Classification of Hepatic Tissues from CT Images Based on Texture Features and Multiclass Support Vector Machines

A computer-aided diagnosis (CAD) of X-ray Computed Tomography (CT) liver images with contrast agent injection is presented. Regions of interests (ROIs) on CT liver images are defined by experienced radiologists. For each ROI, texture features based on first order statistics (FOS), spatial gray level dependence matrix (SGLDM), gray level run length matrix (GLRLM) and gray level difference matrix (GLDM) are extracted. Support vector machine (SVM) is originally for binary classification. In order to classify hepatic tissues from CT images into primary hepatic carcinoma, hemangioma and normal liver, we utilize two methods to construct multiclass SVMs: one-against-all (OAA), one-against-one (OAO) and compare their performance. The result shows that a total accuracy rate of 97.78% is obtained with the multiclass SVM using the OAO method. Our study has some practical significance for clinical diagnosis.