Classification of benign and malignant breast cancer based on DWI texture features

Abstract

In this paper, the purpose was to develop a computer-aided diagnosis (CADs) system that use diffusion-weighted Magnetic resonance image (DW-MRI) for distinguishing benign and malignant breast tumors. Sixty-one cases were collected, including 23 patients with benign tumors, and 38 patients with malignant tumors. Two types of texture features were obtained from each lesion, including 6 histogram statistical features and 16 gray-level co-occurrence matrix (GLCM) features. The feature selection was based on Random Forest-Recursive Feature Elimination (RF-RFE) method. Random Forest was utiliezed to build the classification model, and the classifier performance were evaluated based on area under the receiver operating characteristic curve (AUC), and using leave-one-out cross validation(LOOCV). 6 texture features (including 3 histogram statistical features and 3 GLCM features) are selected via this approach, an AUC of 0.76 was obtained, and the classification accuracy, sensitivity, and specificity were show to be 77.05%, 84.21%, 65.21%, respectively. The results suggest that the texture features can be used for developing CADs of breast cancer, and show high sensitivity.