Segmentation of Mammography Images Based on Spectrum Clustering Method

Mass segmentation in mammography images is one of the effective ways to screen breast cancer. The accurate segmentation of the pectoral muscle can improve the accuracy of mass recognition. However, the results of traditional mammography image segmentation methods often appear incomplete segmentation and over-segmentation, the accuracy is low, which directly affects the accuracy of breast cancer screening. To solve these problems, a segmentation method of mammography images based on spectral clustering is proposed in this paper. Firstly, we use the spectral clustering to segment the pectoral muscle preliminarily. In view of the stratification of pectoral muscle and the unclear boundary of breast muscle and breast tissue, we use the maximum grayscale difference constraint and shape constraint to achieve accurate breast muscle segmentation. The mass is recognized accurately with the segmented image. The experimental results of the MIAS breast image database show that the proposed method can effectively segment the uneven grayscale pectoral muscle caused by the overlap of the pectoral muscle tissues, and it is robust to the segmentation of tumors of different sizes.