Delia Campanella MD, Presenter: Consultant, im3D SpA, Turin, Italy

Loredana Correale PhD, Abstract Co-Author: Researcher, I-m3d S.p.a., Turin, Italy

Maria Carla Cassinis MD, Abstract Co-Author: Nothing to Disclose

Cristina Martina MD, Abstract Co-Author: Nothing to Disclose

Vincenzo Tartaglia MD, Abstract Co-Author: Nothing to Disclose

Daniele Regge MD, Abstract Co-Author: Research Consultant, im3D SpA

Teresa Maria Gallo MD, Abstract Co-Author: Nothing to Disclose

et al, Abstract Co-Author: Nothing to Disclose

et al, Abstract Co-Author: Nothing to Disclose

To assess if the number of FP candidates prompted by CAD affects the reader’s performance.

100 CTC were obtained in 50 patients (30 with at least one polyp ≥5mm) that performed CTC following cathartic preparation. We simulated two CAD systems that prompted for all true positive (TP) lesions, producing however different number of FP prompts. CAD1 included all TP lesions in a list of 5 prompts/dataset, while CAD2 in a list of 10 prompts/dataset. Therefore using CAD1 as first reader, 5 experienced (>300 validated cases) radiologists classified 70 TP and 430 FP candidates. Datasets were reread after 6 weeks using CAD2 (70 TP, 930 FP candidates) and the same reading paradigm. In both experimental runs reading times, TP and FP findings per dataset were recorded for each radiologist.

Mean reading times for CAD1 and CAD2 were 4min:34sec±12sec (error of the mean) and 4min:13sec±10sec respectively and they can be thus considered as statistically compatible. Mean per-lesion sensitivity values were respectively 0.74±0.04 and 0.79±0.04 for the 1st and 2nd experimental run respectively (not significant different). Mean FP findings per dataset recorded by readers using CAD1 and CAD2 were respectively 0.21±0.02 and 0.27±0.03 indicating for values compatible within error bars. Average per-patient sensitivity and specificity was respectively 0.78 and 0.80 for CAD1 and 0.80 and 0.74 for CAD2. Using the two CADs, all radiologists maintained statistically compatible values of their diagnostic performance indices.

On the basis of these preliminary results we conclude that reducing the number of FP CAD prompts from 10 to 5 does not improve per-lesion and per-patient sensitivity, if CTC is read by experienced radiologists. There was no statistical difference in FP findings and per-patient specificity. Reading times were statistically similar. Further studies will be necessary to confirm these results and to investigate the dependence of false negative and FP findings on colon preparation.

This study suggests that the number of erroneous CAD prompts, within certain limits, does not seem to impair the experienced radiologists’ performance.

Campanella, D, Correale, L, Cassinis, M, Martina, C, Tartaglia, V, Regge, D, Gallo, T, et al, , et al, , CT Colonography (CTC) Using Different Theoretical Computer-aided Detection (CAD) Software: Influence of False Positive (FP) Prompts Number on Experienced Readers’ Performance: Preliminary Results.  Radiological Society of North America 2007 Scientific Assembly and Annual Meeting, November 25 - November 30, 2007 ,Chicago IL. http://archive.rsna.org/2007/5003869.html