Establishment of Computer-Aided Diagnosis System for Liver Tumor CT Based on SVM

In this paper, we establish a computer-aided diagnosis (CAD) system that can distinguish normal liver regions and hepatocellular carcinoma (HCC) regions in the liver tumor spectral computed tomography (CT) images to help doctors diagnose and treat them. Methods: First-order statistical texture features and gray-level co-occurrence matrix (GLCM) texture features are extracted from the HCC and normal liver region of interest (ROI), respectively. We use the classification model base on support vector machine (SVM), according to the experimental results of different feature preprocessing methods and kernel functions, the best preprocessing method and SVM parameter are selected. Finally, we use principal component analysis (PCA) to reduce dimensions of the features and improve the classification accuracy, accelerate the classification model's learning speed, obtain the CAD system. The average classification accuracy of normal ROI in the training set reaches 92.22%, and the average classification accuracy of HCC ROI reaches 87.93%. The average classification accuracy of normal ROI in the testing set reaches 90.62%, and the average classification accuracy of HCC ROI reaches 86.36%. The experimental results have been recognized by professional physicians.