| The technique of virtual colon visualization represents a large number ofthree dimensional colon display and inspection methods. Virtual colonvisualization is a kind of noninvasive procedure developed for screeningcolorectal cancer. Colorectal tumors are believed to derive from colorectalpolyps, which might take a few years to grow up before becoming cancerous. Earlydetection and successful resection of the precursor polyps can effectivelyreduce the cancer-related mortality rates.The technique of virtual colon visualization involves a series of modules,including data acquisition, colon segmentation, colon centerline computationand colonic data visualization. In this dissertation, our work is concentratedon two of the complicated modules: colon segmentation and colonic datavisualization, so as to improve the performance of virtual colon visualization.Two problems impede reliable colon segmentation. The first is the use oforal contrast agents for bowel preparation. Oral contrast agents could make thoseissues of high CT values, such as bones and small intestines, be easily andmistakenly included into the segmentation results. The second is the applicationof insufflations when conducting CT scanning. Insufflations could causeoverdistention/underdistention to the colon. Overdistention might produce loopsin the segmented colon, while underdistention might lead to collapse in thesegmented colon. To correct the topological errors caused by the two problems,a fully automated colon segmentation method is proposed in this dissertation.The proposed method takes advantages of colonic topological information, aimingto fully automate the process from data acquisition to colon centerlinecomputation. The proposed segmentation method contains four novel algorithms:1) automated seed placement algorithm for region growing,2)automated anatomicallandmark location algorithm based on single-connected morphological closingoperation,3) colonic segments selection algorithm based on shortest path principle and4) loop and non-colonic tissue removal algorithm based oncomplementary geodesic distance map.170CT abdominal data are used to validatethe proposed method. Computer-generated centerlines are compared tohuman-generated centerlines. The proposed method yields a90.56%correctcoverage rate with respect to the human-generated centerlines.In the module of colonic data visualization, we focus on how to improve theperformance of the technique of virtual colon flattening. Virtual colonflattening technique is able to map the three dimensional colonic wall onto atwo dimensional surface, and thus can provide a global field of view of the entirecolonic wall on a single inspection window. Unfortunately, great distortionsof colonic wall might occur during the mapping process from three dimensionalspace to two dimensional surface, leading to missed polyps and false polypsdetection.Therefore, in this dissertation, a novel virtual colon flatteningmethod based on colonic outer surface is proposed. The proposed method is capableto adaptively correct distortions by using colonic outer surface instead ofcolonic inner surface.The proposed colonic outer surface based flattening methodis mainly composed of two parts:1) improve the algorithm for colonic outersurface extraction and2) develop a novel algorithm to correct the nonlinearityfor sampling planes based on colonic outer surface.We test the proposed methodon60colon cases which contain200annotated polyps in total. Visual inspectionsby three independent operators demonstrate that the proposed method can achieve88.0%correct dectection rate and the false detection rate is only0.16falsedetections per case.In this dissertation, we fulfill the fully automated colon segmentation andgreatly improve the performance of virtual colon flattening. We believe thatthe proposed methods in our work would be valuable to enhance the performanceof colonic polyp detection for virtual colon visualization system. |
PDF Full Text Request