See more, detect more: investigating colonic inspection techniques and polyp detection using machine learning

Area of Science:

• Medical technology
• Artificial intelligence
• Gastroenterology

Background:

• Colorectal cancer (CRC) remains a significant mortality cause despite screening.
• Adenoma detection rate (ADR) is crucial for colonoscopy effectiveness but is limited by inspection quality.
• Current simulation-based training (SBT) lacks objective assessment of inspection skills.

Purpose of the Study:

• To develop an automated machine learning (ML) model for objective assessment of colonoscopy inspection performance.
• To identify key feedback metrics for improving adenoma detection rate (ADR).

Main Methods:

• A Vision Transformer (ViT)-based ML model was created to recognize inspection techniques during SBT.
• Model performance was validated using accuracy, precision, recall, and F1 score.
• The model analyzed videos from experts, intermediate endoscopists, and novices using physical simulators with polyps.

Main Results:

• The ViT model demonstrated high accuracy in recognizing inspection techniques.
• Increased time inspecting obscured regions correlated with higher polyp detection rates (PDR).
• Experts showed superior PDR and spent more time on detailed inspection compared to novices.

Conclusions:

• The ViT ML model objectively assesses colonoscopy inspection skills and differentiates performance levels.
• This technology enables targeted feedback for improving inspection techniques and ADR.
• The model has the potential to enhance colonoscopy effectiveness in CRC screening.