Back to 2025 Abstracts
PROS AND CONS OF ARTIFICIAL INTELLIGENCE IMPLEMENTATION DURING COLONOSCOPY TO ENHANCE ADENOMA DETECTION RATES.
maryam aleissa
*1,2, Micheal Luca
1, Jai Singh
1, Vijay Mittal
1, Ernesto Drelichman
1, Jasneet S. Bhullar
11Henry Ford Providence Hospital, Southfield, MI; 2Princess Nourah bint Abdulrahman University, Riyadh, Saudi Arabia
BackgroundColorectal cancer (CRC) prevention relies heavily on colonoscopy, with adenoma detection rate (ADR) serving as a key quality indicator. However, ADR varies significantly among endoscopists, leading to missed polyps. Artificial intelligence (AI) has shown promise in improving ADR by assisting with real-time polyp identification. While randomized controlled trials (RCTs) and meta-analyses highlight the benefits of AI in increasing detection rates and reducing missed adenomas, concerns remain about its real-world applicability, effect on procedure time, and cost-effectiveness. This review examines the current status and practical utility of AI in real-time colonoscopy practice.
MethodsFollowing PRISMA guidelines, we conducted a systematic review of PubMed and Web of Science databases for articles in English published between January 2000 and August 2024. Meta-analyses and systematic reviews assessing AI’s role in ADR during colonoscopy were included, while articles on non-adenoma indications were excluded. Of 24 identified articles, 22 met the inclusion criteria. Data extraction was independently verified by two researchers for accuracy and consistency.
ResultsThe 22 included articles showed moderate heterogeneity 42%-63% across most studies, with significant heterogeneity 87%-95% in studies focusing on ADR and real-world data. The number of studies per meta-analysis ranged from 5 to 28, with higher heterogeneity in analyses involving more than 18 RCTs. Subgroup analyses considered adenoma size, polyp characteristics, endoscopist experience, and geographic differences, while sensitivity analyses addressed factors such as study design, withdrawal time, and imaging techniques.
DiscussionAI improves ADR by approximately 20% across studies, excelling in detecting small and challenging lesions, such as proximal and flat adenomas. However, in real-world practice, AI underperforms compared to expert endoscopists, particularly in non-randomized settings, showing limited benefits outside controlled conditions. Additionally, no meta-analyses have explored AI’s impact on training fellows and residents. Some experts caution that AI reliance may hinder trainees’ development of essential observational skills, potentially leading to less thorough examinations and reduced vigilance in assessing difficult-to-reach areas.
ConclusionAI effectively enhances ADR and aids in detecting complex lesions but often falls short compared to expert endoscopists in real-world settings. The lack of research on AI’s role in training raises concerns about over-reliance on technology, which could impede critical skill development. Future research should focus on optimizing AI integration into training frameworks to support skill acquisition and retention in clinical practice.
Back to 2025 Abstracts
