A prospective multicenter randomized controlled trial on artificial intelligence assisted colonoscopy for enhanced...

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Gastroenterology

Background:

• Colorectal cancer (CRC) screening and prevention rely heavily on colon polyp detection and removal during colonoscopy.
• Early detection of polyps is crucial for reducing CRC incidence and mortality.

Purpose of the Study:

• To develop an artificial intelligence (AI)-assisted colon polyp detection program using the RetinaNet algorithm.
• To verify the clinical utility and performance of the AI program in a real-world setting.

Main Methods:

• Development of an AI model using a fully anonymized dataset with 10-fold cross-validation (9,639 training images, 1,070 validation images per fold).
• Transfer learning was applied using a still image-based model to video data from 56 patients for real-time endoscopy application.
• Prospective randomized controlled trial across six institutions involving 805 patients to compare AI-assisted colonoscopy with standard colonoscopy.

Main Results:

• The AI-assisted group demonstrated significantly higher polyp detection rates (PDR) compared to the standard colonoscopy group.
• Adenoma detection rates (ADR) were also significantly improved in the AI-assisted group.
• Multivariate analysis indicated an odds ratio (OR) of 1.50 for polyp detection when using the AI model.

Conclusions:

• The AI-assisted colon polyp detection program is clinically beneficial, enhancing polyp identification during colonoscopy.
• Implementation of this AI tool can lead to improved adenoma detection rates, thereby strengthening colorectal cancer prevention strategies.