Breast Cancer Diagnosis by Weighted Blocking Threshold Segmentation and Fuzzy SVM Classifier in Thermography Images

Abstract

Breast cancer is the most common cancer among women and the second most common cancer among cancer patients around the world. Approximately 70% of patients could be diagnosed through screening and early imaging when no tangible mass exists in patient's body. In this step overcoming the disease is much easier and also much less cost. To provide a noninvasive method for diagnosis of this type of cancer, researchers turned to infrared images processing methods anew, though mostly focused on the segmentation and ROI detection step. In this paper, a method for early detection of breast cancer in thermal images is proposed. In the proposed method, after blocking and segmentation of the images toward ROI detection, both feature extraction and dimension reduction with manifold learning are performed. Finally data features matrix is classified with the Fuzzy Support Vector Machine technique into the three classes: Healthy, benign, and malignant. Then, the areas suspected to cancer are determined. After feature extraction, Fuzzy Kernel Isomap algorithm was applied to reduce the size of the feature vectors. After dimension reduction, the SVM algorithm separated the normal and abnormal classes. Then based on a fuzzy function; areas which have not conformity with the other side of breast were extracted. Best accuracy of the proposed method measured on the 50 thermography photos is 74.12%. Based on our methodology, a single stand-alone computer can be usedfor screening as computer aided thermography approach. Moreover, it would be applied for initial checkup in the rural and remote areas with fewer treatment possibilities.