An artificial intelligence system for qualified mucosal observation time during colonoscopic withdrawal

Area of Science:

• Gastroenterology
• Medical Imaging
• Artificial Intelligence

Background:

• Colonoscopic withdrawal time is critical for adenoma detection rate (ADR) and reducing colorectal cancer risk.
• Manual quantification of qualified mucosal observation time (QMOT) is challenging in routine colonoscopies.
• Enhanced mucosal observation during colonoscopy is linked to improved ADR.

Purpose of the Study:

• To develop and validate an artificial intelligence (AI) system, QAMaster, for automatic QMOT calculation.
• To assess the correlation between QMOT and ADR.
• To evaluate QAMaster's potential as a tool for improving colonoscopy quality assessment.

Main Methods:

• Development of QAMaster with two AI models: Model I for image quality analysis and Model II for anatomical landmark identification.
• Training datasets included 57,235 images from 64 patients for Model I and 7,712 images from 3,013 patients for Model II.
• Patients were stratified into high (≥90s) and low (<90s) QMOT groups to compare ADR.

Main Results:

• Model I achieved AUCs of 0.980-0.991, and Model II achieved AUCs of 0.977-0.997, demonstrating high accuracy.
• In a cohort of 482 patients, the high-QMOT group showed a significantly higher ADR (36.54%) compared to the low-QMOT group (19.94%).
• The adjusted odds ratio for detecting adenomas in the high-QMOT group was 2.02 (95% CI 1.23-3.33).

Conclusions:

• QAMaster accurately calculates QMOT during colonoscopy withdrawal.
• Higher QMOT is significantly associated with an increased ADR.
• QAMaster offers a promising, automated solution for assessing colonoscopy withdrawal quality and potentially improving patient outcomes.