Abstract Archives of the RSNA, 2005


Reader Performance Using CT Colonography Computer-assisted Reader Software: Benefit of Dedicated CT Colonography Training

Scientific Papers

Presented on November 29, 2005
Presented as part of SSG10: Gastrointestinal (CT Colonography: Computer-aided Diagnosis)


Stuart Andrew Taylor MD, Presenter: Nothing to Disclose
David Nigel Burling MBChB, Abstract Co-Author: Nothing to Disclose
Steve Halligan MD, Abstract Co-Author: Nothing to Disclose
Mary Roddie MD, Abstract Co-Author: Nothing to Disclose
Julian Tsang, Abstract Co-Author: Nothing to Disclose
Jamshid Dehmeshki, Abstract Co-Author: Nothing to Disclose


To investigate the additional benefit of dedicated CT colonography (CTC) training on reader performance when using computer assisted reader (CAR) software


6 readers were recruited from a separate study investigating the impact of CAR software on reader performance. As part of this study, readers (board certified without any prior CTC training) had read a dataset of 120 endoscopically proven CTC cases (50% abnormal), using CAR software (ColonCAR 1.2, Medicsight plc), The software highlights polyp candidates via circles superimposed on the 2D axial image during the primary read. Reporting times and diagnostic confidence (1-100) were recorded. No performance feedback was given. Two months later readers underwent a 1-day training course given by two CTC experts, consisting of five short lectures and five hours of workstation training. One week later, using CAR, readers re-read 20 abnormal datasets (randomly chosen from 120 case database), blinded to the case source or prevalence of abnormality. On statistical advice, pre and post training polyp detection, reading times, confidence levels and false positives were compared using the paired exact test, paired t test, Mann Whitney and Wilcoxen matched pairs test respectively


The 20 cases contained 55 polyps (twenty-two 1-5mm, twenty-four 6-9mm and nine 10mm+). CAR software in isolation highlighted 55%, 67% and 89% respectively. Four of the 6 readers significantly increased polyp detection after training. Pooled reader sensitivity increased by 18% (CI 14-24) p<0.001 overall and by 11% (CI 5-18) p=0.002, 26% (CI 18-34) p<0.001 and 19% (CI 6-31) p=0.01 for polyps 1-5mm, 6-9mm and 10mm+ respectively. Post-training pooled sensitivities were 23%, 51% and 74% respectively. Diagnostic confidence increased (median 70 IQ range 50-87.5 vs. median 80 IQ range 60-90), p=0.007. The total number of false positives was small but significantly increased after training (median 2.5 IQ range 1-5 vs. median 5.5 IQ 3.75-8), p=0.03. Mean reading time (minutes) increased from 11 (sd 3.1) to 16.5 (sd 3.4), p=0.01


One day of dedicated CTC training, whilst increasing reporting time, significantly improves reader performance when using CAR software


D.N.B.: advisor to Barco plc and Medicsight plcS.A.T.,S.H.: consultants to Medicsight plcM.R.,J.T.,J.D.,H.A.: Employees of Medicsight plcS.A.T.: Grant support from EZEMS.H.: Research agreement with Barco plc

Cite This Abstract

Taylor, S, Burling, D, Halligan, S, Roddie, M, Tsang, J, Dehmeshki, J, Reader Performance Using CT Colonography Computer-assisted Reader Software: Benefit of Dedicated CT Colonography Training.  Radiological Society of North America 2005 Scientific Assembly and Annual Meeting, November 27 - December 2, 2005 ,Chicago IL. http://archive.rsna.org/2005/4410328.html