Colorectal mucosal exposure area assessment using artificial intelligence: a multicenter prospective observational...

Area of Science:

• Gastroenterology
• Medical Technology
• Artificial Intelligence in Medicine

Background:

• Colonoscopy quality is crucial for detecting colorectal neoplasia.
• Current quality indicators may not fully capture mucosal exposure during the procedure.
• A novel indicator, cumulative colorectal mucosal exposure area (CCMEA), was developed using deep learning.

Purpose of the Study:

• To introduce and validate the cumulative colorectal mucosal exposure area (CCMEA) as a novel quality control indicator for colonoscopy.
• To assess the performance of a deep learning-based CCMEA system in a clinical setting.
• To determine if CCMEA can differentiate colonoscopies based on adenoma detection rates.

Main Methods:

• A deep learning system utilizing ResNet50 and UNet++ was developed to calculate CCMEA.
• A CCMEA threshold of 2000 was established based on the 25% adenoma detection rate (ADR) quality benchmark.
• A multi-center prospective observational study evaluated 510 patients, comparing ADR and other detection rates between CCMEA-qualified and unqualified groups.

Main Results:

• The CCMEA system classified 270 patients as qualified and 240 as unqualified (CCMEA threshold: 2000).
• The qualified group exhibited a significantly higher ADR (53.70% vs. 14.17%, aOR=7.97), particularly for small adenomas (42.22% vs. 10.00%).
• Higher polyp detection rates (89.63% vs. 40.00%), adenoma counts (1.00 vs 0.17), and polyp counts (5.79 vs 1.32) were observed in the qualified group.

Conclusions:

• The cumulative colorectal mucosal exposure area (CCMEA) system, with a threshold of 2000, effectively identifies higher quality colonoscopies.
• CCMEA-qualified colonoscopies are associated with significantly improved adenoma detection rates.
• CCMEA shows promise as a valuable objective indicator for colonoscopy quality control.