We also have X and podcasts
AI-Assisted Polyp Detection Shows No Advantage in Real-World Practice
A large randomized controlled trial found that real-time AI-assisted colonoscopy did not significantly improve adenoma detection rates compared to traditional colonoscopy, challenging assumptions about computer-aided detection systems in routine clinical practice.
Background
Computer-aided detection (CAD) systems have been proposed to enhance polyp detection during colonoscopy. This multicenter randomized controlled trial evaluated EndoMind, a real-time polyp detection system, in a practical setting across five outpatient centers in private practice from November 2021 to November 2022.
Study Design
The trial enrolled 914 patients undergoing screening or surveillance colonoscopy, randomized to either CAD-assisted colonoscopy (452 patients) or traditional colonoscopy (462 patients). Over 94% of procedures were for screening or post-polypectomy surveillance. All 10 examiners had more than 10 years of experience.
Key Findings
The adenoma detection rate (ADR) was not significantly different between groups: CAD-assisted colonoscopy achieved 34.5% compared to 32.9% in traditional colonoscopy (p=0.656). This lack of significant improvement occurs despite EndoMind’s real-time detection capability, suggesting the effect of computer-aided colonoscopy on adenoma detection remains controversial.
Why It Matters
The results underscore that diverging study designs and patient populations complicate consistent evaluation of CAD systems across trials. The authors advocate for large-scale real-world studies to clarify the practical clinical value of computer-assisted detection in routine colonoscopy.
Limitations
Study reproducibility was rated as low, and findings may be specific to the private practice setting and experienced endoscopists evaluated.
Original paper: Artificial intelligence assisted colorectal lesion detection in private practices a randomized controlled study. — NPJ digital medicine (2026).
