Segmentation Of Mammographic Masses Based On The Gmac Model

Breast cancer is a kind of common malignancies which seriously endangers females’ health of mind and body. Early detection is a key to breast cancer that could reduce mortality and raise recovery rate. Breast tumors segmentation in the Computer-Aided Diagnosis (CAD) is the preparation step, which influences the following classification and detection directly.In this paper, the Gmac model is applied for the breast mass segmentation for the first time, meanwhile3improved algorithms based on Gmac model are proposed, including the improved variational level set method, the improved dual method and the improved split bregman method. The modified strategy contains rough selection and fine selection. The function fl is settled by the image information such as mean gray value, area and distance. Then the suspicious area is obtained by choosing the maximum value area of fl, where the gray values are higher than the lower threshold. The above step is named rough selection. After that, function f2is settled by the rough selection area. Then we can get a fine selection area using f2, where the gray value is higher than the higher threshold. Two standards are established to get the best results. They are area radio and distance sent between edge points farthest from the center and that nearest to the center.483images are selected. The CM mean values of the three algorithms are:64%,74%and76%and the AMED mean values of the three algorithms are:4.4750,1.6961and1.4602respectively. The experimental results show that split bregman method preforms best of the3algorithms. In this paper, we also use classical methods such as geodesic active contour model and active contour without edge model for breast tumors segmentation. The compared results illustrates that the global optimal solution could be received by using split bregman method.