Multi-Phase CT Image Based Hepatic Lesion Diagnosis by SVM

In this paper, a novel liver lesion diagnosis approach based on multi-phase enhanced CT images is proposed. Regions of Interest (ROIs) which are drawn by an experienced radiologist are categorized into 4 classes: normal, cyst, haemangioma and hepatic cellular carcinoma. The diagnosis scheme includes 3 steps: feature extraction, feature selection and classification. For each ROI, 3 distinct kinds of features are extracted using First Order Statistics (FOS), Second Order Statistics (SGLCM), and Temporal Features, where 5 different feature sets are constructed respectively to be fed into a SVM-based classifier. Both classification accuracy statistics and Receiver Operating Characteristic (ROC) curve are employed to evaluate performance of different feature sets. Finally, a mixed feature set consisting of reduced FOS, SGLCM and temporal features gives the best classification accuracy of 0.955, 0.972 and 0.964 for normal-abnormal, cyst-otherdisease and carcinoma-haemangioma sub problems respectively.