Colorectal carcinoma is the second leading cause of cancer deaths in the United States today. In an effort to reduce mortality, Congress recently included a provision in the Balanced Budget Act of 1997 to support screening colonoscopy as a means for early detection and removal of colorectal polyps, the precursors to cancer. In this country alone, more than 68 million people are eligible for colorectal screening, but the majority are unlikely to comply with screening recommendations because of the costs, risks, discomfort, and inconvenience associated with traditional endoscopy. Furthermore, even if a small fraction of eligible persons are examined, the number of available gastroenterologists would be insufficient to perform so many procedures. We have developed a new technique, called virtual colonoscopy (VC), as an alternative to screening diagnostic colonoscopy (DC). The procedure consists of cleansing a patient's colon, inflating the colon with air, scanning the abdomen with helical computed tomography (CT), and generating a rapid sequence of three-dimensional (3D) images of the colon by means of virtual reality computer technology. Although VC makes possible the visualization of 3D images of the colon in a manner similar to that of DC, a correct diagnosis depends upon a physician's ability to identify small and sometimes subtle polyps within hundreds of 3D images. The absence of visual cues that normally occur with DC makes VC interpretation tedious and susceptible to error. With support from a National Science Foundation (NSF) grant, we have developed a computer-assisted polyp detection (CAPD) system that calculates areas of abnormal colon wall thickness in helical CT image data in order to highlight potential polyps in the 3D images. A physician ultimately determines if each detected lesion represents a true abnormality. Although we have found CAPD to be sensitive for finding subtle abnormalities, poor specificity can be attributed to several obstacles, including imprecise image segmentation, limited feature analysis, and suboptimal bowel preparation prior to helical CT scanning. With these challenges in mind, we propose research to perfect CAPD.
Our specific aims are as follows: 1. To develop an image segmentation algorithm that accurately isolates the colon from helical CT image data; 2. To improve our polyp detection algorithm with expanded feature analysis and artificial intelligence methods; 3. To optimize bowel preparation with digital subtraction of opacified feces and controlled gas distention; and 4. To validate the accuracy of VC, with the modifications achieved in the stated aims, by comparing the results of VC and DC in 200 patients undergoing usual-care colonoscopy. If VC with CAPD proves accurate and efficient in the diagnosis of colorectal polyps, it could evolve into a simple laboratory test, thereby meeting the demand for worldwide colorectal cancer screening.
