Classification of focal liver disease in egyptian patients using ultrasound images and convolutional neural networks

Abstract

Recently, computer-aided diagnostic systems for various diseases have received great attention. One of the latest technologies used is deep learning architectures for analyzing and classifying medical images. In this paper, a new system that uses deep learning to classify three focal diseases in the liver besides the normal liver is proposed. A pre-trained convolutional neural network is utilized. Two types of networks are used, ResNet50 and AlexNet with fully connected networks (FCNs). After extracting the deep features using deep learning, FCNs can input images in different states of the disease, such as Normal, Hem, HCC, and Cyst. Dataset is obtained from the Egyptian Liver Research Institute. Two classifiers are utilized, the first includes two classes (Normal/Cyst, Normal/Hem, Normal/HCC, HCC/Cyst, HCC/Hem, Cyst/Hem) and the second contains four classes (Normal/Cyst/ HCC/Hem) to distinguish liver images. Using performance criteria, it has been shown that the two-category classifiers have given better results than the four-class classifier, and accordingly a hybrid classifier was suggested to merge the weighted probabilities of the classes obtained by each singular classifier. Experimental results have achieved an accuracy of 96.1% using ResNet50 which means that it can be used as an assistive diagnostic method for classification of focal liver disease.