Impact of an automated system for endocytoscopic diagnosis of small colorectal lesions: an international web-based study

Background and study aims: Optical diagnosis of colorectal polyps is expected to improve the cost-effectiveness of colonoscopy, but achieving a high accuracy is difficult for trainees. Computer-aided diagnosis (CAD) is therefore receiving attention as an attractive tool. This study aimed to validate the efficacy of the latest CAD model for endocytoscopy (380-fold ultra-magnifying endoscopy). Patients and methods: This international web-based trial was conducted between August and November 2015. A web-based test comprising one white-light and one endocytoscopic image of 205 small colorectal polyps (<= 10mm) from 123 patients was undertaken by both CAD and by endoscopists (three experts and ten non-experts from three countries). Outcome measures were accuracy in identifying neoplastic change in diminutive (<= 5 mm) and small (<= 10 mm) polyps, and accuracy in predicting post-polypectomy surveillance intervals according to current guidelines for high confidence optical diagnoses of diminutive polyps. Results: Of the 205 small polyps (147 neoplastic and 58 non-neoplastic), 139 were diminutive. CAD was accurate for 89% (95% confidence interval [CI] 83%-94%) of diminutive polyps and 89% (84%-93%) of small polyps, which was significantly greater than results for the non-experts (73% [71%-76%], P<0.001; and 76% [74%-78%], P<0.001, respectively) and comparable with the experts' results (90% [87%-93%], P=0.703; and 91% [89%-93%], P=0.106, respectively). The surveillance interval predicted by CAD provided 98% (93%-100%) and 96% (91%-99%) agreement with pathology-directed intervals of the European and American guidelines, respectively. Conclusions: The use of CAD in endocytoscopy can be effective in the management of diminutive/small colorectal polyps.