Texture features are useful in analyzing the hepatic fibrosis on CT images, however properly selecting features and slice thickness is still uncertain. In this paper, five types of slice thickness and 15 features extracted from co-occurrence matrix are investigated to select the optimal parameters. Each combination will be checked by using SVM (Support Vector machine) with leave-one-case-out method. 149 cases including 6 grades of hepatic fibrosis are acquired by CT scanner and divided into two groups: normal & mild fibrosis vs severe fibrosis & typical cirrhosis. Iteration test on all of the subsets indicates that 5 to 7 features with slice thickness of 1.25mm is the optimal combination with relative higher accuracy in classification of fibrosis.
展开▼
