AI Tool Measures Colonoscopy Quality Across Thousands of Procedures

AI automates colonoscopy quality monitoring

Researchers at Northwestern Medicine have developed artificial intelligence software that can assess colonoscopy quality by analyzing procedure recordings, demonstrating accuracy across nearly 19,000 procedures performed by 55 physicians over 11 months.

The tool addresses a persistent challenge in gastroenterology: medical societies recommend routine quality reviews for every colonoscopy provider, but manual assessment remains time-consuming and difficult to implement at scale. The research, published in The American Journal of Gastroenterology, represents the first demonstration of AI comprehensively measuring quality across thousands of colonoscopies.

How the system works

The AI software reviews colonoscopy footage and identifies critical moments during procedures, including when the scope reaches the colon's beginning, when it's withdrawn, and when polyps are removed. The system measures withdrawal time—a key quality metric tracking how long physicians examine the colon during scope removal—and its measurements closely matched those recorded by nurses, validating the technology's accuracy.

Beyond replicating human measurements, the tool tracks quality indicators that would be impractical for manual review at large scale. These include counting polyps removed during each procedure and monitoring how frequently physicians use cold snare polypectomy, a guideline-recommended technique for removing small polyps.

Why it matters

Colorectal cancer rates are rising among young adults in the United States, making colonoscopy quality increasingly critical. Previous research has shown that measuring colonoscopy performance improves overall quality and reduces colorectal cancer mortality. By automating quality monitoring, this tool could enable systematic feedback across entire hospital systems—something currently beyond most institutions' capacity. High-quality colonoscopies require complete colon inspection, adequate examination time, and proper polyp removal techniques, but quality varies substantially between physicians.

Questions about AI in clinical practice

Dr. Rajesh Keswani, the study's lead author and associate professor of medicine in Northwestern's division of gastroenterology and hepatology, noted that his tool only assesses quality after procedures are completed, distinguishing it from AI systems that assist during colonoscopies. A 2024 Lancet study raised concerns that colonoscopists using AI detection assistance during procedures showed declining proficiency over time.

"One possibility is that physicians rely too much on AI, which leads to deskilling," Keswani said. "Alternatively, AI can teach us about blind spots and make us better clinicians."

Keswani's research team is currently studying AI's role in teaching colonoscopy to trainees. He emphasized that measuring quality is essential for providing safe and effective screening colonoscopy care.

The findings were first reported by News Medical and involved additional Northwestern authors including Dr. John Pandolfino, Dr. Mozziyar Etemadi, Matthew Wittbrodt, Alex Heller, Kristjana Kristinsdottir, and Evandros Kaklamanos.