Uma metodologia para construção de redes bayesianas com base em ontologias de domínio na área da saúde para suporte à decisão clínica

Abstract

Through the union of two approaches of Artificial Intelligence, Knowledge Representation via ontologies, and the treatment of incomplete information through the use of Bayesian Networks (BNs), this work aims to create a methodology for the construction of BNs based on ontologies of the health domain in order to provide clinical decision support. In order to evaluate the methodology, it was applied to an ontology of the Nephrology domain, more specifically of Chronic Kidney Disease (CKD) from which a BN was built. To obtain the probabilities of the Bayesian Network generated, it was used real clinical cases from a database of patients from the Lauro Wanderley University Hospital in the State of Paraíba. For that, it was used techniques such as: direct probability specification, marginal probability and conditional probability. In this way, the conditional probability table was constructed, for the obtained nodes of the BN. Given the results obtained with the experimental evaluation, where the methodology was applied, it was possible to observe the creation of new knowledge based on pre-existing knowledge. The generated network also enabled the extraction of probabilistic knowledge of an ontology by the use of BN, allowing the obtaining of knowledge not provided by the ontology, due to non-existence and/or uncertain information.