Effect of artificial intelligence on novice-performed colonoscopy: a multicenter randomized controlled tandem study

Area of Science:

• Gastroenterology
• Medical Technology
• Artificial Intelligence in Medicine

Background:

• The effectiveness and safety of colonoscopies performed by novices using artificial intelligence (AI) assistance are not well-established.
• There is a need to evaluate the impact of AI on the diagnostic capabilities of less experienced endoscopists.

Purpose of the Study:

• To compare the lesion detection capabilities of novices, AI-assisted novices, and expert endoscopists.
• To assess whether AI assistance can make novice endoscopists non-inferior to experts in colonoscopy.

Main Methods:

• A multicenter, randomized, noninferiority tandem study involving 685 patients across 3 hospitals.
• Patients were randomized into three groups: control novice (CN), AI-assisted novice (AN), and control expert (CE).
• Adenoma miss rate (AMR) and polyp miss rate were primary outcomes, with repeat colonoscopies by AI-assisted experts to confirm detection.

Main Results:

• AI-assisted novices (AN group) demonstrated significantly lower adenoma miss rates (18.82%) and polyp miss rates (21.23%) compared to control novices (CN group: 43.69% and 35.38%, respectively).
• The AI-assisted novice group met the noninferiority margin when compared to the control expert group for both AMR (18.82% vs 26.97%) and polyp miss rate (21.23% vs 24.10%).

Conclusions:

• AI-assisted colonoscopy significantly reduces the adenoma miss rate for novices, achieving performance comparable to experts.
• AI technology can effectively enhance the withdrawal technique and overall competence of new endoscopists.