Neste trabalho descreve-se um sistema de apoio ao diagnóstico médico que baseia-se na regra de Bayes e apresenta as seguintes caracteristicas : elevada precisão de diagnóstico, eficiência computacional, generalidade para aplicação em diferentes dominios da área médica, facilidade de uso, capacidade de trabalhar com dados parciais, suporte a decisões de modo iterativo. A extração do conhecimento é obtida diretamente de casos concretos de pacientes armazenados em um banco de dados. Além disso, mostra-se a aplicação do sistema na área de doença coronariana.

We show an expert system with the objective of aiding physicians in decisions concerning differential diagnosis, test planning, prognosis and therapy selection. This system acquires its knowledge directly from concrete patient cases stored in its database. Aiming to extract the maximum information from patients data, we investigale a Bayes' rule application, considering the dependences and interactions between symptoms. A technique to form clusters of interdependent patient variables was developed. the clusters are treated as mutually independent complex variables. lhe applicalion of this approach to coronary artery diseases diagnosis allowed significant gains in classification performance, when compared with bayesian models using the conditional independence assumption. Compulationally efficient algorithms for training and inference were developed based on a sparse representalion of joint condilional probabilities. The work and memory space of these algorithms depend on the training data set size, being not affected, in practical situalions, by the exponential growth of lhe number of parameters experienced when weaker interactions between symptoms are considered. Moreover, we implemented an algorithm for creation of medical invesligation plans based on heurislic programming techniques, capable to select the best sequence of tests to a specific palient. Finally, a shell was developed to permit an easy and flexible implementation of decision support syslems for different medical diagnosis problem domains.