Automatic 3D Prostate MR Image Segmentation Using Graph Cuts and Level Sets with Shape Prior

Wei Xiong, Anthony Lianjie Li, Sim Heng Ong, Ying Sun

Abstract: Automatic segmentation for 3D magnetic resonance images of the prostate is a challenging task due to its varying shapes and sizes. Most recent techniques are focused on using variations of the Active Appearance Model (AAM) approach as the main segmentation method. In this paper, an alternative approach using a hybrid of the graph cut technique and the geodesic active contour shape prior level set method is presented. Despite being relatively accurate, level set methods are not commonly used for 3D segmentation purposes because they are computationally expensive. This paper shows that, with 3D graph cut results as initialization for level sets, the processing time for such level set based methods can be substantially reduced while preserving the accuracy of the segmentation.