Researchers have unearthed that inexperienced physicians conducting AI-assisted colonoscopies significantly enhanced the detection of tumors. By integrating AI with this medical tool, the likelihood of overlooking potential precursors to colorectal cancer can be reduced.
The advancement of AI has enhanced various clinical diagnostic tools such as MRI, ultrasound, and mammography. Colonoscopy is now poised to benefit from a computer-assisted transition.
Colonoscopy, a procedure that entails inserting an endoscope into the colon to examine its inner walls and identify and remove premalignant polyps (adenomas), plays a crucial role in reducing the risk of death related to gastrointestinal cancer. However, the diagnostic accuracy of colonoscopy can be compromised, with up to 26% of adenomas and 9% of advanced adenomas being missed, thereby increasing the risk of mortality and unfavorable outcomes. Factors contributing to adenoma oversight include flat morphology, inadequate colon preparation, and insufficient endoscopist experience.
A team of researchers from the Faculty of Medicine at the Chinese University of Hong Kong (CUHK) investigated the impact of AI-assisted colonoscopy on adenoma detection rates (ADR) when utilized by inexperienced physicians.
Lead author of the study, Louis Lau Ho-shing, emphasized the significance of their analysis in shaping the future integration of AI in medical practice and procedural training. Given that novice endoscopists often lack proficiency and require additional support in the initial phases of their training, the study underscores how AI can provide visual guidance to enhance skill development uniformly. The results indicated that the technology enhances endoscopists’ adenoma detection capabilities across different experience levels.
The AI system employed in the study was a deep learning model called computer-aided polyp detection (CADe), which has demonstrated significant real-time improvements in adenoma detection in previous trials. Over the course of the study conducted between April 2021 and July 2022, 22 novice endoscopists with limited personal experience (less than 500 procedures) and under three years of training were tasked with evaluating their performance using the AI-assisted endoscopy system. The participants were categorized into beginner (less than 200 procedures) and intermediate (200 to 500 procedures) groups.
The primary focus of the study was on ADR, with secondary endpoints including the detection of adenomas of varying sizes (less than 5 mm, 5 to 10 mm, greater than 10 mm, etc.). Notably, an advanced adenoma, considered a precursor to colon cancer, must be at least 10 mm in size.
During the colonoscopies conducted on 766 patients, the trainee endoscopists performed procedures on 386 patients with the assistance of CADe, while the remaining patients underwent standard colonoscopy. Overall, the ADR in the CADe group was significantly higher than that in the control group, with rates of 57.5% and 44.5%, respectively. Specifically, the CADe group exhibited ADRs of 40.4% for adenomas smaller than 5 mm compared to 25.0% in the control group, and 36.8% versus 29.2%, respectively. The ADRs for advanced adenomas did not show significant differences. Furthermore, beginner endoscopists in the CADe group demonstrated higher ADRs (60.0% vs. 41.9%) compared to the control group, as did intermediate-level endoscopists (56.5% versus 45.45%).
The researchers noted that the benefits of CADe for large and advanced adenomas remain to be fully elucidated. They recommend the simultaneous development of computer-aided polyp detection systems and the optimization of individual performance. Nevertheless, based on their findings, they advocate for the inclusion of AI tools in colonoscopy training programs.
