HeC Technical Review, Work Package 6 LINKING EXTERNAL KNOWLEDGE TO THE PATIENT CASE

1
HeC Technical Review, Work Package 6 LINKING
EXTERNAL KNOWLEDGE TO THE PATIENT CASE
Maat Gknowledge Jaume I University (TKBG
group) PhD Student Ernesto Jiménez-Ruiz

• 14-16 January 2008, Geneva

2
Outline

• Introduction
• Motivation
• Use Case
• Objectives
• Used state-of-the-art techniques
• Text mining
• Ontology Fragment Extraction and Segmentation
• 3D Brower-like Tool
• Methodology
• Prototype
• WP6 Workflows

3
Motivation

• Work complementary to HeC Toolbar
• Linking, accessing and using external resources
• Medline
• Public online databases (e.g. ORPHANET, GeneMap)
• Knowledge resources (e.g. biomedical ontologies)
• Helping the clinician in integrating with her
  everyday environment
• Clinicians use a web browser (IE or FF) to find
  papers/search public databases for information
• Novel research in a particular topic, association
  of one or more clinical conditions/ pathologies/
  malformations etc.

4
Use Case (I)

• Clinicians have created a new Patient Folder
  within the Hypertrophic Cardiomyopathy (HCM)
  study (in the HeC database).
• Clinicians want to gets back the list of HCM HeC
  related patients diagnosed of a rare subtype of
  the disease, and which is of interest to him for
  obtaining interesting treatments/opinions.
• Clinicians want to find research articles that
  describe related research work.

5
Use Case (II)

• Clinicians presents a text based query describing
  a HeC patient case.
• Clinicians want to get related patient cases and
  research articles in order to
• obtain additional knowledge
• be able to refine and resubmit the query if
  necessary
• Clinicians want to use Ontologies to help them
  in
• the query refinement specification or
  generalization of query-concepts
• the representation of known links between
  query-concepts
• the search look for a semantic type (Disease or
  Syndromes)

6
Motivation (III)

• Objectives
• Creation of a browser-like tool to establish a
  connection between
• Patient-based (Text-like) queries,
• Medical terminologies like UMLS Metathesaurus
• Textual resources research articles
• Fragments from domain ontologies.
• Needed HeC related resources
• A collection of text-rich resources annotated
  with UMLS
• A set of domain ontologies mapped to UMLS.
• Patient data annotated with UMLS

7
UNDERLYING STATE-OF-THE-PART TECHNIQUES

• Text mining techniques
• Ontology Segmentation techniques

8
Text Mining Techniques

• Collaboration with the Rebholzs Text Mining
  group at European Bioinformatics Institute,
  Cambridge
• EBI Infrastructure and techniques
• Biomedical Entity Recognition
• Flexible processing of a text input stream
• Flexibility in the use of different lexicon
  dictionaries
• Contribution
• Treatment of the UMLS Metathesaurus
• Creation of UMLS-based input lexicon-dictionary
• Relation Discovery
• Currently based on co-occurrences at sentence
  level
• Used abstracts coming from Medline

9
Ontology Fragment Extraction Techniques

• Ontologies provide Surrounding information about
  the knowledge of interest
• Galen, NCI, UMLS Semantic Network, UMLS
  Metathesaurus
• Domain ontologies in Medicine are rather huge so
  we need to deal with them in a modular-fragmented
  way.
• OntoPath Language and Tool
• Flexibility in the extraction of fragments
  (knowledge on demand) ??
• Small and specific fragments can be obtained ?
• It lacks a underlying formal approach ?
• It does not allow approximate queries ?
• Locality-based modularization (Collaboration with
  IMG from Manchester)
• Proposes a formal approach ??
• Problems dealing with big knowledge resources ?
• It does not allow approximate queries ?
• Approximate Queries (ArHex)
• Extraction of approximate fragments by means tree
  patterns. ??
• The idea is to use the three techniques (combined
  or independently) depending on the case.

10
BROWSER TOOL

• Methodology
• Prototype

11
Methodology for the Browser Tool

• Step 0 Annotation/Alignment of resources with
  UMLS
• Extraction and annotation of a collection of
  text-rich resources, related to a specific HeC
  domain (i.e. JIA ,Cardiomyophaties or Brain
  Tumors).
• Selection of domain ontologies of interest
  (Galen, NCI and the UMLS itself) and alignment
  with UMLS.
• Patient data annotated should be annotated with
  UMLS.

12
Methodology for the Browser Tool

• Step 1 Definition of a text-based query
• Gene mutations in sarcomere involved in
  cardiomyopathies related to patient relatives
• ltze ids"C0596611" sem"Genetic Function"gtGene
  mutationslt/zegt in ltze ids"C0036225" sem"Cell
  Component"gtsarcomerelt/zegt involved in ltze
  ids"C0878544" sem"Disease or Syndrome"gtcardiomyo
  pathieslt/zegt related to ltze ids"C0080103"
  sem"Family Group"gtpatient relativeslt/zegt

13
Methodology for the Browser Tool

• Step 2 Extraction of an ontology fragment (I)
• Knowledge represented in NCI w.r.t.
  Cardiomyopathy

14
Methodology for the Browser Tool

• Step 2 Extraction of an ontology fragment (II)
• Knowledge represented in Galen relating/linking
  sarcomere with cardiomyopathies DCM and HCM

15
Methodology for the Browser Tool

• Step 3 Extraction of Medline abstracts and
  patient cases
• In Step 0 a set of abstracts related to the
  domain were extracted.
• The most relevant abstracts, w.r.t. the query
  entities are selected and ranked given a
  relevance measure
• Query Entities Patient Relatives, DCM, HCM,
  Myocardium, Sarcomere Kind of Genes (UMLS
  semantic type)
• For the retrieval of patient records the idea is
  similar, we should look for the occurrence of the
  given entities within the patient data, and to
  extract the most relevant/interesting cases.
• Other techniques could be applied

16
Methodology for the Browser Tool

• Step 4 Extraction of co-occurrences
• Entities Patient Relatives, DCM, HCM,
  Myocardium, Sarcomere Kind of Genes (UMLS
  semantic type)
• Gene Mutation is a rather general entity so we
  look for the semantic type.

17
Methodology for the Browser Tool

• Step 5 Customization of the conceptual map and
  query
• We want to focus on HCM

18
Prototype for the Browser Tool (I)

• Integration with HeC Gateway by means of an
  AJAX-like client

19
Prototype for the Browser Tool (II)
20
WP6 WORKFLOW
21
Workflow between WP6 efforts (I)

• Joint use with HeC Toolbar
• HeC Toolbar
• Integration between HeC Client application (where
  a patient of interest is identified) and a web
  browser
• URLs of interesting resources could be saved in
  the system in the form of a bookmark which
  associates the resource with the case or the
  patient or part of the case, like an image.
• Extracted URL of Medline abstracts with the 3D
  Knowledge Browser could be selected and saved by
  means the HeC Toolbar.
• HeC Toolbar Firefox http//tamas.web.cern.ch/tama
  s/hectoolbar/

22
Workflow between WP6 efforts (II)

• Joint use with HeC CaseReasoner
• The 3D Knowledge Browser extracts patient cases
  of interest
• From patient cases can also be extracted
  co-occurrences and interesting relations
• However used techniques (eg clustering) when
  working with structured data are different from
  those used in text mining
• The tools for global similarity search (HeC
  CaseReasoner) analyse similarity patterns of
  entities within the extracted set of patients,
  which can have useful information as treatments
  or problems in the diagnosis.

23
Questions and Feedback

• Thank you!!
• Interesting Links
• 3D Browser draft report
• https//www.health-e-child.org/wps/wp6/documents/
  3d-browser
• EBI Work
• Report http//krono.act.uji.es/publications/techr
  ep/tkbg-ebi-report
• Software http//www.ebi.ac.uk/Rebholz/software.ht
  ml
• Contact
• E-mail ejimenez_at_uji.es
• Web page
• http//krono.act.uji.es/people/Ernesto
• http//www3.uji.es/ejimenez