Quantitative tissue characterization of diffuse liver diseases from ultrasound images by neural network

Abstract

The aim of the study is to establish a computer-aided diagnosis system for diffuse liver diseases such as chronic active hepatitis (CAH) and liver cirrhosis (LC). The authors introduced an artificial neural network in the classification of these diseases. In this system the neural network was trained by feature parameters extracted from B-mode ultrasonic images of normal liver (NL), CAH and LC. Therefore one need not input any a priori information about these diseases. For input data the authors used 7 parameters calculated by 5 regions of interest (ROIs) in each image. They are variance of pixel values in an ROI, coefficient of variation, annular Fourier power spectrum, longitudinal Fourier power spectrum, and variation of the means of the 5 ROIs. In addition, the authors used 2 more parameters calculated from a co-occurrence matrix of pixel values in an ROI. The results showed that the accuracies of the neural network were 83.8% for LC, 90.0% for CAH and 93.6% for NL, and that the system was considered to be helpful for clinical and educational use.