Feature Extraction for the Prediction of Liver Fibrosis Stages in Chronic Hepatitis C

Many conventional studies have predicted the liver fibrosis stages in chronic hepatitis C at a certain time using the clinical test results on blood and urine obtained at the same time as inputs to classifiers. However, given the mechanism of liver fibrosis progress through time-varying inflammation, it is considered to be more effective to use the time series of test results obtained from the past to the present. This study aims to extract features on the dynamics of such time series and to examine how they are effective for the prediction of the stages of liver fibrosis. We propose the combination of the mean, standard deviation, and linear predictive coding cepstrum as a feature and conduct experiments on the prediction performance of this proposed feature and competitive ones, such as the waveform and other types of spectrum estimators. The experimental results suggest the effectiveness of the proposed feature, since it consistently achieves a performance that is better than those of the other features.