AUTOMATIC COLON SEGMENTATION FOR 3D VIRTUAL COLONOSCOPY

点击翻译按钮获取中文摘要

A cost-effective, non-invasive, and patient-comfortable procedure for detecting colon polyps of a size less than 1 cm in diameter is extremely valuable for health care. Since there is a known progression of colonic polyps toward malignant carcinoma, detection and removal of small polyps can totally cure patients. Currently available diagnosis utilizes optical colonoscopy. Although it is accurate and can biopsy colonic polyps, optical colonoscopy has several drawbacks. It is expensive, time consuming, invasive, requires intravenous sedation and carries a small risk of perforation. Thus it is impractical for massive patient screening. To overcome these drawbacks, we have been developing an innovative technology, called three-dimensional (3D) Virtual Colonoscopy, and have demonstrated its effectiveness in imaging polyps as small as 3mm in diameter. In this new technology, a computed tomography (CT) scan of the patient's abdomen is taken and then a computer visualization system is used to virtually navigate within a 3D model of the colon searching for polyps. The goal of this project is to investigate the means for colon segmentation and polyp identification. It is necessary to differentiate residual stool from polyps for an accurate colon screening. The specific aims are: (1) To study contrast agents for a differentiated image contrast of stool from colon wall on the CT scans. An appropriate amount of barium mixed in meals will be given to the patients prior to the CT scans. The mixture of barium with stool will increase the image density of the stool. Most stool voxels will then be automatically labeled. (2) To develop computer algorithms to segment the residual-stool voxels. Statistical mixture models will be studied to characterize the voxel-density distribution. Markov random-field priors will be investigated to model the local structure of the residual-stool configuration. An optimal Bayesian inference will be used to compute the solution recognizing the residual- stool patterns. (3) To investigate visualization techniques for achieving interactive navigation and differentiating residual stools from polyps. Camera control and interactive surface and volume rendering will be studied to enable the physician to inspect the colonic surface and sub-surface tissue intuitively and interactively. With success of the specific aims, an accurate and cost-effective procedure for massive colon screening should be possible.