An Interactive Segmentation Method Using Graph Cuts for Mammographic Masses

In this paper graph cuts method combined with watershed algorithm is applied for segmentation of breast masses in digitized mammograms. Graph Cuts have been applied widely for image segmentation but seldom on mammograms. In the proposed method, mammogram is partitioned with watershed algorithm into small and homogeneous regions. And a region adjacency graph is extracted from the watershed transform and then graph cuts method is applied for the following energy optimization step. The proposed approach can increase efficiency and precision of the mass segmentation, as well as decrease the computations. In the experiment based on USF-DDSM database, the obtained contours show good concordance with manual segmentation by the radiologists. And the final experimental results show higher accuracy and efficiency of the proposed segmentation method than those of GVF snake.