Speaker: Robert Van Uitert

Title: CT Virtual Colonoscopy Computer-Aided Polyp Detection at the NIH

Abstract:

With the increase in the number of slices and resolution of images acquired by medical imaging machines, the use of computers for diagnostic purposes has become increasingly important as an aid for radiologists. Computer-aided detection (CAD) alleviates some of the interpretation burden placed on radiologists by highlighting specific regions of the acquired images on which the physician may want to concentrate. CAD has been shown to allow for improved consistency and sensitivity in diagnosis of medical scans.

One area where CAD might be of significant benefit is in the detection of colorectal cancer, which is the second leading cause of cancer related deaths in the U.S. With proper screening, colorectal cancer can be prevented. Unfortunately, many patients do not undergo screening due to perceived inconvenience and discomfort of existing screening tests. Virtual colonoscopy, a CT scan-based imaging method, shows promise as a method of colorectal cancer screening that may be acceptable to many patients. Recent studies have suggested that virtual colonoscopy may have a high sensitivity and specificity for polyp detection.

In this presentation, we will discuss the CT virtual colonoscopy computer-aided detection system developed at the National Institutes of Health. This system automatically segments the colon from a abdominal CT scan, determines potential problematic regions of the colon surface, calculates features of these regions, and then uses a support vector machine to decide if each region is a polyp and should be presented to the radiologist for further diagnosis. This system was validated on over 1000 patients from 3 medical centers and was shown to have a sensitivity of 89.3% for detecting adenomatous polyps at least 1cm in size with a false-positive rate of 2.1 false polyps per patient.