CT Colonography: Role of a second reader CAD paradigm in the initial training of radiologists

Purpose: To evaluate the influence of CAD for the evaluation of CT colonography (CTC) datasets by inexperienced readers during the attendance of a dedicated hands-on training course. Method and materials: Twenty-seven radiologists inexperienced in CTC (11 with no CTC training at all, 16 having previously reviewed no more than 10 CTC cases overall) attended a hands-on training course based on direct teaching on fifteen workstations (four Advantage Windows 4.4 with Colon VCAR software, GE; six CADCOLON, Im3D; five ColonScreen (Toshiba/Voxar) with ColonCAD™ API, Medicsight). During the course, readers were instructed to analyze 26 CTC cases including 38 colonic lesions obtained through low-dose MDCT acquisitions, consisting of 12 polyps sized less than 6 mm, 9 polyps sized between 6 and 10 mm, 12 polyps sized between 11 mm and 30 mm, and 5 colonic masses sized >3 cm. CTC images were reviewed by each reader both in 2D and 3D mode, respectively by direct evaluation of native axial images and MPR reconstructions, and virtual endoscopy or dissected views. Each reader had 15 min time for assessing each dataset without CAD, after which results were compared with those provided by CAD software. Global rater sensitivity for each lesion size before and after CAD usage was compared by means of two-tailed Student's t test, while sensitivity of each single reader before and after CAD usage was assessed with the McNemar test. Results: For lesions sized 30 mm, sensitivity before CAD-assisted reading was 0.3556 ± 0.3105 and did not change after CAD usage (p = 1). Sensitivity of each single rater did not significantly differ before and after CAD for any lesion size category (McNemar test, p > 0.05). Specificity was not significantly different before and after CAD for any lesion size (>96% for all size categories). Conclusion: CAD usage led to increased overall sensitivity of inexperienced readers for all polyps sizes, except for lesions >30 mm, but sensitivity of individual raters was not significantly higher compared with CAD-unassisted reading.