Interpreting Patient Data using Medical Background Knowledge

Transcription

1 Interpreting Patient Data using Medical Background Knowledge 1, Sonja Zillner 1, Bernhard Bauer 2, Matthias Hammon 3 1 Siemens AG, Corporate Technology, Munich, Germany 2 Software Methodologies for Distributed Systems, University Augsburg, Germany 3 University Hospital Erlangen, Germany Page 1

2 Overview 1. Problem: Integrate patient-data in clinical decision support systems 2. Our Approach: Development of a Disease-Symptom Knowledge Model 3. Prototype 4. Conclusion and Future Work Page 2

3 Annotations to unstructured clinical data Medical Images clinical data Problem: Reports Annotations representing the descriptive content of the clinical data need to be understood with respect to Ø diseases Ø examinations Ø treatment evaluation MEDICO-Annotation-Ontology lymph node lymph Problems: node of trunk lymph node of head Ø Data is mostly unstructured Lingual lymph node Ø Difficult facial to access lymph node all information needed in a specific clinical task malar lymph node size modifier Ø Clinicians don t use the mandibular full information lymph node contained shrunken in existing patient data enlarged Ø Ranking of likely diseases Ø Planning next examinations Annotations Interpretation e.g. Disease-Symptom-Ontology Clinical Decision Support Page 3

4 Search in the Diagnosis Process Search for Cancer-indicating symptoms ü There exist medical ontologies that cover symptom-related and disease-related information Ø There exist no medical ontologies covering the relationship between symptoms and diseases Page 4

5 Human Disease Ontology (DOID) ü Contains about 8000 diseases ü Well linked to MSH, SNOMED, UMLS, ICD, ü About 15 ObjectProperty relations owl:objectproperty total use described diseases doid:has_symptom doid:located_in doid:has_material_basis_in doid:transmitted_by doid:derives_from Ref: Page 5

6 Towards a Disease Symptom Knowledge Model Therapy Staging Etiology Importance Incidence Risk-factors Epidemiology Typical age has Differential Diagnosis Disease Probability Has (Leading) Symptom is (Leading)Symptom for Significance value Might detect Examination Risk Costs Knowledge Resource: Herold Innere Medizin Patient Symptom Shows Symptom Age Indicates to check for Anat. Region Modification Importance Intensity Gender Temporal Relevance Page 6

7 Disease Symptom Ontology disy:disease_symptom_relation disy:has_probability xsd:decimal disy:has_disease disy:has_symptom Outgoing from the Lymphoma use-case we included 5 diseases and 40 symptoms in the Disease-Symtpom-Ontology disy:disease disy:symptom disy:might_detect disy:examination disy:has_symptom RadLex SNOMED CT FMA DOID disy:patient disy:has_patient disy:patient_symptom_relation disy:datetime xsd:datetime disy:is_present disy:has_intensity xsd:boolean xsd:int Page 7

8 Linkage from disy to Other Ontologies lymph node lymph node of head lymph node of trunk MEDICO-Annotation-Ontology enlarged facial lymph node lingual lymph node malar lymph node size modifier mandibular lymph node shrunken disy:hasanatomicalregion_radlex disy:hasmodifier_radlex disy:enlarged_lymph_node disy:lymphoma disy:has_symptom is-a is-a disy:symptom disy:disease Disease-Symptom-Ontology Page 8

9 Use in Clinical Decision Support: Ranking likely Diseases Ranking Factors: ü proportion of present over absent symptoms of a disease ü age and gender specific incidence proportion ü leading symptoms ü symptom intensity ü intrinsic importance of symptoms Clinical Decision Support Systems: MYCIN, INTERNIST, CASNET, DXplain, CADIAG Page 9

10 Conclusion and Future Work Relations between diseases and symptoms make possible to integrate unstructured clinical data in decision support systems. Future Work: Ø Populating the Disease-Symptom Ontology / extending DOID Ø Creating Annotations: problems with German Ø Temporal information Ø Big Data: combining structured and unstructured data Page 10

11 Questions? Contact: Page 11