Classification of mammographic images via computer aided diagnosis system

Breast cancer which is one of the most common cancers causing death, particularly among women, constitutes %23 of all types of cancerous cases among women. There is no way to prevent breast cancer yet. To fight against breast cancer, it is crucial to detect it in an early stage. Today, some researches on automated recognition systems which will help radiologists to scan mammograms are being done. These automated systems are called as ?Computer Aided Diagnosis Systems" (CAD). CAD software is used to ensure a second look on mammograms to assist radiologists using mammogram. There is no question that the ultimate determination is made by radiologists.In this study, a hybrid CAD system which consists of Curvelet Transform (CT), Wave Atom Transform (WAT), Principle Component Analysis (PCA), Linear Discriminant Analysis (LDA), Support Vector Machines (SVM), k-Nearest Neighbor (k-NN), Least Squares Support Vector Machines (LS-SVM) are presented. In the generated systems, firstly, suspicious areas in mammograms are determined by using top-hat transform, bottom-hat transform and average filter automatically and sets of sub-images are created. Following this process, feature extraction and classification operation are applied to the data set obtained from sub-images. Classification is performed in two stages as abnormal-normal of all the mammogram images and benign-malignant of the separated abnormal images. CT and WAT are used for feature extraction and SVM, k-NN and LS-SVM are applied for classification comparatively. LDA and PCA are used for feature selection. Successful classification results have been achieved at %100.