Is the Transverse Colon Overlooked? Establishing a Comprehensive Colonoscopy Database from a Multicenter...

Area of Science:

• Gastroenterology
• Medical Informatics

Background:

• Colonoscopy is a high-volume endoscopic procedure.
• Large colonoscopy datasets are valuable for quality assurance and AI development.

Purpose of the Study:

• To build a comprehensive colonoscopy database for quality assurance.
• To train artificial intelligence (AI) models for colonoscopy analysis.

Main Methods:

• Data collected via a cluster-randomized controlled trial by a designated medical student.
• Included full colonoscopy videos, colonoscope position, timestamps, pathology reports, endoscopist descriptions, and patient discomfort.
• 1447 patients included; 1191 completed colonoscopies.

Main Results:

• Withdrawal time was significantly longer in colonoscopies with polypectomies for most colon segments (p<0.001).
• No significant difference in withdrawal time was observed in the transverse colon (p=0.92).
• An AI was trained to objectively evaluate bowel preparation using the database.

Conclusions:

• The established database is the most thorough to date, containing unique information.
• The transverse colon exhibits different withdrawal time characteristics during polypectomies compared to other segments.
• An AI for bowel preparation evaluation was developed, with potential for future AI collaborations.