COMPUTER-AIDED DETECTION OF COLONIC DIVERTICULAR DISEASE

Abstract

While CT colonography (CTC) is becoming a more prevalent and accepted method to diagnose colon cancer, the leading cause of nondiagnostic segmental evaluation of CT colonography is colonic diverticular disease (CDD). An essential element of detecting CDD in conjunction with CT colonography (CTC) is the accurate segmentation of the colonic wall. We have developed a level set based method to determine from a CTC scan the location of the outer wall of the colon, the serosal-tissue boundary. The algorithm then segments the entire outer colon wall at subvoxel precision and determines the thickness of the wall throughout the colon. Potential CDD detections are clustered based on the thickness of the colon wall and CT intensity values at the outer wall's position. Finally, a support-vector machine classifier is used to determine the location of diverticular disease. The algorithm has been validated on 10 CTC datasets, half of which have CDD present. At 100% sensitivity for the diverticular disease detections, the system has 0.2 false positives per patient.