Maruf Haider, Abstract Co-Author: Nothing to Disclose

Ronald M. Summers MD, PhD, Abstract Co-Author: Nothing to Disclose

Edward Mark Iuliano DO, Abstract Co-Author: Nothing to Disclose

Linda Brown MD, Abstract Co-Author: Nothing to Disclose

Nicholas Petrick PhD, Presenter: Nothing to Disclose

Perry Joseph Pickhardt MD, Abstract Co-Author: Nothing to Disclose

Jong-Ho Richard Choi MD, Abstract Co-Author: Nothing to Disclose

et al, Abstract Co-Author: Nothing to Disclose

To assess the added benefits of computer-aided detection (CAD) for radiologists interpreting CT colonography (CTC) images.

The study was a multi-reader multi-case receiver operating characteristic (ROC) experiment. 60 CTC cases were analyzed, with a 20 minute reading limit, using a commercially available CTC review system (Viatronix V3D Colon) by 3 board-certified radiologists. Reader 1 had the most experience with the CTC review system, Reader 2 had CTC experience but not with the review system and Reader 3 had no prior CTC experience. There were 20 training cases given to all readers to familiarize them with the workstation and ROC experiment. Two-dimensional (2D) axial views were used for initial polyp detection while a three-dimensional (3D) endoluminal view and multiplanar 2D views were used for problem solving. After an initial interpretation and detection without CAD were completed, the reader was shown CAD-identified polyp candidates. The reader was then allowed to modify their original detections and diagnosis based on the CAD marks. Polyp location, characteristics and reader confidence as to the likelihood of any identified structure being a polyp were recorded pre- and post-CAD. ROC curves, sensitivity and specificity for the reader alone, and reader+CAD were calculated using optical colonoscopy as a reference standard.

For polyps 1 cm and above, the sensitivity (Se) and specificity (Sp) for each reader alone were: Reader 1, 100% and 97%; Reader 2, 83% and 94%; Reader 3, 67% and 94%. Reader+CAD Se and Sp for polyps 1cm and above were: 100% and 97%; 100% and 91%; and 83% and 91% for readers 1, 2 and 3, respectively. The pre-CAD areas under the ROC curves (AUC) for Readers 1, 2 and 3 were 0.725, 0.642, and 0.625 respectively. The post-CAD AUC values improved to 0.770, 0.749, and 0.661 for the same readers using the CAD aid. The average time to read each case alone was 4.6±2.9 minutes. CAD added 2.8±4.4 minutes yielding an average total read time of 7.5±5.6 minutes per patient.

CAD increased each reader’s sensitivity and AUC so that less experienced readers with CAD performed similarly to more experienced readers without CAD.

J.R.C.: On the medical advisory board of Viatronix, Inc.P.J.P.: On the medical advisory boards of Viatronix, Inc. and Medicsight, Inc.R.M.S.,P.J.P.: Patents pending and/or awarded in the subject area of this abstract

Haider, M, Summers, R, Iuliano, E, Brown, L, Petrick, N, Pickhardt, P, Choi, J, et al, , An Observer Performance Study: Computed Tomographic Colonography and Computer-aided Detection as a Second Reader.  Radiological Society of North America 2005 Scientific Assembly and Annual Meeting, November 27 - December 2, 2005 ,Chicago IL. http://archive.rsna.org/2005/4406306.html