How AI can boost colonoscopy quality in real time

Researchers out of Wuhan, China, have developed a new AI-based quality improvement system for colonoscopies, sharing their findings in The Lancet: Gastroenterology & Hepatology. Could such a system lead to more consistent colonoscopies across the board?

“Colonoscopy has an important role in the diagnosis and treatment of lower gastrointestinal lesions,” wrote lead author Dexin Gong, MD, Renmin Hospital of Wuhan University, and colleagues. “However, findings of a meta-analysis showed that the proportion of adenomas missed during colonoscopy is as high as 26%, while 12% of patients with colorectal cancer were not diagnosed at a previous colonoscopy done within 5 years. As endoscopic technology continues to improve, colonoscopy quality is gaining further attention.”

The real-time quality improvement system, ENDOANGEL, was designed to achieve three primary goals: identify whether an endoscope was inside or outside the body and locate the caecum, avoid blind spots that sometimes occur due to slipping, and monitor withdrawal speed during the procedure. Deep convolutional neural networks (DCNNs) were trained on still images, and the team conducted a single-center trial to grade the system’s overall performance when it comes to colonoscopy quality.

For that trial, Gong et al. explored data from 704 patients treated from June 18 to Sept. 6, 2019, paying close attention to the adenoma detection rate (ADR) of each colonoscopy. Overall, the ADR was “significantly better” in the ENDOANGEL group than in a control group where the system was not used. No adverse events were reported during the trial.

These findings, the authors explained, show that the ENDOANGEL system can boost the ADR during colonoscopies, enhancing the overall quality and providing improved consistency. This comes at a time, they added, which improvements in colonoscopy quality seem to be lacking.

“Changes in the medical training environment, which greatly emphasize patients' safety and efficiency of endoscopy procedures, have highlighted the need for assessment tools capable of objectively assessing endoscopy procedures to improve procedural quality,” the authors wrote. “Colonoscopy quality is important, and efforts to improve quality include education and videorecording for possible quality review, but no substantial improvement in ADR has been achieved, nor is the effect of increasing ADR known, by these methods.”