Intelligent Diagnosis of Liver Diseases from Ultrasonic Liver Images: Neural Network Approach

Abstract

The main objective of this study is to develop an optimal neural network based DSS, which is aimed at precise and reliable diagnosis of chronic active hepatitis (CAH) and cirrhosis (CRH). Multilayer perceptron (MLP) neural network is designed scrupulously for classification of these diseases. The neural network is trained by eight quantified texture features, which were extracted from five different region of interests (ROIs) uniformly distributed in each B-mode ultrasonic image of normal liver (NL), CAH and CRH. The proposed MLP NN classifier is the most efficient learning machine that is able to classify all three cases of diffused liver with average classification accuracy of 96.55%; 6 cases of cirrhosis out of 7 (6/7), all 7 cases of chronic active hepatitis (7/7) and all 15 cases of normal liver (15/15). The advantage of proposed MLP NN based Decision Support System (DSS) is its hardware compactness and computational simplicity.