[Prognosis of the severity of acute pancreatitis by the neural network method]

Abstract

For prognozing the character of the development of acute pancreatitis a neural network of the direct spread of the signal was developed with 15 inlets, 5 neurons of the latent layer and two outlets. The network was trained with the help of an algorithm of an inverse spread of the error. The sensitivity of the produced prognostic model was 100%, specificity at admission--70%, on the first day--80%, on the second and third day--90%. Before the moment of hospitalization 57 patients were observed during 72 hours. The diagnosis was confirmed in all cases by the methods of computed tomography with contrast amplification.