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

This study proposed a new quality control indicator for colonoscopy, the cumulative colorectal mucosal exposure area (CCMEA), to assess mucosal exposure, constructed a CCMEA system based on deep learning, and validated the indicator in a multicenter prospective observational study.The CCMEA system was based on ResNet50 and UNet++. A CCMEA threshold was determined on the basis of an adenoma detection rate (ADR) of 25%. A multicenter prospective observational study was conducted to evaluate the system and the threshold in clinical practice. Based on the CCMEA threshold, patients were divided into qualified and unqualified colonoscopy groups. The ADR and other lesion detection rates were then compared between the two groups.510 participants who underwent colonoscopy were evaluated, being grouped as having qualified (n = 270) or unqualified (n = 240) colonoscopies based on a CCMEA qualification threshold of 2000. The ADR was 39.5 percentage points higher in the qualified group than in the unqualified group (53.7% vs. 14.2%; adjusted odds ratio [aOR] 8.0, 95%CI 5.0-12.8; <i>P</i> < 0.001), and notably was higher for lesions ≤5 mm (42.2% vs. 10.0%; aOR 6.9, 95%CI 4.1-11.5; <i>P</i> < 0.001). The qualified group also had a significantly higher polyp detection rate (89.6% vs. 40.0%; aOR 13.1, 95%CI 7.8-21.8; <i>P</i> < 0.001) and higher mean numbers of both adenomas (1.0 vs. 0.2; adjusted incident rate ratio [aIRR] 5.9, 95%CI 4.3-8.4; <i>P</i> < 0.001) and polyps (5.8 vs. 1.3; aIRR 4.0, 95%CI 3.5-4.5; <i>P</i> < 0.001).The CCMEA qualified group, based on a CCMEA threshold of 2000, showed a higher ADR than the unqualified group, indicating CCMEA could be a promising colonoscopy quality indicator.