An automatic method for colon segmentation in virtual colonoscopy

Colon segmentation from computed tomography (CT) data is a requirement for the construction of three-dimensional (3D) colon in the virtual colonoscopy (VC) system. There are two main challenges in the colon segmentation: 1) many non-colonic objects might attach to the colon, 2) under-distension of the colon might lead to numbers of collapsed portions. In this paper, we proposed a method that can remove non-colonic attachments and connect collapsed colonic segments together for the VC examination. The proposed method is mainly composed by a non-colonic attachment classification algorithm and a heuristic connection algorithm. We use 30 CT scans to evaluate our method. Computer-generated colons are compared with human-generated colons which are manually extracted by two radiologists. The proposed method could achieve 92.86% coverage of human-generated colons, which is of 13.68% higher than the conventional method.