Ontology-based Diagnostic Decision Support for Radiology Charles E. - - PDF document

Ontology-based Diagnostic Decision Support for Radiology

Charles E. Kahn, Jr., MD, MS

Medical College of Wisconsin Milwaukee, Wisconsin, USA

MIE 2014

Goals

• Apply an ontology of disorders and associated

imaging findings for differential diagnosis

• Develop user interface and web service API for

knowledge query

• Demonstrate valid application of the ontology’s

knowledge

Radiology Gamuts Ontology (RGO)

• Large and growing knowledge model

▫ Diseases ▫ Imaging observations

• Constructed from several differential diagnosis

references

• Serves as a form of "computable knowledge" to

aid in radiological diagnosis

Ontology Structure

• Entity

▫ disorder chronic hepatitis ▫ intervention steroid therapy ▫ observation misty mesentery

• Relation

▫ subsumption (“is a”) ▫ causality (“may cause”)

Radiology Gamuts Ontology

• Standardized knowledge representation

▫ Normalized entity names + synonyms ▫ Ability to reason over domain

• Interoperability / knowledge integration

▫ Mappings to other ontologies ▫ Knowledge-based applications

Gamuts Ontology

• Content

▫ 16,822 terms ▫ 1,755 subsumption relations ▫ 53,638 causal relations

• User interface

▫ Web site (www.gamuts.net) ▫ RESTful web service / JSON

Imaging findings and diseases Causal relationships

gamuts.net

Differential Diagnosis (“DDx”)

• Identify the cause(s) of a set of findings

▫ Each diagnosis explains all imaging findings that have been asserted

• “Discriminators”

▫ Union of findings of each listed diagnosis

• exclude asserted findings
• exclude findings common to all listed diagnoses

▫ Additional findings that can reduce the DDx list

DDx Example

• Finding

▫ hypertelorism

• Diagnosis

▫ 151 diagnoses ▫ “Aarskog syndrome” to “XXXY syndrome”

• Discriminators

▫ 976 discriminators ▫ including acro-osteolysis, abnormal sternum, clubfoot, tracheomalacia, Wormian bones

DDx Example: 2nd finding

• Findings

▫ hypertelorism ▫ abnormal sternum

• Diagnosis (disorders that cause both findings)
• Discriminators

▫ Brachmann-de Lange syndrome ▫ cleidocranial dysostosis ▫ Noonan syndrome ▫ Rubinstein-Taybi syndrome ▫ Seckel syndrome ▫ XXXXY syndrome

DDx Example: 3rd finding

• Findings

▫ hypertelorism ▫ abnormal sternum ▫ Wormian bones

• Diagnosis (disorders that cause all 3 findings)

▫ cleidocranial dysostosis

• Discriminators

▫ (none)

Gamuts DDx

• User interfaces

▫ Web site (gamuts.net/ddx) ▫ Web service (see gamuts.net/dev)

gamuts.net/ddx

gamuts.net/ddx

"finding": [ "hypertelorism", "Wormian bones" ], "diagnosis": [ "aminopterin fetopathy", "cleidocranial dysostosis", "metaphyseal chondrodysplasia Jansen type", "normal variant", "Ritscher-Shinzel syndrome", "Schinzel-Giedion syndrome", "sclerosteosis" ], "discriminator": [ { "name": "abnormal fibula", "n": "1" }, { "name": "widespread predominantly medullary

• steosclerosis",

"n": "1" }

Mappings

• RadLex
• SNOMED CT
• OrphaNet Rare Disease Ontology (ORDO)
• Bone Dysplasia Ontology (BDO)

Conclusions

• Working system for differential diagnosis
• Valid application of diagnostic knowledge
• Requires diagnoses to cause all asserted findings

Future Work

• Test on clinical cases
• Apply “transitive closure” over ontology
• Integrate with NLP tools
• Incorporate probabilistic knowledge + reasoning

Thank you !