FCAViewFCAView Tab:Tab:A ConceptA Concept--Oriented View Generation Tool for Oriented View Generation Tool for Clinical Data Using Formal Concept AnalysisClinical Data Using Formal Concept Analysis

GuoqianGuoqian JiangJiang, Katsuhiko Ogasawara, Naoki , Katsuhiko Ogasawara, Naoki NishimotoNishimoto, ,

Akira Akira EndohEndoh, , TsunetaroTsunetaro SakuraiSakurai

Department of Medical Informatics, Hokkaido University Department of Medical Informatics, Hokkaido University Graduate School of Medicine, Sapporo, JapanGraduate School of Medicine, Sapporo, Japan

The 8The 8thth International ProtInternational Protééggéé ConferenceConferenceMadrid, Spain, July 20, 2005Madrid, Spain, July 20, 2005

Classification of Views for Electronic Patient RecordsClassification of Views for Electronic Patient Records

From Q.Zeng, et al. Journal of Biomedical Informatics. 2001;34(2):112-128

ConceptConcept--Oriented ViewsOriented Views

?? Several studies have demonstrated that Several studies have demonstrated that computercomputer--generated, conceptgenerated, concept--oriented views for oriented views for clinical data can be used to reduce cliniciansclinical data can be used to reduce clinicians’’information overload and improve the accuracy information overload and improve the accuracy of clinical data retrieval.of clinical data retrieval.

?? However, the tasks for automating the However, the tasks for automating the generation of conceptgeneration of concept--oriented views are oriented views are challenging.challenging.

Formal Concept AnalysisFormal Concept Analysis?? Formal concept analysis (FCA), a mathematical Formal concept analysis (FCA), a mathematical

approach to data analysis based on lattice theory, approach to data analysis based on lattice theory, has been advocated to represent and process has been advocated to represent and process medical knowledge in different topics related to medical knowledge in different topics related to decision support systems.decision support systems.

Example: A familyExample: A family

Study PurposeStudy Purpose

?? In the study, the potential role of formal In the study, the potential role of formal concept analysis on conceptconcept analysis on concept--oriented view oriented view generation for clinical data is explored.generation for clinical data is explored.

MaterialsMaterials?? Clinical DataClinical Data??A collection of 386 Japanese discharge summaries A collection of 386 Japanese discharge summaries

obtained from cardiovascular domain.obtained from cardiovascular domain.?? Knowledge sourcesKnowledge sources??The Japanese version of The Japanese version of MeSHMeSH (MSHJPN) in (MSHJPN) in

UMLS2005AA (54,550 entries).UMLS2005AA (54,550 entries).??Developed by the NLMDeveloped by the NLM

??MEDIS version 2.3 MEDIS version 2.3 ----The Japanese standard The Japanese standard dictionaries of diagnosis terms (19,660 entries) and dictionaries of diagnosis terms (19,660 entries) and procedural terms (9,686 entries).procedural terms (9,686 entries).??Developed by the MEDISDeveloped by the MEDIS--DC in JapanDC in Japan

System ConstructionSystem Construction??FCAViewFCAView Tab:Tab: A ProtA Protééggéé--2000 tab plug2000 tab plug--inin??A virtual patient records systemA virtual patient records system??A ontologyA ontology--based knowledge basebased knowledge base??Medical conceptsMedical concepts??Formal concept analysisFormal concept analysis

??A natural language processing (NLP) moduleA natural language processing (NLP) module??A formal concept analysis (FCA) moduleA formal concept analysis (FCA) module

The structure of The structure of discharge summary discharge summary

represented by a represented by a XML fileXML file

<?xml version="1.0" encoding="Shift_JIS"?><!DOCTYPE DischargeSummary SYSTEM "DischargeSummary.dtd"><DischargeSummary>

<PatientDischargeSummary><FileInformation>

<FileID>pcdata</FileID><PatientID>pcdata</PatientID>

</FileInformation><PatientInformation>

<PatientAge>pcdata</PatientAge><PatientSex>pcdata</PatientSex><PatientOccupation>pcdata</PatientOccupation>

</PatientInformation><DateInformation>

<AdmissionDate>pcdata</AdmissionDate><DischargeDate>pcdata</DischargeDate>

</DateInformation><DiagnosisList>

<PrimaryDiagnosis>pcdata</PrimaryDiagnosis> <SecondaryDiagnosis>pcdata</SecondaryDiagnosis>

</DiagnosisList><PatientSummary>

<Complaint>pcdata</Complaint><CurrentHistory>pcdata</CurrentHistory><PastHistory>pcdata</PastHistory><OtherHistory>pcdata</OtherHistory><ExaminationRecords>pcdata</ExaminationRecords><InvestigationRecords>pcdata</InvestigationRecords><ProgressRecords>pcdata</ProgressRecords>

</PatientSummary></PatientDischargeSummary>

</DischargeSummary>

An Example of the Virtual Patient Records An Example of the Virtual Patient Records

OntologyOntology--based Knowledge Basebased Knowledge Base

Class

Slot

Instance Case id

NLP ModuleNLP Module?? ChaSenChaSen version 2.3version 2.3

?? A Japanese morphological analysis systemA Japanese morphological analysis system?? Open sourceOpen source?? Developed by Nara Institute of Science and Technology Developed by Nara Institute of Science and Technology

?? MACDMACD?? Morphological Analyzer Connectivity DriverMorphological Analyzer Connectivity Driver--model model ?? As the Java interface of As the Java interface of ChaSenChaSen?? Developed by Nara Institute of Science and TechnologyDeveloped by Nara Institute of Science and Technology

?? User dictionaries:User dictionaries:?? MSHJPNMSHJPN?? MEDIS version 2.3MEDIS version 2.3

FCA ModuleFCA Module

?? The Java API of Concept Explorer version 1.2The Java API of Concept Explorer version 1.2??Open source projectOpen source project

?? To generate the concept oriented views for the To generate the concept oriented views for the clinical data clinical data

Results (I)Results (I)?? View generation for individual patient recordsView generation for individual patient records??The items about the structure knowledge of the patient The items about the structure knowledge of the patient

records used as the formal objectsrecords used as the formal objects??Primary diagnosisPrimary diagnosis??Secondary diagnosesSecondary diagnoses??Chief ComplaintsChief Complaints??Current historyCurrent history??Past historyPast history??Other historiesOther histories??Examination recordsExamination records?? Investigation recordsInvestigation records??Progress recordsProgress records

??The medical concepts extracted from each part used as The medical concepts extracted from each part used as the formal attributesthe formal attributes

Context of an Individual Case by CrossContext of an Individual Case by Cross--tabletable

Medical Concepts

Structural Items

ConceptConcept--Oriented View of an Individual Case by LatticeOriented View of an Individual Case by Lattice

Effort Angina

Shortness of breath,A feeling of general fatigue

Impaired liver function

Results (II)Results (II)

??View generation for the similar casesView generation for the similar cases??The similar cases indicated by the case ids The similar cases indicated by the case ids

were used as the formal objectswere used as the formal objects??The medical concepts extracted from each The medical concepts extracted from each

case were used as the formal attributescase were used as the formal attributes

Context of the Similar Cases by CrossContext of the Similar Cases by Cross--tabletable

Medical ConceptsCase Ids

ConceptConcept--Oriented View of the Similar Cases by LatticeOriented View of the Similar Cases by Lattice

Effort Angina

Chest Pain

Chest Discomfort

Hyperlipidemia

Shortness ofBreath

Diabetes Mellitus Type II

Results (III)Results (III)

?? Implementation of information model for Implementation of information model for practical usepractical use??Converting the lattice diagram into a concept Converting the lattice diagram into a concept

treetree??Developing the filters to restrict the context sizeDeveloping the filters to restrict the context size

For Context OfThe Similar Cases

For Context OfThe Individual CaseImpaired Liver FunctionEffort Angina

?? Context filter using Context filter using Structure KnowledgeStructure Knowledge?? Primary diagnosisPrimary diagnosis?? Secondary diagnosisSecondary diagnosis?? Chief complaintsChief complaints?? Current historyCurrent history?? Past historyPast history?? Other historyOther history?? Examination recordsExamination records?? Investigation recordsInvestigation records?? Progress recordsProgress records

Effort Angina

?? Context filter using Context filter using Semantic KnowledgeSemantic Knowledge of of medical conceptsmedical concepts??NounNoun--Diagnostic TermDiagnostic Term??NounNoun--Procedural TermProcedural Term??NounNoun--ModifierModifier??NounNoun--MeSHMeSH TermTerm??NounNoun--Compound TermCompound Term??NounNoun--OthersOthers

Effort Angina

?? The filtered context:The filtered context:?? Formal objects: Formal objects:

??3232 cases with primary cases with primary diagnosis of diagnosis of Effort Effort AnginaAngina

?? Formal attributes:Formal attributes:??All medical concepts All medical concepts

with semantic type of with semantic type of Diagnosis TermDiagnosis Term, , extracted from the extracted from the Chief ComplaintsChief Complaints part part of recordsof records

Effort Angina

Chest Pain

DiscussionDiscussion?? Automating view generation by integrating the Automating view generation by integrating the

NLP, OntologyNLP, Ontology--based knowledge base, with the based knowledge base, with the FCA technique.FCA technique.??The individual patient recordsThe individual patient records??The similar casesThe similar cases

?? Features of the systemFeatures of the system?? Processing freeProcessing free--text clinical datatext clinical data?? Providing flexible way Providing flexible way

?? for browsing using concept treefor browsing using concept tree?? for adapting the information needs of clinicians using for adapting the information needs of clinicians using

context filters that are based on structural knowledge and context filters that are based on structural knowledge and semantic knowledgesemantic knowledge

DiscussionDiscussionGeneralization Consideration (I)Generalization Consideration (I)

?? To support HLTo support HL--7 Clinical Data Architecture (CDA) as 7 Clinical Data Architecture (CDA) as input format.input format.

LocalClinical

Documents

LocalClinical

Documents

HL7CDA

Transformation Input

Concept-Oriented View System

DiscussionDiscussionGeneralization Consideration (II)Generalization Consideration (II)

?? To visualize the ProtTo visualize the Protééggéé--based based ontologiesontologies. . ?? We assume that the instances (i.e. the formal We assume that the instances (i.e. the formal

objects) of a class in Protobjects) of a class in Protééggéé--2000, together 2000, together with its own slots (i.e. the formal attributes), with its own slots (i.e. the formal attributes), could form a formal context that would interests could form a formal context that would interests the users. the users. ?? the instances of a class with its slots in Boolean typethe instances of a class with its slots in Boolean type?? the instances of a class with it slots in Multiple the instances of a class with it slots in Multiple

Instance type. Instance type.

AcknowledgementsAcknowledgements

?? The study was supported in part by the GrantsThe study was supported in part by the Grants--inin--Aid of the Japan Society for the Promotion of Aid of the Japan Society for the Promotion of Science. Science.

?? Special thanks to Mr. Harold R. Special thanks to Mr. Harold R. SolbrigSolbrig in in Division of Division of BioMedicalBioMedical Informatics, Mayo Clinic, Informatics, Mayo Clinic, Rochester, MN, who contributed the idea about Rochester, MN, who contributed the idea about the generalization of our the generalization of our FCAViewFCAView Tab.Tab.