Segmentation of Volume Data Based on 3D Lazy Snapping

Abstract — The image segmentation techniques for medical image play important roles in computer-aided detection, analysis and measurement. Due to the computed tomography (CT) and the magnetic resonance imaging (MRI) technologies, the doctor can obtain volume images to visualize the structures of internal organ of human body. In this paper, we present a volume segmentation method by extending the Lazy Snapping [1] to 3D. A high-level painting tool is provided for the user to indicate which part is object and which is background in volume by marking some lines. The segmentation result can be determined with little interaction without voxel-precise selection. In order to improve the performance, we adopt the watershed used in [1] to pre-process the volume data. And we implement the 3D watershed to handle the volume data. However, large size of volume data still needs great computation. It is hard to achieve a real time response. To improve the efficiency without losing the accuracy, we introduce a volume of interest (VOI) specified by the user to restrict the data size for processing. Our system also implements a user-friendly interface to browse the volume of interest area in three cutting planes. I