An improved method of automatic colon segmentation for virtual colon unfolding

The technique of virtual colon unfolding (VU) is a non-invasive procedure to detect polyps on the colon inner wall. Compared with conventional virtual colonoscopy, VU is faster and results in fewer uninspected regions. However, the performance of VU is more vulnerable to the quality of colon segmentation. In this paper, an improved colon segmentation method is proposed to enhance the performance of VU. The improved method is with the use of a novel post-processing scheme, which is composed of two parts: attain more accurate centerlines with the help of scalar complementary geodesic distance field and compensate gap-like artifacts based on local morphological information. We validated the improved method on twenty colon cases via two widely used VU techniques, the ray-casting technique and the conformal-mapping technique. Experimental results indicated that with the use of the improved method, the rates of correct response via ray-casting and conformal-mapping techniques were respectively elevated by 14.9% and 13.1%, while the rates of false response were respectively reduced by 8.4% and 10.8%.