Translator Disclaimer
9 May 2002 Computer-aided diagnosis in CT colonography: detection of polyps based on geometric and texture features
Author Affiliations +
A computer-aided diagnosis scheme for the detection of colonic polyps in CT colonography has been developed, and its performance has been assessed based on clinical cases with colonoscopy-confirmed polyps. In the scheme, the colon was automatically segmented by use of knowledge-guided segmentation from 3-dimensional isotropic volumes reconstructed from axial CT slices in CT colonography. Polyp candidates are detected by first computing of 3-dimensional geometric features that characterize polyps, and then segmenting of connected components corresponding to suspicious regions by hysteresis thresholding and fuzzy clustering based on these geometric features. False-positive detections are reduced by computation of 3-dimensional texture features characterizing the internal structures of the polyp candidates, followed by application of discriminant analysis to the feature space generated by the geometric and texture features. We applied our scheme to 43 CT colonographic cases with cleansed colon, including 12 polyps larger than 5 mm. In a by-dataset analysis, the CAD scheme yielded a sensitivity of 95% with 1.2 false positives per data set. The false negative was one of the two polyps in a single patient. Consequently, in by-patient analysis, our method yielded 100\% sensitivity with 2.0 false positives per patient. The results indicate that our CAD scheme has the potential to detect clinically important polyp cases with a high sensitivity and a relatively low false-positive rate.
© (2002) COPYRIGHT Society of Photo-Optical Instrumentation Engineers (SPIE). Downloading of the abstract is permitted for personal use only.
Hiroyuki Yoshida, Janne J. Naeppi, Hans Frimmel, and Abraham H. Dachman "Computer-aided diagnosis in CT colonography: detection of polyps based on geometric and texture features", Proc. SPIE 4684, Medical Imaging 2002: Image Processing, (9 May 2002);

Back to Top