Hybrid Segmentation of Colon Boundaries in CT Images based on Geometric Deformable Model

One of the most important stages of virtual colonoscopy is colon segmentation, since an incorrect segmentation may lead to a misdiagnosis. Geometric deformable models (GDM) are found as efficient and attractive tools for structural based segmentation in particular to extract objects with complicated topology. There are two parameters influencing the overall performance of GDM algorithm; the distance between the initial contour and the actual object's contours and the stopping term which controls the deformation. To overcome these limitations, a two-stage based segmentation is utilized to extract the rough but precise initial contours at first. Then, extracted boundaries are smoothed and fined using a modified GDM algorithm by improving the stopping terms of the algorithm. The proposed method considerably removes false contours obtained during the first stage. The improvement of 6% on the accuracy of the segmentation method in comparison with region growing method is also achieved.