Prediction of cirrhosis in patients with chronic hepatitis C infection by artificial neural network analysis of virus and clinical factors.

The diagnosis of cirrhosis in patients with hepatitis C virus (HCV) infection is currently made using a liver biopsy. In this study we have trained and validated artificial neural networks (ANN) with routine clinical host and viral parameters to predict the presence or absence of cirrhosis in patients with chronic HCV infection and assessed and interpreted the role of the different inputs on the ANN classification. Fifteen routine clinical and virological factors were collated from 112 patients who were HCV RNA positive by reverse transcriptase-polymerase chain reaction (RT-PCR). Standard and Ward-type feed-forward fully-connected ANN analyses were carried out both by training the networks with data from 82 patients and subsequently testing with data from 30 patients plus performing leave-one-out tests for the whole patient data set. The ANN results were also compared with those from multiple logistic regression. The performance of both ANN methods was superior compared with the logistic regression. The best performance was obtained with the Ward-type ANNs resulting in a sensitivity of 92% and a specificity of 98.9% together with a predictive value of a positive test of 95% and a predictive value of a negative test of 97% in the leave-one-out test. Hence, further validation of the ANN analysis is likely to provide a non-invasive test for diagnosing cirrhosis in HCV-infected patients.