Artificial intelligence and computer-aided diagnosis in colonoscopy: current evidence and future directions

Area of Science:

• Gastroenterology
• Medical Imaging
• Artificial Intelligence

Background:

• Colonoscopy has high polyp miss rates (up to 22%), leading to interval colorectal cancers.
• Optical biopsy for in-vivo polyp classification is limited by interobserver variability and expert dependency.
• Computer-aided diagnosis (CAD) and AI offer potential solutions to enhance colonoscopy performance.

Purpose of the Study:

• To review the clinical applications of CAD and AI in colonoscopy.
• To summarize evidence on AI-powered decision-support software for polyp detection and characterization.
• To assess the impact of AI on improving colonoscopy quality and reducing diagnostic errors.

Main Methods:

• Review of existing literature on computer-aided diagnosis and artificial intelligence in colonoscopy.
• Analysis of studies evaluating real-time decision-support software for polyp detection, characterization, and quality feedback.
• Synthesis of evidence on AI algorithm performance compared to human experts for optical biopsy.

Main Results:

• AI algorithms, particularly those using deep learning, demonstrate performance comparable to human experts in optical biopsy.
• Real-time decision-support software shows promise in detecting and characterizing polyps during colonoscopy.
• AI can provide feedback on the technical quality of colonoscopic inspection, aiding in standardization.

Conclusions:

• Computer-aided diagnosis and AI are emerging as powerful tools to improve colonoscopy accuracy and consistency.
• These technologies have the potential to overcome limitations of traditional colonoscopy, such as high miss rates and interobserver variability.
• Further integration of AI into routine colonoscopy practice could significantly reduce colorectal cancer incidence and improve patient outcomes.