Diagnostic performance of real-time characterization in artificial intelligence-assisted colonoscopy

Area of Science:

• Gastroenterology
• Medical Imaging
• Artificial Intelligence

Background:

• Most colorectal polyps are small with low cancer risk.
• Artificial intelligence (AI) could enable strategies like leave-in-situ and resect-and-discard for polyps.
• AI diagnostic systems require predefined quality benchmarks for implementation.

Purpose of the Study:

• To assess the diagnostic performance of a computer-aided polyp characterization system (CADx).
• To predict the histology of diminutive rectosigmoid polyps during colonoscopy.
• To evaluate AI system performance in a Danish population.

Main Methods:

• Prospective diagnostic accuracy study across four endoscopy centers.
• Inclusion of adults referred for colonoscopy with various indications.
• Histopathology as the reference standard for polyp classification (adenoma vs. non-adenoma).

Main Results:

• Analysis included 278 patients and 772 polypectomies.
• For diminutive rectosigmoid polyps (n=184), CADx achieved 93% sensitivity and 33% specificity.
• Overall accuracy was 71%, with a positive predictive value of 70% and negative predictive value of 74%.

Conclusions:

• The CADx system demonstrated high sensitivity but lower specificity than expected, due to a high false-positive rate.
• Inconsistent findings across studies highlight challenges in standardizing AI characterization systems.
• Current AI performance is insufficient for reliable implementation in resect-and-discard or leave-in-situ strategies, despite being a promising adjunct.