ARTIFICIAL NEURAL NETWORKS AND LIVER DISEASES: AN ECONOMIC AND PRE-IMAGING DIAGNOSIS

Background e Aims: We investigated if an Artificial Neural Network ANN) is able to identify hepatobiliary disease in selected patients affected with several, already diagnosed, hepatobiliary diseases, using only clinical and few laboratory findings, to provide a tool for early and "pre-imaging "(i.e. without using radiologic techniques) diagnosis of patients in real-life context. Methods: We used data from medical records of 270 patients affected with several hepatobiliary diseases. Patients were divided in three groups: G(train), (with clinical paradigmatic characteristics), to train network; G(train)., ("clinically similar" to those of G(trai)) to test the trained network; and, finally, Gvai significantly different from the above sets), to validate ANN diagnostic capabilities. Results: After training, the network provided right answer 96% of times, while in remaining 4% network outputs were only partly wrong. Comparing sets Girain and Giew we deduce that ANN is stable under minor modifications. Considering G(train), 1, right answer was given 80% of cases, while remaining results were, again, enough correct, an evidence of ANN 's stability under major modifications. Conclusions: our ANN works well for patients with known hepatobiliary diseases. Next step will be to use ANN for patients with suspected hepatobiliary diseases, and to extend our ANN to other human diseases.