set of 8
kernels is produced by taking one of the kernels and rotating its
coefficients circularly. Each of the resulting kernels is sensitive to an
edge orientation ranging from 0° to 315° in steps of 45°, where 0°
corresponds to a vertical edge.
The maximum response for each pixel is the value of the
corresponding pixel in the output magnitude image. The values for the
output orientation image lie between 1 and 8, depending on which of the
8 kernels produced the maximum response.
This edge detection method is also called edge template matching,
because a set of edge templates is matched to the image, each
representing an edge in a certain orientation. The edge magnitude and
orientation of a pixel is then determined by the template that matches the
local area of the pixel the best.
The Prewitt edge detector is an appropriate way to estimate the
magnitude and orientation of an edge. Although differential gradient edge
detection needs a rather time-consuming calculation to estimate the
orientation from the magnitudes in the x- and y-directions, the Prewitt
edge detection obtains the orientation directly from the kernel with the
maximum response. The set of kernels is limited to 8 possible
orientations; however experience shows that most direct orientation
estimates are not much more accurate.
On the other hand, the set of kernels needs 8 convolutions for each
pixel, whereas the set of kernel in gradient method needs only 2, one
kernel being sensitive to edges in the vertical direction and one to the
horizontal direction. The result for the edge magnitude image is very
similar with both methods, provided the same convolving kernel is used.
57
Figure 3.4 Output of Edge Detection by Prewitt Operator
Fig 3.4 shows the edge detection output by applying the Prewitt
operator. Like the Sobel operator, Prewitt operator detects only the
boundary of object.
3.1.4 ROBERT’S CROSS OPERATOR
The Roberts Cross operator performs a simple, quick to compute,
2-D spatial gradient measurement on an image. Pixel values at each point
in the output represent the estimated absolute magnitude of the spatial
gradient of the input image at that point.
The operator consists of a pair of 2×2 convolution kernels as
shown in Figure. One kernel is simply the other rotated by 90°. This is
very similar to the Sobel operator.
58
These kernels are designed to respond maximally to edges running
at 45° to the pixel grid, one kernel for each of the two perpendicular
orientations. The kernels can be applied separately to the input image, to
produce separate measurements of the gradient component in each
orientation (call these Gx and Gy). These can then be combined together
to find the absolute magnitude of the gradient at each point and the
orientation of that gradient. The gradient magnitude is given by equation
3.4,
(3.4)
Although typically, an approximate magnitude is computed using
equation 3.5,
(3.5)
which is much faster to compute.
The angle of orientation of the edge giving rise to the spatial gradient
(relative to the pixel grid orientation) is given by:
59
Figure 3.5 Output of Edge Detection by Roberts Operator
Fig 3.5 shows the edge detection output by applying the Robert
operator. From the above outputs, all operators have failed to detect the
tumor location.
3.2 HISTOGRAM EQUALIZATION
Histogram equalization is a method in image processing of contrast
adjustment using the image's histogram. This method usually increases
the global contrast of many images, especially when the usable data of
the image is represented by close contrast values. Through this
adjustment, the intensities can be better distributed on the histogram. This
allows for areas of lower local contrast to gain a higher contrast without
affecting the global contrast. Histogram equalization accomplishes this
by effectively spreading out the most frequent intensity values.
The method is useful in images with backgrounds and foregrounds
that are both bright or both dark. In particular, the method can lead to
better views of bone structure in x-ray images, and to better detail in
60
photographs that are over or under-exposed. A key advantage of the
method is that it is a fairly straightforward technique and an invertible
operator. So in theory, if the histogram equalization function is known,
then the original histogram can be recovered. The calculation is not
computationally intensive. A disadvantage of the method is that it is
indiscriminate. It may increase the contrast of background noise, while
decreasing the usable signal.
Figure 3.6 Histogram
Figure 3.7 Output of Histogram equalized image
61
The spatial domain enhancement technique, histogram equalization
improves contrast of the MR image by reassigning the brightness values
of pixels based on the image histogram. Generally, images have unique
brightness histograms. Even images of different areas of the same
sample, in which the various structures present have consistent brightness
levels wherever they occur, will have different histograms, depending on
the area fraction of each structure. Here the pixel intensities are modified
by a position invariant transformation function. The traditional histogram
equalization method for MR image suffers from the following
drawbacks:
It lacks of a mechanism to adjust the degree of enhancement.
It often causes unpleasant visual artifacts, such as over
enhancement, level saturation and raised noise level.
It could dramatically change the character of the image, e.g., the
average luminance (mean) of the image. Changing the overall
illumination of MR image will shifts the peaks in the histogram,
there is a very little scope to improve contrast by global
transformation.
3.3 THRESHOLDING TECHNIQUES
Thresholding is the simplest method of image segmentation. From
a grayscale image, thresholding can be used to create binary images.
During the thresholding process, individual pixels in an image are
marked as “object” pixels if their value is greater than some threshold
value (assuming an object to be brighter than the background) and as
“background” pixels otherwise. This convention is known as threshold
62
above. Variants include threshold below, which is opposite of threshold
above; threshold inside, where a pixel is labeled "object" if its value is
between two thresholds; and threshold outside, which is the opposite of
threshold inside (Shapiro, et al. 2001:83). Typically, an object pixel is
given a value of “1” while a background pixel is given a value of “0.”
Finally, a binary image is created by coloring each pixel white or black,
depending on a pixel's label.
Thresholding Between 100-200 Thresholding Between 175-200
Thresholding Between 200-225 Thresholding above 240
Figure 3.8 Output for various Threshold values
63
Fig 3.8 shows the output images by applying various threshold values.
The drawbacks of thresholding includes
• Threshold selection is not always straightforward.
• Pixels assigned to a single class need not form coherent regions as
the spatial locations of pixels are completely ignored.
3.4 REGION BASED SEGMENTATION
Region-based segmentation methods attempt to partition or group
regions according to common image properties. These image properties
consist of
1. Intensity values from original images, or computed values based
on an image operator
2. Textures or patterns that are unique to each type of region
3. Spectral profiles that provide multidimensional image data
These can be classified as two main classes
Merging Algorithms - in which neighboring regions are compared
and merged if they are close enough in some property.
Splitting Algorithms – in which large non-uniform regions are
broken up into small areas which may be uniform.
These algorithms which are combination of splitting and merging.
In all cases some uniformity criterion must be applied to decide if a
region should be split or two regions should be merged. This criterion is
based on some region property which will be decided by the application
64
and could be one of the measurable image attributes such as image mean
intensity, color, etc.,
Fig 3.9 Output of region based segmentation
Fig 3.9 shows the segmented image by applying the region based
algorithm. From the output tumor regions are segmented exactly but the
drawback of region based algorithm is it is difficult to identify the seed
points.
3.5 FUZZY C-MEANS ALGORITHM
Fuzzy C-Means Clustering (FCM) is also known as Fuzzy
ISODATA, for clustering technique. The aim of FCM is to find cluster
centers (centroids) that minimize a dissimilarity function. The fuzzified
c-means algorithm (Bezdek in Jang et al., 1997) allows each data point to
belong to a cluster to a degree specified by a membership grade, and thus
each point may belong to several clusters. The FCM employs fuzzy
partitioning such that a data point can belong to all groups with different
65
membership grades between 0 and 1. FCM is an iterative algorithm and it
is a method of grouping the similar types of pixels in the image.
Fuzzy c-means is different from hard c-means, mainly because it
employs fuzzy partitioning, where a point can belong to several clusters
with degrees of membership.
Clustering of numerical data forms the basis of many segmentation
and system modeling algorithms. The purpose of clustering is to identify
natural groupings of data from a large data set to produce a concise
representation of a system's behavior.
Fuzzy c-means (FCM) is a data clustering technique wherein each
data point belongs to a cluster to some degree that is specified by a
membership grade. This technique was originally introduced by Jim
Bezdek in 1981 [Bez81] as an improvement on earlier clustering
methods. It provides a method that shows how to group data points that
populate some multidimensional space into a specific number of different
clusters.
FCM starts with an initial guess for the cluster centers, which are
intended to mark the mean location of each cluster. The initial guess for
these cluster centers is most likely incorrect. Additionally, FCM assigns
every data point a membership grade for each cluster. By iteratively
updating the cluster centers and the membership grades for each data
point, FCM iteratively moves the cluster centers to the right location
within a data set. This iteration is based on minimizing an objective
function that represents the distance from any given data point to a
cluster center weighted by that data point's membership grade. By using
information returned by FCM to represent the fuzzy qualities of each
cluster.
66
A new cluster validity index is proposed that determines the
optimal partition and optimal number of clusters for fuzzy partitions
obtained from the fuzzy c-means algorithm. The proposed validity index
exploits an overlap measure and a separation measure between clusters.
The overlap measure, which indicates the degree of overlap between
fuzzy clusters, is obtained by computing an inter-cluster overlap. The
separation measure, which indicates the isolation distance between fuzzy
clusters, is obtained by computing a distance between fuzzy clusters.
A good fuzzy partition is expected to have a low degree of overlap
and a larger separation distance. Fuzzy cluster-validity criterion tends to
evaluate the quality of fuzzy c-partitions produced by fuzzy clustering
algorithms. Many functions have been proposed. Some methods use only
the properties of fuzzy membership degrees to evaluate partitions. Others
techniques combine the properties of membership degrees and the
structure of data. Major problems exist in both crisp and fuzzy clustering
algorithms. The fuzzy c-means type of algorithms use weights
determined by a power m of inverse distances that remains fixed over all
iterations and over all clusters, even though smaller clusters should have
a larger.
This method uses a different “distance” for each cluster that changes
over the early iterations to fit the clusters. Clustering refers to the process
of unsupervised partitioning of a data set based on a dissimilarity
measure, which determines the cluster shape. Considering that cluster
shapes may change from one cluster to another, it would be of the utmost
importance to extract the dissimilarity measure directly from the data by
means of a data model.
67
The fuzzy c-means (FCM) clustering algorithm has been
extensively used for pattern recognition. It has also been used in the
process of generating fuzzy rules from data. It has been used with success
in the soft segmentation of MR images and for the estimation of partial
volumes.
FCM partitions a collection of n vector Xi,i=1,2,3,……..,n. into
‘C’ fuzzy groups and finds the cluster center in each group such that a
cost function of dissimilarity measure is minimized. FCM employs fuzzy
partitioning such that a given data point can belong to several groups
with the degree of belongingness specified by membership grades
between 0 and 1.
The FCM algorithm is simply an iterative procedure. In a batch
mode operation FCM determines the cluster centers Ci and the
membership matrix U using following steps.
Step 1: Intialize the cluster centers the membership matrix U with
random values between 0 and 1 such that the following constraints are
satisfied
Step 2: Calculate ‘C’ fuzzy cluster centers Ci,i=1,2,……..,C
Step 3: Compute the cost functions
Where,
Y={yi },is the set of centers of clusters.
68
Ej(xk), is a dissimilarity measure between the sample xk and the center yj
of a specific cluster j.
U=[ujk], is the c x n fuzzy c-partition matrix, containing the membership
values of all samples in all clusters.
m (1, ), is a control parameter of fuzziness.
Stop if either Jm below a certain tolerance or it is improved over
previous iteration.
Step 4: Compute a new U and repeat the steps until an optimum result is
obtained.
The performance depends on the initial cluster centers, thereby allowing
to run FCM several times, each starting with a different set of initial
cluster centers.
Figure 3.10 Output of FCM Algorithm
69
Chapter-4Project description
70
CHAPTER 4
PROJECT DESCRIPTION
4.1 BLOCK DIAGRAM
Fig 4.1 BLOCK DIAGRAM
4.2 WATERSHED SEGMENTATION
The watershed algorithm is an image processing segmentation
algorithm that splits an image into areas, based on the topology of the
image. The length of the gradients is interpreted as elevation information.
During the successive flooding of the grey value relief, watersheds with
adjacent catchment’s basins are constructed. This flooding process is
performed on the gradient image, i.e. the basins should emerge along the
edges. Normally this will lead to an over-segmentation of the image,
especially for noisy image material, e.g. medical CT data. Either the
71
image must be pre-processed or the regions must be merged on the basis
of a similarity criterion afterwards.
A hierarchic watershed transformation converts the result into
a graph display (i.e. the neighbor relationships of the segmented regions
are determined) and applies further watershed transformations
recursively. A problem is that the watersheds will increase in width.
The marker based watershed transformation performs
flooding starting from specific marker positions which have been either
explicitly defined by the user or determined with morphological
operators. Interactive watershed transformations allow to determine
include and exclude points to construct artificial watersheds. This can
enhance the result of segmentation.
Fig 4.2 Segmentation using Watershed Algorithm
The image on the left represents the type of result obtained from the
thresholding of classical images where Watershed segmentation is
efficient. This could be a picture of coffee beans, blood cells, sand ...
Concepts of Watershed segmentation is
The concepts of watersheds and catchment basins are well known
in topography.
72
Watershed lines divide individual catchment basins.
The North American Continental Divide is a textbook example of
a watershed line with catchment basins formed by the Atlantic and
Pacific Oceans.
Working the 2D function presentations, image data may be
interpreted as a topographic surface where the image gray-levels
represent altitudes.
Thus, region edges correspond to high watersheds and low-
gradient region interiors correspond to catchment basins.
The goal of region growing segmentation is to create homogeneous
regions.
In watershed segmentation, catchment basins of the topographic
surface are homogeneous in the sense that all pixels belonging to
the same catchment basin are connected with the basin's region of
minimum altitude (gray-level) by a simple path of pixels that have
monotonically decreasing altitude (gray-level) along the path.
Such catchment basins then represent the regions of the segmented
image.
One of the important drawbacks of watershed segmentation
algorithm is producing severe oversegmentation due sensitivity of noise.
73
Fig 4.3 Original MRI image
Fig 4.4 Enhanced Image
74
Fig 4.5 Boundary extraction of reconstructed image using watershed
algorithm
Fig 4.6 Boundary superimposed on original image
75
4.3 INDEPENDENT COMPONENT ANALYSIS
4.3.1 INTRODUCTION
Independent component analysis (ICA) is a computational method
for separating a multivariate signal into additive subcomponents
supposing the mutual statistical independence of the non-Gaussian source
signals.
A simple application of ICA is the “cocktail party problem”,
where the underlying speech signals are separated from a sample data
consisting of people talking simultaneously in a room. The problem is
simplified by assuming no time delays and echoes. If N number of source
present, at least N observations are needed to get the original signals.
This constitutes the square
J = D
Where,
D = input dimension of the data
J = dimension of the model
There are two cases:
1. If (J < D) is underdetermined
2. If (J>D) is overdetermined
TYPES OF ICA
There are two types of ICA.They are
1 .Non Linear ICA
2. Linear ICA
(a) Linear Noiseless ICA
(b) Linear Noisy ICA
76
4.3.1.1 LINEAR NOISELESS ICA
The components xi is a random vector are generated as a sum of
the independent components sk, weighted by the mixing weights ai,k.The
generative formula is given by
x = As
X=Mixture; A=Mixing coefficients; S=Sources.
This is called ICA MODEL. This is done by adaptively calculating
the w vectors and setting up a cost function which either maximizes the
nongaussianity of the calculated by
sk = (wT * x)
ASSUMPTIONS
1. Linear mixing
2. Independence of sources
3. Non Gaussianity
(A) LINEAR MIXING
Linear mixing based on first and second order stastics are usually
optimal. When the linear transformation takes place it leads to
gaussianty.So limited amount of information can be separated into
independent components. But when this phenomenon takes place with
higher order stastics then it does not miss out extra information which
enhances the image quality.
77
(B) INDEPENDENCE OF SOURCE
If two random variables x and y are present in an image, they are
said to be independent if the information regarding x does not dependent
on y this is one of the key concept in independent component analysis.
(C) NON GAUSSIANITY
According to central limit theorem sum of non Gaussian variables
is closer to Gaussian original ones. Its non gaussianity will attain the
local maximum equal to independent components. This is because, if it
were the mixture of two are more components it would be closer to
Gaussian distribution but this is eliminated by central limit theorem. If
the contained data in an image is non Gaussian then their high order
statistics would contain extra information which makes the process
easier.
4.3.3 NEED FOR CLASSIFICATION
In magnetic resonance imaging (MRI), a set of slices are acquired
over time, and small differences in the intensity of the signal over time
are extracted. The first application of ICA to MRI data used spatial ICA
(SICA). SICA when applying ICA to MRI have several reasons:
The most important is that the spatial dimension is much larger
than the temporal dimension in MRI. By choosing a particular
component is examined through the spatial map using knowledge of
brain structure and function. A spatial ICA analysis is performed on the
data.
The application of SICA to MRI data is typically done in one of
two ways:
78
Consistently task-related components are then chosen by
correlating their time courses with the predicted waveform. Transiently
task-related components are also extracted by examination of those
components that are correlated, but not as highly correlated as the
consistently task-related component.
CLASSIFICATION OF ICA
1. Spatial Independent Component Analysis (SICA)
2. Temporal Independent Component Analysis (TICA)
Fig 4.7 BLOCK DIAGRAM OF SPATIAL AND TEMPORAL ICA
79
In spatial ICA, suppose X is an N-by-M matrix, where N is the
number of time points and M is the number of voxels. The “signals” are
the M spatial voxels , flattened to a 1-D vector, and there are thus N
different instances of these signals whereas TICA would consider the
signals the N individual time courses of which there are M instances.
The SICA decomposition can then be described as
C = W* X
W= N-by-N estimated linear mixing matrix
C = N-by-M matrix with N independent components.
Now,
X = Wˆ -1*C
Where the spatially independent components (images) are located in the
rows of C.
In temporal ICA, X is an M-by-N matrix. The decomposition is
C = W* X
W= M-by-M estimated linear mixing matrix
C =M-by-N matrix containing the M independent component
Now, we can write
X = Wˆ -1*C
Where the temporally independent time courses are located in the
rows of C and the associated temporally independent maps (images) are
found in the columns of Wˆ -1.
80
Spatial independent components suits for MRI application because
number of voxels in MRI is independent of time space so it is
unpredictable. So SICA has good practical feasibility than the Temporal
independent component analysis.
4.3.4 PREPROCESSING STEPS IN ICA
The preprocessing method is used to segment the MR image and it
consists of two types:
1. CENTERING
2. WHITENING
4.3.4.1 CENTERING
The most basic and necessary pre-processing is to centre x, subtract its
mean vector m = E{x} where X is a zero-mean variable. This implies that
s is zero-mean as well, as can be seen by taking expectations on both
sides. This pre-processing is made solely to simplify the ICA algorithms.
4.3.4.2 WHITENING
ICA or statistical model is represented as X=AS, Where W= A-1,
this transformation takes place through the observed vector x linearly as
˜x which is white. The covariance matrix of ˜x equals the identity matrix
E{˜x˜xT } = I.
It reduces the number of parameters to be estimated by considering
the original matrix A, there are n2 parameters but we only need to
estimate the new, orthogonal mixing matrix ˜A.
81
Fig 4.8 Orthogonal Mixing Matrix
JOINT DISTRIBUTION OF WHITENED MIXTURES
Because whitening is a very simple and standard procedure, it
reduces the complexity and reduces the dimension of the data. When
considering in PCA it proceeds as follows:
1. Obtain data
2. Subtract the mean
3. Calculate the covariance matrix
4. Calculate the Eigen vector and Eigen value.
Thus the highest Eigen value is obtained as principle and it also
retains the lowest Eigen value which produces noise. But in ICA Eigen
values which are too small are discarded. Thus it enhances in reducing
the noise in an image.
82
4.4 COMPARISON OF PCA AND ICA
The basis images found by PCA depend only on pair wise
relationships between pixels in the image database. In a task such as
brain tumor detection, in which important information may be contained
in the high-order relationships among pixel so Independent component
analysis (ICA), a generalization of PCA, is one such method.
Applying PCA on MR Images where pixel location and brain
images are treated as observation and measures respectively which leads
to maximum variability in pixels so the input does not throw high order
statistics. So, maximum amount of data cannot be separated.
Fig 4.8 PLOT OF ICA AND PCA
83
Chapter-5Result
84
CHAPTER 5
RESULT
OUTPUT OF CLASSIFIED TUMOR
Tumor Image 1 GLIOBLASTOMA
Tumor Image 2 ASTROCYTOMA
85
Tumor Image 3 LYMPHOMA
Tumor Image 4 MENINGLOMA
86
Chapter-6Conclusion
87
CHAPTER 6
CONCLUSION
The results show that Watershed Segmentation can successfully
segment a tumor provided the parameters are set properly. The watershed
method did not require an initialization while the others require an
initialization inside the tumor. The visualization and quantitative
evaluations of the segmentation results demonstrate the effectiveness of
this approach. Watershed Segmentation algorithm performance is better
for the cases where the intensity level difference between the tumor and
non tumor regions is higher. It can also segment non homogenous tumors
providing the non homogeneity is within the tumor region. This paper
proves that methods aimed at general purpose segmentation tools in
medical imaging can be used for automatic segmentation of brain tumors.
The quality of the segmentation was similar to manual
segmentation and will speed up segmentation in operative imaging.
Among the segmentation methods investigated, the watershed
segmentation is marked out best out of all others. The user interface in
the main application must be extended to allow activation of the
segmentation and to collect initialization points from a pointing device
and transfer them to the segmentation module. Finally the main program
must receive the segmented image and present the image as an opaque
volume and the type of the tumor is also detected using ICA
Algorithm.
88
appendix
89
APPENDIX
clc;
s=input('ENTER THE IMAGE FILE NAME TO
TRAIN::','s');
i=imread(s);
k=trainimage_filtering(i);
figure,imshow(k);
title('FILTERED IMAGE');
n1=imcrop(k);
n2=imcrop(k);
dd=trainimage_segment(n2);
f=ica_training(n1,n2);
disp(f);
if(f>248 || f<256)
figure,imshow('tissue1.bmp');
title('ASTROCYTOMA');
90
disp('detected tumour is astrocytoma'); display disease name
elseif(f>224 || f<228)
figure,imshow('tissue2.bmp');
title(' GLIOBLASTOMA');
disp('detected tumour is glioblastoma'); display disease name
elseif(f>238 || f<240)
figure,imshow('tissue3.bmp');
title(' LYMPHOMA');
disp('detected tumour is lymphoma');
elseif(f>263 || f<290)
figure,imshow('tissue4.bmp');
title(' MENINGLOMA');
disp('detected tumour is meningioma');
else
disp('unknown detected or no tumour found');
end
disp('project completed successfully');
function [k]=trainimage_filtering(i)
d=rgb2gray(i);
91
c=input('ENTER THE CORRESPONDING VALUE FOR
FILTERING:1-SOBEL,2-PREWITT,3-MEDIAN,4-LAPLACIAN:');
switch (c)
case 1
h=fspecial('sobel');
k=imfilter(d,h);
case 2
h=fspecial('prewitt');
k=imfilter(d,h);
case 3
k= medfilt2(d,[5 5]);
otherwise
h=fspecial('laplacian');
k=imfilter(d,h);
end
Explanation for User Defined Function Segmentation:
function [n1]=trainimage_segment(g)
BW = edge(g,'canny',0.2);
[imx,imy]=size(BW);
92
msk=[0 0 0 0 0;
0 1 1 1 0;
0 1 1 1 0;
0 1 1 1 0;
0 0 0 0 0;];
B=conv2(double(BW),double(msk));
L = bwlabel(B,8);
mx=max(max(L));
[r,c] = find(L==2);
rc = [r c];
[sx sy]=size(rc);
n1=zeros(imx,imy);
for i=1:sx
x1=rc(i,1);
y1=rc(i,2);
n1(x1,y1)=255;
end
RGB = label2rgb(B);
figure,imshow(RGB,[]);
93
title('SEGMENTED IMAGE(AFFECTED REGION');
end
function[hss]=ica_training(v1,v2)
M = 2;
N = 100;
v1=double(v1);
v2=double(v2);
v1=v1(1:N);
v2=v2(1:N);
v=[v1,v2];
A=ones(1,N*2);
x =v.*A;
W = eye(1,N*2);
y = x.*W;
maxiter=100;
eta=1;
/*****************************************************/
94
CHAPTER 7References
95
CHAPTER 7
BIBLIOGRAPHY
1. ICGST-GVIP Journal, ISSN 1687-398X, Volume (9), Issue (III),
June’09
2. L.P. Clarke, R.P.Velthuizen, M.A. Camacho, J.J. Heine, M
Vaidyanathan, L.O. Hall, R.W. Thatcher, and M.L. Silbinger: MRI
Segmentation: Methods and Applications. Magnetic Resonance
Imaging, 1995.
3. Medical image analysis, volume 2, issue 2,march 1998.
4. Information Technology in Biomedicine, IEEE Transactions on
sep 2005.
5. Medical Image Computing and Computer-Assisted Intervention
6. MICCAI,2002. Indian Journal of Science and Technology Vol.2
No 2 (Feb. 2009) ISSN: 0974- 6846
7. Jonathan sachs (1996)”Digital Image Basics”.
8. www.icgst.com
9. www.ieeeexplorer.com
96
97