Artificial Intelligence-Driven Colonoscopy: A Systematic Review and Network Meta-Analysis on System Performance for...

Area of Science:

• Gastroenterology
• Medical Technology
• Oncology

Background:

• Colorectal cancer (CRC) poses a significant global health challenge, with adenomatous polyps as primary precursors.
• Standard colonoscopy, while effective, is limited by operator variability and potential for missed lesions.
• Artificial Intelligence (AI)-assisted colonoscopy is an emerging technology to enhance polyp detection during procedures.

Purpose of the Study:

• To systematically compare the performance of various AI-assisted colonoscopy systems.
• To evaluate the impact of different AI systems on adenoma detection rates (ADR) and adenomas per colonoscopy (APC).
• To assess the comparative effectiveness of AI systems against conventional colonoscopy in CRC screening.

Main Methods:

• A systematic review and Bayesian network meta-analysis of 48 randomized controlled trials (RCTs) involving 34,106 participants.
• Searched major databases (PubMed, Scopus, Google Scholar) up to November 4, 2025.
• Evaluated five commercial AI systems (EndoAngel, EndoAID, CAD-EYE, GI Genius, EndoScreener) and local platforms, focusing on ADR and APC.

Main Results:

• All evaluated AI systems demonstrated a significant improvement in ADR compared to conventional colonoscopy.
• EndoAngel showed the highest effect on ADR (OR 1.84), followed by EndoAID (OR 1.64).
• No AI system significantly improved the detection of high-risk lesions, with moderate evidence quality.

Conclusions:

• AI-assisted colonoscopy enhances adenoma detection, with EndoAngel and EndoAID showing the most substantial improvements.
• Performance variations among AI systems are modest, and the detection of high-risk lesions requires further investigation.
• Future head-to-head trials and cost-effectiveness analyses are crucial for optimizing AI implementation in CRC screening.