Application of neural network to the interpretation of laboratory data for the diagnosis of two forms of chronic active hepatitis

Abstract

The purpose of this study is to investigate the ability of neural network to discriminate the two subtypes, mild and severe form, of chronic active hepatitis, on the basis of the patterns of the five blood biochemical parameters. Serum levels of cholinesterase, albumin, alkaline phosphatase, type IV collagen and hyaluronate were used as variables. The neural network trained with the data from 31 patients: 11 of mild form and 20 of severe form of chronic active hepatitis. The ability of the network to predict the diagnosis of the patients who were additionally recruited was tested with a separate group (cross-validation group) of 9 patients with chronic active hepatitis. A neural network with 5 input neurons, 10 hidden neurons and 2 output neurons correctly classified all 31 patients. This network correctly predicted the diagnoses for 78% of the cross-validation group. These results suggested that neural network are useful for the differentiation of two forms of chronic active hepatitis by a less invasive blood biochemical analysis.