Distinction of Liver Disease from CT Images Using Kernel-based Classifiers

Abstract

In this paper, a kernel-based classifier for liver disease distinction of computer tomography (CT) images is introduced. Three kinds of liver diseases are identified including cyst, hepatoma and cavernous hemangioma. The diagnosis scheme includes two steps: features extraction and classification. The features, derived from gray levels, co-occurrence matrix, and shape descriptors, are obtained from the region of interests (ROIs) among the normal and abnormal CT images. The sequential forward selection (SFS) algorithm selects the certain features for the specific diseases, and also reduces the features space for classification. In the classification phase, a 4-layer hierarchical scheme is adopted in the classifier. In the first layer, the classifier distinguishes the normal tissue from the abnormal tissues. The second layer classifier differentiates cyst from the other abnormal tissues. Cavernous hemangioma is identified in the third layer, while hepatoma is recognized from the undefined tissues in the last layer. Finally, we use the receiver operating characteristic (ROC) curve to evaluate the performance of the diagnosis system.