Where technology meets technique: computer-aided detection and mucosal exposure device to improve adenoma detection

Article information

Clin Endosc. 2025;58(3):404-405

Publication date (electronic) : 2025 May 23

doi : https://doi.org/10.5946/ce.2025.079

James Weiquan Li^,1,2,3

¹Gastroenterology and Hepatology, Department of Medicine, Sengkang General Hospital, Singapore Health Services, Singapore, Singapore

²Academic Medicine Centre, Duke-NUS Medical School, Singapore, Singapore

³Yong Loo Lin School of Medicine, National University of Singapore, Singapore, Singapore

Correspondence: James Weiquan Li Gastroenterology and Hepatology, Department of Medicine, Sengkang General Hospital, Singapore Health Services, 110 Sengkang East Way, Singapore, Singapore E-mail: MDCLWJ@nus.edu.sg

Received 2025 March 15; Revised 2025 April 26; Accepted 2025 April 28.

Computer-aided detection (CADe) has raised expectations for improving adenoma detection rates (ADR) during colonoscopy, based on the consistent efficacy of randomized controlled trials.1 However, a recent meta-analysis failed to confirm the effectiveness of CADe in increasing ADR in a less controlled, non-randomized, real-world setting.2 This finding suggests the need for a more pragmatic study design to determine whether general implementation is advisable in hospital settings. Another limitation of CADe is its inability to detect polyps that are not visible on endoscopy.3 False negatives may also occur when polyps appear near the corners of the monitor or briefly enter and exit the field of view; for example, when the colonoscope tip brushes across a mucosal fold concealing a polyp.4 To address this, prospective randomized studies have evaluated the combination of CADe with the Endocuff (Olympus), a mucosal exposure device.5-7 These studies demonstrated that this combination could increase the ADR in controlled settings.

In this issue of Clinical Endoscopy, Kim et al.8 conducted a retrospective review of screening and surveillance colonoscopies performed between January 2022 and December 2022 by experienced endoscopists with high baseline ADRs at a single tertiary center. This study uniquely evaluated the combined use of CADe and the Endocuff for ADR in a non-randomized, real-world setting. The authors also categorized the cases into three groups—pre-artificial intelligence (AI), AI-assisted, and post-AI groups—for analysis, likely to examine any negative impact on the endoscopists’ ability to detect polyps after withdrawal of CADe. Although the study did not find significant differences in the primary outcomes of interest (ADR, adenoma per colonoscopy, sessile serrated lesions per colonoscopy and sessile serrated lesion detection rate [SSLDR]), there was an increase in the ADR for diminutive polyps. This finding is consistent with most published randomized controlled trials on CADe.1 Furthermore, a secondary analysis stratified by Endocuff use demonstrated significant increases in right-sided ADR (42.7% vs. 30.8%, p=0.03) and SSLDR (24.8% vs. 12.3%, p<0.01) with CADe. The efficacy of combining CADe with the Endocuff has been demonstrated in previous clinical trials. The current study’s non-randomized, real-world design attempted to bridge the efficacy-effectiveness gap discussed earlier in this editorial,1,2 although this was only evident in the secondary analysis. However, this study has some limitations. First, the retrospective nature of the study using historical controls introduced a significant selection bias. For example, the Endocuff usage rate differed substantially between the AI and pre-AI groups (65.7% and 37.8%, respectively). As only screening and surveillance colonoscopies were included, the participating endoscopists likely prioritized adenoma detection during these procedures. Endoscopists can decide on the use of CADe with or without an Endocuff at their discretion. The higher Endocuff usage in the AI group suggests the unintentional allocation of higher-risk patients (i.e., a higher likelihood of having polyps or having more polyps) or behavioral differences when performing colonoscopy if the endoscopists knew about the hypothesis before the CADe system was introduced for the trial period. Second, although 16 experienced endoscopists participated, some performed as few as two AI-assisted colonoscopies. Endoscopists with <20 AI-assisted cases were matched with at least 20 pre- and post-AI historical controls, whereas those with higher case volumes were matched equally across the pre-, with-, and post-AI groups. Retrospective studies with historical controls are more susceptible to selection bias than prospective observational studies with concurrent cohorts. This has been demonstrated in non-randomized CADe studies,9,10 which often report no differences in ADR. Lastly, the CADe system was tested for only two months in this center, limiting the study’s ability to assess lasting effects on endoscopists’ polyp detection performance.

The combined use of CADe and Endocuff has demonstrated synergistic efficacy in increasing ADR in randomized controlled trials. The present study takes a step toward evaluating the effectiveness of such a combination in routine clinical practice.