Transcript MedAT - Knowledge Engineering Group

MedAT:
Medical Resources
Annotation Tool
Monika Žáková*, Olga Štěpánková*, Taťána Maříková
*Department
of Cybernetics, CTU Prague
Institute of Biology and Medical Genetics, Prague
[email protected], [email protected]
1
Outline
1.
Motivation
2.
System Description
3.
Creating Annotation
4.
Additional Functionalities
5.
Knowledge Representation
6.
Ontologies

Task Ontologies
 Domain Ontologies
7.
2 / 20
Results and Conclusion
KEG seminar 8.12.2005
Motivation

Patients’ records represent a valuable source of information

Records stored in semi-structured text files

For sharing and data mining format such as ontology or
relational database needed

Currently known methods for text mining not applicable,
since
Records heterogeneous – type of examination, personality of doctor
 Abbreviations used (some non-standard)
 Gazetteers not available in Czech

3 / 20
KEG seminar 8.12.2005
Motivation II

Grant “Relational ML for analysis of biomedical data” of the
Czech national research program Information Society in
cooperation with the Institute of Biology and Medical
Genetics, 2nd Medical Faculty of Charles University

Relational data mining using subgroup discovery
methodology

Need to transform data from text files into a form suitable
for relational data mining i.e. relational database and rules
4 / 20
KEG seminar 8.12.2005
System overview
Ontologies
Forms generator
Medical record
Knowledge base
5 / 20
Relational
database
KEG seminar 8.12.2005
System description

Creating semantic annotations of medical records

Based on Dynamic Narrative Authoring Tool

Modular architecture

Export to knowledge base in OCML, OWL

Export to relational database

Visualization – genealogical tree
6 / 20
KEG seminar 8.12.2005
MedAT GUI
7 / 20
KEG seminar 8.12.2005
Creating Annotations

Dynamically generated forms

A form  one major class in ontology  master table in
the database

E.g. Patient, Examination

Adding abbreviations and aliases to the ontology

Filling of forms

Automatically by parsing
 Drag and drop from records in text format
 Manually in case OCR not effective
8 / 20
KEG seminar 8.12.2005
Creating Annotations II
9 / 20
KEG seminar 8.12.2005
Additional functionalities

Exploration of data stored in the relational database



Pre-defined SQL queries – knowledge of SQL not required
Writing queries directly in SQL
Visualization

10 / 20
Genealogy tree
KEG seminar 8.12.2005
Knowledge representation

Core formalism – Apollet
Apollet

Frame-based formalism based on OCML

Formalism used by Apollo ontology editor => possibility to
use I/O modules of Apollo

Export to lisp available

Inference engine available

Disadvantage: rules very often just lisp functions
11 / 20
KEG seminar 8.12.2005
Relational database

Tables of the relational database generated automatically
from the ontology

Semantic description of the database given by an ontology

Export done in a batch for a particular version of ontology
and knowledge base

Export intended for a data mining experiment

Currently PostgreSQL database used
12 / 20
KEG seminar 8.12.2005
Ontologies
MedAT relies on ontologies on 2 levels:

Task ontologies
 Describe

structure of different medical records
Domain ontologies
 Formalize
knowledge about a specific domain
e.g. diseases, family relations, time points
13 / 20
KEG seminar 8.12.2005
Task Ontologies

Developed on basis of procedures and structure of medical
records in cooperation with medical doctors

Hierarchy induced by part-whole relationship
OCML – slots with facets
 OWL – hasPart, partOf (W3C Working Draft)


Serve as basis for generating of forms and tables in
relational database
14 / 20
KEG seminar 8.12.2005
Task Ontologies - Example

Classes - elements of medical
records e.g. object of
examination, therapy

Slots – description of
composition of medical
records e.g. class
examination has slots date,
doctor, has_therapy
15 / 20
KEG seminar 8.12.2005
Domain ontologies

Use of third party ontologies e.g. GALEN, Gene Ontology

Ontology of family relations

Need for rules e.g. hasHalfBrother(x,y)
hasHalfBro ther ( x, y )  male ( y ) 
a , b:( hasParent( x, q )  hasParent( y , a ))
 ( hasParent( x, b )  hasParent( y , b ))
OWL – no standardized rule language (ORL)
 OCML – lisp functions

16 / 20
KEG seminar 8.12.2005
Time ontology
Time ontology

Developed originally for historical narratives

Based on Allen’s algebra

Uncertain time points and intervals



Extended to cover time events specific for medical domain


Uncertain temporal position
Uncertain granularity
E.g. before surgery, during infancy
Available in OCML
17 / 20
KEG seminar 8.12.2005
Results

Easily transfer information from medical reports to
dynamically generated forms

Data from forms saved to a knowledge base and relational
database

Iterative extending of ontologies, adding aliases and
abbreviations

Tool currently being tested at the Institute of Biology and
Medical Genetics for patients with neurofibromatosis type 1
18 / 20
KEG seminar 8.12.2005
Future work

Text mining methods for semi-automatic annotation

Tool for semantic search and retrieval of a relevant subset
of data and visualization of retrieved data

Use of annotated data along with information about
genotype for data mining using subgroup discovery
methodology
19 / 20
KEG seminar 8.12.2005
Questions
Thank you for your attention
Questions???
20 / 20
KEG seminar 8.12.2005