Automatic and Robust Vessel Segmentation in CT Volumes Using Submodular Constrained Graph

Graph cut is one of segmentation method that can give us the good result on natural image and large organ segmentation of medical image. However, we cannot get the correct or accurate results by using graph cut on detailed structures such as, tree branch, or blood vessel because the property of smoothness in graph cuts energy function will completely remove the small branch of the detailed structure to minimize its cost. We propose the vessel extraction method which combine graph cut and concept of submodular function. The conventional graph cuts will be use to obtain initial segmentation while graph cut with submodular function will be use to refine the initial segmentation. Submodular function can solve the problem of smoothness of graph cut in detail structure as shown in result that less segment and more united vessel tree than conventional graph cuts. The experimental result shows that our method can segment blood vessels of liver with higher accuracy while graph cut lead to a lot of loss of the detail branches in the liver vessel. With submodular constraint, we can connect the segment branch of vessel into united vessel tree which conventional graph cut still remain the segment of vessel’s branches.