We will analyze CT colonography (virtual colonoscopy) data using computer-assisted diagnosis methods. These methods attempt to identify and characterize colonic polyps automatically, thereby improving physician accuracy and efficiency. We will compare the results of the computer analyses with the ground truth data (conventional colonoscopy, pathologic analysis). We made a number of advances over the past year, including advances in colon and polyp segmentation, feature extraction methods, false positive reduction and classifier optimization. We just had accepted for publication a major trial of computer-aided polyp detection (Gastroenterology, accepted for publication). This trial showed that computer-aided polyp detection is robust, reproducible and accurate for polyps 8 mm and larger. We also made a clinical observation that oral contrast sticks to some polyps, especially those with villous histology. This finding has implications for radiologic interpretation.
Showing the most recent 10 out of 37 publications
