A Novel Dual LevelSets Competition Model for Colon Region Segmentation.
Huafeng Wang,Wenfeng Song,Lihong Li,Haixia Pan,Ming Ma,Weifeng Lv,Zhaohui Zhong,Zhengrong Liang
DOI: https://doi.org/10.1007/978-3-319-29965-5_12
2015-01-01
Abstract:To segment the colon region is of much significance for colonic polyp's detection in Computed Tomographic Colonoscopy CTC. However, not only the low contrast between CT attenuation values of the colon wall and the various surrounding tissues but also the pseudo enhancement effect by tag materials limit many traditional algorithms to achieve this task. Though few approaches suggested to depict colon walls by exploiting two steps: 1 find the inner colon wall; and 2 apply geodesic active contour based level set to extract outer boundary of colon wall, the failures happened when encounter the merging around houstral folds or adhesions of two very adjacent outer walls. Motivated by the observation that the interaction among 'forces' lead to a balance between the objects who caused those 'forces', we proposed a dual LevelSets competition model to simulate the mutual interference relationships among those compositions of the colon walls. Differ from the traditional LevelSet approach, the dual LevelSets competition model has a comprehensive cost function which take fully advantage of the essential characteristics of colon such as mixture, weak boundaries, volumetric, and so on. Compared with two already proved to be effective methods in literature: the graph cut and the geodesic active contour method, the proposed method has a much better performance to segment both the inner wall and the outer wall of colon. Both the comparison on if the method works well on weak boundaries of colon but also if it is capable of distinguishing the sticking boundaries of two very close walls is given. 200 CTC datasets are used to validate our proposed method. In conclusion, since the colon consists of various tissues, and they depend on and interact with each other, we could not consider the segmentation task in a static way, but a dynamic view works well.