Enrichment of Medical Ontologies from Textual Clinical Reports: Towards Improving Linking Human Diseases and Signs

Abstract

Healing a sick patient requires a medical diagnosis before proposing appropriate treatment. With the explosion of medical knowledges, we are interested in their exploitation to help clinician in collecting informations during diagnostic process. This article focuses on the development of a data model targeting knowledges available in both formal and non-formal resources. Our goal is to merge the strengths of all these resources to provide access to a variety of shared knowledges facilitating the identification and association of human diseases and to all of their available relevant characteristic signs such as symptoms and clinical signs. On one side, we propose an ontology produced from a merging of several existing and open medical ontologies and terminologies. On another side, we exploit real cases of patients whose diagnosis has already been confirmed by clinicians. They are transcribed in textual reports in natural language, and we show that their analysis improves the list of signs of each disease. This work results in a knowledges base loaded from the known target ontologies on the bioportal platform such as DOID, MESH and SNOMED for diseases and, SYMP and CSSO ontologies for all existing signs.