Classification of mass type based on segmentation techniques with support vector machine model for diagnosis of breast cancer

Abstract

The use of technology in medical imaging is highly increased due to improved accuracy in radiologist's decisions. Computer Aided Diagnosis (CAD) tools helps radiologists to rule out the indirect symptoms which signs for false identification. Breast mass extraction from background is crucial step in processing of mammography. Hence, the proposed method primarily contemplates on three different segmentation techniques such as adaptive threshold based, modified watershed and energy based contour segmentation techniques and then relevant features extracted by Gray Level Covariance Matric (GLCM), Segmentation-based Fractal Texture Analysis (SFTA) and Shape features then passed to Support Vector Machine (SVM) classifier to classify mass type as benign or malignant. The experimental results show that the energy based contour segmentation techniques is more suitable for discriminating the mass type with highly promising results of accuracy, specificity and sensitivity as 98.26%, 100% and 96.83% respectively comparing to other techniques. The results of proposed methods experimented on MIAS dataset.