Reader error during CT colonography: causes and implications for training

This study investigated the variability in baseline computed tomography colonography (CTC) performance using untrained readers by documenting sources of error to guide future training requirements. Twenty CTC endoscopically validated data sets containing 32 polyps were consensus read by three unblinded radiologists experienced in CTC, creating a reference standard. Six readers without prior CTC training [four residents and two board-certified subspecialty gastrointestinal (GI) radiologists] read the 20 cases. Readers drew a region of interest (ROI) around every area they considered a potential colonic lesion, even if subsequently dismissed, before creating a final report. Using this final report, reader ROIs were classified as true positive detections, true negatives correctly dismissed, true detections incorrectly dismissed (i.e., classification error), or perceptual errors. Detection of polyps 1-5 mm, 6-9 mm, and ≥10 mm ranged from 7.1% to 28.6%, 16.7% to 41.7%, and 16.7% to 83.3%, respectively. There was no significant difference between polyp detection or false positives for the GI radiologists compared with residents (p=0.67, p=0.4 respectively). Most missed polyps were due to failure of detection rather than characterization (range 82-95%). Untrained reader performance is variable but generally poor. Most missed polyps are due perceptual error rather than characterization, suggesting basic training should focus heavily on lesion detection. (orig.)