Semi-automatic bayesian network construction from ontology for e-health applications

Abstract

Ontology is widely used for representing and sharing bio information. However, ontology alone can not be used to represent non-deterministic relationship among E-health information such as symptoms and diseases. Although Bayesian network (BN) is one of the popular methods to represent and reason with probabilistic knowledge for E-health applications, it is normally a complex task to construct a BN model by collecting and analyzing domain knowledge. Therefore, we propose a semi-automatic way of constructing BN models from ontology for diagnosing diseases in the E-health applications. Our method allows users to select abstract levels to specify important terms for a specific application and to determine complexity. Our approach automatically generates nodes in a BN model out of E-health ontology, and allows users to easily establish links among nodes based on a meta model that represents cause-and-effect relationships among ontologies. In addition, to reduce the complexity and to improve the robustness of a BN model, we use an intermediate ontology that makes an application-specific abstraction of E-health ontology. We also present a diagnosis process using the E-health ontologies and constructed BN models. We tested our approach with an obesity management application. For diagnosing obesity, we set a scenario about service providing on the web portal, constructed E-health ontology, built BN model out of the ontology, and implemented an obesity-diagnosing system by adapting the diagnosis process. We also evaluate the effectiveness of our approach by comparing it with the methodologies used in related works of BN construction from ontology.