Ontologies and Bayesian Networks in Medical Diagnosis

The amount of information that must be taken into ac count in medical diagnosis is huge and subject to evolution. Ontologies are a means for formalizing the concepts of the domain of interest. Open, interoperable ontologies already exist for the biomedical field, enabling scientists to communicate with minimum ambiguity. Unfortunately, reasoners acting upon ontologies operate in a deterministic manner, which is unsuitable for the medical domain, where uncertainty must also be taken into account. Bayesian networks (BNs) offer a coherent and intuitive representation of uncertain domain knowledge. This paper presents an approach to the use of ontologies and BNs in medical diagnosis. The approach is based on the adoption of predefined structures for the BNs. These lead to reduced extensions to the domain ontology, yet allowing probabilistic analysis.