University of Alabama at Birmingham Birmingham, AL
Erik Roepke, MD1, John Cooper, MD2, Dalton A. Norwood, MD1, Amanda Cartee, MD3, Fayez Sarkis, MD3, Patricia Ajayi-Fox, MD1, Anam Hameed, MD3, Ramzi Mulki, MD4, Sergio Sánchez-Luna, MD1, Douglas Morgan, MD, MPH1, Shajan Peter, MD1 1University of Alabama at Birmingham, Birmingham, AL; 2University of Alabama at Birmingham Heersink School of Medicine, Birmingham, AL; 3University of Alabama at Birmingham Hospital, Birmingham, AL; 4Basil I. Hirschowitz Endoscopic Center of Excellence, The University of Alabama at Birmingham Heersink School of Medicine, Birmingham, AL Introduction: Colonoscopy remains the gold standard for colorectal cancer (CRC) screening, though variability in adenoma detection rates (ADR) among gastroenterologists persists. Recent advances in artificial intelligence (AI), such as computer aided detection (CAD-e) systems, may enhance real-time polyp recognition and reduce ADR variability. This study compares the efficacy between AI-assisted colonoscopy versus traditional colonoscopy in high-risk patients. Methods: We performed a retrospective chart review of screening and surveillance colonoscopies in high-risk patients between February 2023 and September 2024 at a large tertiary academic center. High-risk criteria included personal or family history of CRC or adenomas, IBD, or genetic syndromes such as familial adenomatous polyposis and Lynch syndrome. Two different CAD-e systems were utilized. We evaluated patient demographics, cecal intubation rate, ADR, number of adenomas identified and removed, and sessile serrated polyp (SSP) detection rate. Categorical variables were analyzed using chi-square test, and continuous variables were analyzed using two-sided t-tests. Results: 3,023 colonoscopies were analyzed, with CAD-e used in 33.5% of cases. Patients in the AI-assisted cases were 1.9 years older (p< 0.001), had longer withdrawal times (863 vs 789 seconds, p< 0.001), and had higher cecal intubation rates (96.9% vs 94.4%, p< 0.001). AI-assisted cases had a 33% increase in ADR (95% CI 1.24-1.43, p< 0.001). No significant differences were found in detection of adenocarcinomas, advanced neoplasms, or >10 adenomas. However, AI-assisted cases increased the detection rate of hyperplastic polyps, 1-2 adenomas, and 3-10 adenomas (p< 0.001 for each). For SSPs < 1 cm, detection increased by 78% (95% CI 1.34-2.36, p< 0.001). Unassisted cases retrieved 0.78 fewer polyps on average compared to AI-assisted (p< 0.001). Discussion: Use of CAD-e systems may increase ADR in high-risk patients, especially when adenomas are rare and diminutive. Cecal intubation rate is unlikely to be truly affected by use of AI. Age may have impacted the results, as older patients are more likely to have adenomas. Lastly, the use of CAD-e systems may precipitate longer case and withdrawal times due to retrieval of more adenomas and hyperplastic polyps; however, we expect withdrawal times to decrease with improvement in AI technology that facilitates the differentiation between adenomatous and hyperplastic polyps.
Figure: Table 1: Comparison of performance metrics between AI-assisted versus traditional colonoscopy
Disclosures: Erik Roepke indicated no relevant financial relationships. John Cooper indicated no relevant financial relationships. Dalton Norwood indicated no relevant financial relationships. Amanda Cartee indicated no relevant financial relationships. Fayez Sarkis indicated no relevant financial relationships. Patricia Ajayi-Fox indicated no relevant financial relationships. Anam Hameed indicated no relevant financial relationships. Ramzi Mulki indicated no relevant financial relationships. Sergio Sánchez-Luna indicated no relevant financial relationships. Douglas Morgan indicated no relevant financial relationships. Shajan Peter: Castle biosciences – Advisory Committee/Board Member. Olympus corporation – Advisory Committee/Board Member.
Erik Roepke, MD1, John Cooper, MD2, Dalton A. Norwood, MD1, Amanda Cartee, MD3, Fayez Sarkis, MD3, Patricia Ajayi-Fox, MD1, Anam Hameed, MD3, Ramzi Mulki, MD4, Sergio Sánchez-Luna, MD1, Douglas Morgan, MD, MPH1, Shajan Peter, MD1. P2998 - A Comparative Analysis of AI-Assisted vs Traditional Screening and Surveillance Colonoscopies in High-Risk Patients, ACG 2025 Annual Scientific Meeting Abstracts. Phoenix, AZ: American College of Gastroenterology.
