Sharing knowledge in medicine: semantic and ontologic facets of medical concepts

The problem of making medical knowledge and medical data sharable over applications and reusable for several purposes is crucial. Ontologies can be used as a common framework for several systems. Principles for the construction of ontologies have been proposed: parsimony, clarity, categories versus terms, and coherence. Difficulties arise from the fact that medical concepts are empirical rather than perfectly defined, leading to semantic and conceptual adaptation. Medical knowledge has been built by the richness of meticulous patient observations, described by subtle attributes; e.g., ADM, a French program that assists the user in the diagnostic process, contains 1859 items for pain. Medical concepts are not fixed: the plague was well-identified by the XVIlth doctors but knowledge about the plague has changed. Medical terminology may be specific or vague although it is supposed to be universal. Moreover a medical concept occurs in a conceptual representation, so reference to several models is needed even for basic and a priori universal knowledge such as anatomy in medicine. As well as for concepts, sharable basic categories are difficult to obtain: this is illustrated by the SNOMED axes and the UMLS semantic types. Efficient computerised tools adapted to the biomedical domain must implement a sophisticated representation of biomedical concepts, based on some principles, taking into account the richness of medical semantics, and different levels of expertise in medical knowledge, separating roles and concepts, building a multifaceted concept space, modelling the relations between the concepts in order to compensate for the lack of a general theoretical basis of the domain.