MOCABBAN : a modeling case base by a bayesian network applied to the diagnosis of hepatic pathologies

Our work presents a case based reasoning system for the diagnosis of hepatic pathologies according to a Bayesian model. The main idea consists a modelling the case base by a Bayesian network. Bayesian networks are excellent tools for the modelling of the uncertain in terms of their clear graphic representation as well as the conditional probabilities laws defined on this graph. Our network allows a representation of a qualitative and causal knowledge in addition to a quantitative knowledge which expresses the uncertainty. It contains four levels: clinical, biologic, medical imaging and diagnosis-therapy. Each level has a table containing the different conditional probabilities in order to get a good final diagnosis. Each level is composed of a set of attributes; each attribute corresponds to a node of a network. The arcs describe the relations between these attributes as conditional probabilities of attributes in the case. We have used the exact algorithm of inference JLO-Jensen, Lauritzen et Olesen - to calculate the conditional probabilities. When we have a new case, it is inserted in the network and is propagated to the network. When some signs are missed, we have used another algorithm of learning EM-expectation maximisation to estimate the conditional probabilities of the missed variables in order to get a good learning. We have chosen this modelling in order to obtain a good final diagnosis