faculty of computer science © 2006 cmput 605march 3, 2008 concept-based electronic health records:...

15
Faculty of Computer Science CMPUT 605 December 06, 2007March 3, 2008 © 2006 Concept-Based Electronic Health Records: Opportunities and Challenges S. Ebadollahi, S Chang, T. Mahmood, A Coden, A. Amir M. Tanenblatt 14th Annual ACM International Conference on Multimedia (2006) Amit Satsangi [email protected]

Post on 18-Dec-2015

213 views

Category:

Documents


0 download

TRANSCRIPT

Faculty of Computer Science

CMPUT 605 December 06, 2007March 3,

2008© 2006

Concept-Based Electronic Health Records: Opportunities and Challenges

S. Ebadollahi, S Chang, T. Mahmood, A Coden, A. Amir M. Tanenblatt

14th Annual ACM International Conference on Multimedia (2006)

Amit [email protected]

© 2006

Department of Computing Science

CMPUT 605

Focus

ECG Video: document is not important; behavior of sub-organs like valves, ventricles, myocardium is

ECG –Text Report sub organs, diagnosis

Efficient access to the elements of the content of the data ???

New Paradigm – Concept based Multimedia Medical Records

© 2006

Department of Computing Science

CMPUT 605

Problems with the present system

Electronic Health Records (EHR)

Data—mixed format: HIS for lab reports, ECG’s etc. RIS for reports generated after reviewing medical images, and PACS for diagnostic images.

Different Standards: HL7, DICOM, etc.

Information extraction regarding a single concept of interest (Right Atrium) is difficult

Hence the need for (re)organizing the health records at the information level

© 2006

Department of Computing Science

CMPUT 605

Concept-Based Records Organization: Advantages

Goes beyond dealing with data at the document level

Caters to different categories of users of medical records

– Physicians: Ejection fraction of left ventricle measured while reviewing

the ECG. Ideally system should calculate this using quantification

Algorithms. Should also be able to link it with the diagnosis reports,

textbooks, research papers etc.

– Students: Teaching files with history of medical cases + diagnostic

images + medical journals + textbooks

© 2006

Department of Computing Science

CMPUT 605

Concept-Based Records Organization: Advantages

– Patients: Illustrated version of patient’s disease

– Insurance companies: Prevent misuse of expensive tests (MRI) when not justified by the results of earlier, less expensive tests (EKG)

Timely and decision-enabling information extraction

It entails a better organization of medical records from the

scratch in order to deliver all that is promised …

© 2006

Department of Computing Science

CMPUT 605

Architecture

Analytic Engines

—domain knowledge

Heart Chambers in Video

Parse diagnosis report

Relationships b’n concepts

—ontologies (UMLS)

Is a, spatially/temporally/

functionally related to etc.

© 2006

Department of Computing Science

CMPUT 605

Example

© 2006

Department of Computing Science

CMPUT 605

Addendum

New information may need to be added

Graph Structure with Nodes as concepts and links are

relationships between these concepts

Need federation of Ontologies – different concepts of interest in

different domains

Multimedia content restructuring required – Vision, NLP etc.

Not a new way of analyzing data, but a novel way of organizing

the medical records

© 2006

Department of Computing Science

CMPUT 605

Case Study: Video Content Restructuring

Echocardiography – Imaging of the heart in several planes

Inherent spatio-temporal strcuture

Feature-extraction tools used to target areas of interest

Text snippets extracted from diagnosis report

Undirected graphical models used to learn the spatial

arrangement of cardiac chambers

© 2006

Department of Computing Science

CMPUT 605

Schematic

© 2006

Department of Computing Science

CMPUT 605

Text Analytics for Cancer Pathology Reports

MedTAS (Medical Text Analysis System) was used

Several models – conceptually separate pieces of

knowledge

Pieces of knowledge Disease description,

evaluation procedures etc.

4 sub-models: Tumor model, Specimen model,

Lymph-node model and the disease model

© 2006

Department of Computing Science

CMPUT 605

Text Analytics for Cancer Pathology Reports

Models are annotators (can be institution specific)

MedTAS built on IBMs Ustructured Information

Management Architecture (UIMA) . (Open Source)

© 2006

Department of Computing Science

CMPUT 605

Models

© 2006

Department of Computing Science

CMPUT 605

Potential Avenues

Three main issues

— Determining the unifying architecture

— Determining the concepts that need to be extracted

— Development of robust Analytic engines

Testing & Feedback issues when such records in use

Seamless Integration with existing data

© 2006

Department of Computing Science

CMPUT 605

Thank You For Your Attention!