Uma aplicação de redes bayesianas no auxílio à tomada de decisões médicas.

Abstract

Medical Diagnosis belongs to a wide category of problems, where decision makingis accomplished considering the known evidences with different trust levels. Moreover, it is not rare the occurrence of different pathologies with common symptoms. The particularity of this scenario is present in Neurology, where rare pathologies with similar symptoms make differential diagnosis difficult and even imprecise. With the objective of reducing the degree of uncertainties involved, acquisition tools of new evidences are usually used, like clinical and neurological exams. However, these new evidences usually do not reduce the complexity of the emission of one differential diagnosis. Thus, i t is necessary to make use of computational tools that help decision making. In this work the performance of probabilistic inference in Bayesian Networks is evaluated as an aid to medical decisions. The performance of this technique is evaluated under databases with different scenarios. Moreover, new algorithms for reduction of the computational complexity of the probabilistic inference are considered, which use concepts of Information Theory. The results show that the probabilistic inference in Bayesian Networks can be promising as an aid to medical diagnosis.