Sistema Especialista Híbrido: Uma Aplicação para Diagnóstico de Múltiplas Doenças

Abstract

This work describes the implementation of a Hybrid Expert System (HES) applied to multiple disease diagnosis. The implementation process begins with the Knowledge Acquisition (KA), originated from a series of clinical parameters regarding each specific domain. In the proposed HES, the clinical parameters are used as input data for the Neural Network Based Expert System (NNES), hence, all obtained knowledge is converted to fuzzy rules. The NNES learning process and optimization is performed through the Genetic-Backpropagation Based Learning algorithm (GENBACK). The abstracted NNES knowledge, which is already refined, trained and tested, is used to form the Rule Based Expert System (RBES) knowledge base. The rule extraction algorithm is the Fuzzy Rule Extraction Algorithm (FUZZYRULEXT), and it is used to better explain the answer provided by the connectionist system output. In this context, the HES (v.1.0) has been used in epileptic crisis classification and also in breast cancer. For the first domain, it has presented a hit rate varying from 63,6% to 83,3%, and for the second, the partial tests have showed a variation of hits from 50% to 70%. At the moment, the HES (v.2.0) is also being tested to help defining the therapeutic conduct in coronary.