Effect of a deep-learning computer-aided detection system on adenoma detection during colonoscopy (CADe-DB trial): a double-blind randomised study

Detection and removal of adenomatous polyps during colonoscopy is the most effective way to decrease the incidence and mortality of colorectal cancer, 1–6 which is one of the leading causes of cancer-related death.7,8 The number of colorectal adenomas detected during colonoscopy, referred to as the adenoma detection rate (ADR), is regarded as the main quality indicator of colonoscopy. Patients benefit from endoscopists with a high ADR because with every 1·0% increase in ADR there is an associated 3·0% decrease in risk for interval colorectal cancer.9,10 In view of the importance of improving the ADR, research has focused on improving endoscopic hardware technology,11,12 bowel preparation methods,13,14 and observation strategies,15 with the aim to increase adenoma detection during colonoscopy. However, polyps can still be overlooked, with up to 27% of polyps missed for various reasons.16,17 Polyps that lie outside the visual field are one reason for not noticing adenomas, but unrecognised polyps within the visual field are equally important to address.18 

post1

Use of a high-performance computer-aided detection (CADe) system designed to detect polyps during colonos[1]copy19 has been shown in non-blinded trials20 to improve the ADR and the proportion of hyperplastic polyps detected in the colon (referred to as the polyp detection rate [PDR]) by identifying polyps that appear in the visual field but might be otherwise missed by the endoscopist. However, one major limitation of these non-blinded studies is operational bias; with the aid of the CADe system, operating endoscopists might either give more attention because of a competitive spirit or relax and rely on the CADe system.21 In both cases, this bias could overestimate or underestimate the effectiveness of the CADe system. Moreover, in the non-blinded studies, the exact contribution of the CADe system to an increase in ADR and PDR could not be pinpointed. Therefore, we aimed to do a double-blinded study using a sham control to more rigorously assess the effectiveness of a CADe system for improving detection of colon adenomas and polyps. We also aimed to analyse the characteristics of polyps missed by endoscopists.