AI Tool Automates Colonoscopy Quality Assessment Across Thousands of Procedures

Colorectal cancer is rising sharply among younger Americans, and colonoscopy remains one of the most reliable ways to catch it early. But there's a persistent problem: the quality of a colonoscopy depends heavily on who performs it. One physician might spend adequate time examining the colon and remove precancerous polyps with precision; another might rush through the same procedure. Medical organizations have long recommended that hospitals regularly audit every colonoscopist's work—a safeguard that sounds straightforward in principle but has proven logistically grueling to implement across busy practices and health systems.

Researchers at Northwestern Medicine have now built an artificial intelligence system that can measure colonoscopy quality automatically by watching the procedure footage. The tool reviews recordings to identify critical moments: when the scope reaches the start of the colon, when it's withdrawn, when polyps are removed. In a study published last month in The American Journal of Gastroenterology, the team analyzed nearly 18,600 colonoscopies performed by 55 physicians over eleven months and found the AI's measurements matched those of trained nurses with high accuracy. Dr. Rajesh Keswani, the study's lead author and an associate professor of gastroenterology at Northwestern's Feinberg School of Medicine, describes this as the first time an AI system has been shown to comprehensively assess the quality of thousands of procedures at once.

What makes the tool particularly valuable is its ability to track metrics that humans simply cannot measure at scale. Beyond withdrawal time—the duration a physician spends examining the colon as the scope is pulled back—the AI counts how many polyps were removed during each procedure and flags how often physicians used cold snare polypectomy, a guideline-recommended technique for removing small polyps. These are the kinds of granular performance details that matter for patient safety but would require someone to manually review thousands of hours of video footage. A high-quality colonoscopy, in the strictest sense, means the physician has inspected the entire colon thoroughly, spent sufficient time looking for abnormalities, and employed proper removal techniques. Research has shown that when hospitals measure and provide feedback on colonoscopy performance, overall quality improves and colorectal cancer mortality declines.

Keswani emphasizes that institutions cannot deliver safe and effective screening colonoscopies without the ability to measure quality. Automating that measurement process opens the door to systematic monitoring across entire hospital networks and healthcare systems—something that would have been prohibitively time-consuming to do by hand. The AI tool offers what he calls a scalable approach to quality assessment and clinician feedback.

Yet the introduction of AI into colonoscopy has raised legitimate concerns. Last year, a study published in The Lancet suggested that colonoscopists who began using an AI assistant to detect polyps showed declining proficiency over time, raising the specter of physicians becoming dependent on the technology and losing their own diagnostic skills. Keswani acknowledges the uncertainty around how AI will ultimately affect clinical practice and physician development. His tool, importantly, only evaluates quality after a procedure is complete—it does not assist during the colonoscopy itself. He frames the challenge as a genuine open question: physicians might over-rely on AI and lose competence, or AI might reveal blind spots in their technique and make them better clinicians. His team is currently investigating how AI affects the training of colonoscopy residents and fellows, trying to understand whether automation helps or hinders the next generation of gastroenterologists learning the procedure.