Rong Shi MD, Presenter: Nothing to Disclose

Pamela Schraedley-Desmond PhD, Abstract Co-Author: Nothing to Disclose

Sandy Napel PhD, Abstract Co-Author: Nothing to Disclose

Daniel Jason Aaron Margolis MD, Abstract Co-Author: Nothing to Disclose

Judy Yee MD, Abstract Co-Author: Nothing to Disclose

Christopher Frederick Beaulieu MD, Abstract Co-Author: Nothing to Disclose

Eric West Olcott MD, Abstract Co-Author: Nothing to Disclose

R. Brooke Jeffrey MD, Abstract Co-Author: Nothing to Disclose

Michael Ethan Zalis MD, Abstract Co-Author: Nothing to Disclose

David Seungwon Paik PhD, Abstract Co-Author: Nothing to Disclose

Anthony J. Sherbondy MS, Abstract Co-Author: Nothing to Disclose

Padmavathi Sundaram, Abstract Co-Author: Nothing to Disclose

et al, Abstract Co-Author: Nothing to Disclose

To determine if 3D viewing improves radiologists’ accuracy in classifying true positive (TP) and false positive (FP) polyp candidates identified by computer aided detection (CAD), and to determine candidate features that are associated with classification accuracy.

Four radiologists interpreted 47 multi-detector row CT colonography (CTC) datasets where polyp candidates were identified by CAD. Radiologists classified the polyp candidates using a five-point confidence scale, initially with 2D images only, then supplemented by 3D volume renderings. There were a total of 705 polyp candidates, comprising 53 TPs (according to optical colonoscopy) and 652 FPs. Another radiologist unblinded to the colonoscopy results characterized all polyp candidates (both TP and FP) for size, shape, colonic segment, attenuation, and surface smoothness, and assessed local colon distension and preparation, in addition to defining the true identity of FP candidates (stool, fold, ileocecal valve, etc.).

The addition of 3D viewing significantly improved classification accuracy for three of the four readers, increasing the area under a receiver operator characteristic (ROC) curve from 0.93-0.95 (2D) to 0.96-0.97 (2D3D; p < 0.001). For TP candidates, three features significantly affected reader accuracy: maximum polyp width (p = 0.038), polyp height (p = 0.019), and preparation (p = 0.004). For FP candidates, six features significantly affected accuracy: colonic segment (p = 0.007), attenuation (p < 0.001), surface smoothness (p < 0.001), distension (p = 0.034), preparation (p < 0.001) and the identity of the candidate lesion (p < 0.001).

3D viewing increases reader accuracy beyond 2D in classifying CAD identified polyp candidates in multi-detector row CTC. The main features of falsely rejected true lesions were: small and flat polyps, and having retained bowel contents in vicinity. Features of falsely accepted candidates were: located at cecum, heterogeneous in attenuation, irregular surface, inadequate bowel distension and preparation, as well as actually being retained stool or the ileocecal valve.

Shi, R, Schraedley-Desmond, P, Napel, S, Margolis, D, Yee, J, Beaulieu, C, Olcott, E, Jeffrey, R, Zalis, M, Paik, D, Sherbondy, A, Sundaram, P, et al, , Computer-aided Polyp Detection in CT Colonography: Influence of 3D Viewing and Features of Polyp Candidates on Interpretation.  Radiological Society of North America 2005 Scientific Assembly and Annual Meeting, November 27 - December 2, 2005 ,Chicago IL. http://archive.rsna.org/2005/4415269.html