Vers la définition automatique des éléments de données des fiches RCP en cancérologie à partir d’une ontologie

Abstract

Semantic interoperability based on ontologies allows systems to combine their information and process them automatically. The ability to extract meaningful fragments from ontology is a key for the ontology re-use and the construction of a subset will help to structure clinical data entries. The aim of this work is to provide a method for extracting a set of concepts for a specific domain, in order to help to define data elements of an EHR. Method: an extraction algorithm was developed to extract, from the National Cancer Institute’s Thesaurus (NCIT) and for a specific disease (i.e. prostate neoplasm), all the concepts of interest into a sub-ontology. The extraction takes as input, one or several key concepts. We compared all the concepts extracted to the concepts encoded manually and contained into the multi-disciplinary meeting report form (MDMRF). Results: NCIT contained 83143 concepts from which we extracted two sub-ontologies: sub-ontology 1 contained 434 concepts (by using a single key concept) and sub-ontology 2 contained 480 concepts (by using 5 additional keywords). The coverage of sub-ontology 2 to the MDMRF concepts was 51%. The low rate of coverage is due to the lack of definition or misclassification of the NCIT concepts. By providing a subset of concepts focused on a particular domain, this extraction method helps to optimize the binding process of data elements and at maintaining and enriching a domain ontology.