Computer-Aided Detection of Colorectal Polyps at CT Colonography

Introduction: Computer aided detection (CAD) of polyps is an emerging area in CT

colonography (CTC) research. We blindly evaluated two techniques for automated polyp detection using common test data.

Methods: The CTC images from five patients were randomly selected from our CTC database. The location of polyps in these patients was verified at conventional colonoscopy. Due the limited number of polyps available in the test set, we further augmented the datasets with computer-generated polyps.

All datasets were analysed using two techniques: surface normal analysis and surface curvature analysis. The output was displayed as a sequence of 2D slices, and the locations of the polyps flagged by computer detection were compared to known polyp locations.

Results: A total of 28 polyps ranging from 5mm-15mm were present, or inserted, in the datasets. Surface normal analysis detected 23 polyps and missed 5 giving a sensitivity of 82%. The five polyps missed were all 7mm or under. The total number of false positive detections was 152, approximately 30 per dataset.

Surface curvature analysis detected 24 polyps and missed 4 giving a sensitivity of 86%. The four polyps missed were all 5mm in size The total number of false positives using this approach was 160, 32

per dataset. Mean time taken for analysis was approximately 30 minutes.

Conclusion: Computer aided detection of polyps using morphological techniques has a high sensitivity. Further refinements in the technology are required to reduce the false positive rate.