SAGES 2026 Annual Meeting Emerging Technology Session

Background

Existing endoscopic databases for AI development rely on retrospective sampling, isolated images, or lack clinical context, limiting their utility for artificial intelligence (AI) validation and clinical research. This study establishes a prospective endoscopy video databank with standardized annotations enabling AI development and clinical outcomes research.

Methods

Prospective video databank established at the Montreal University Hospital Center (NCT06822616):

•Enrollment of patients (June 2022-ongoing) undergoing colonoscopy and EGD, providing informed consent for data collection and subsequent use in AI system development

•Video recordings de-identifiedfrom acquisition, without protected health information, all technical specifications optimized for AI training (full-length, full-frame videos in 1920x1080-pixel resolution and 60fps, and 10-bit color depth).

•Videos linked with structured, real-time clinical documentation:

•450-740 variables per procedure including anatomical landmarks, lesion characteristics, intervention timing, disease-specific scoring (i.e., IBD activity scores, Prague classification for Barrett’s), optical diagnoses, AI-assisted detection outputs, and quality metrics.

•Timestamps link video frames directly to clinical events, enabling frame-level navigation

•Pathology results subsequently integrated, establishing ground-truth correlation for each visualized lesion

Automated algorithms calculate procedure quality metrics: adenoma detection rates, withdrawal times, etc.

Results

As of December 2025, the databank includes 8739 endoscopic procedures in 7139 patients (mean age 62.3, 51.7% female):

•6433 colonoscopies and 2306 EGDs

•1291 colonoscopies in IBD patients (Table 1)

•Performed by 49 endoscopists

•8301 polyps were detected, timestamped, individuallycharacterized, and correlated to pathology results

•8631 biopsies timestamped, annotated, correlated to pathology

•Polyps and biopsies annotated with up to 49 and 45 variables

•

•The databank has supported several AI validation studies:

•For AI-based polyp detection1,2 and diagnosis3,4

•Participating endoscopists demonstrated 12% relative ADR improvement over 24 months through automated ADR monitoring and performance feedback

Conclusion

This prospective, FAIR-compliant databank of >8500 endoscopic procedures provides a high-quality, scalable foundation for the development and validation of AI systems in endoscopy and clinical outcomes research.

Expansion to four international centers targeting 20,000 procedures by 2027 is underway. Future enhancements include digital pathology integration, multi-omics and longitudinal data for IBD patients and registry linkage for long-term cancer incidence tracking. This video and clinical data capture platform enables rigorous AI systems validation, human-AI interaction studies and high-quality labelled data for AI systems development.

SAGES 2026 Annual Meeting Emerging Technology Session

Implementation of a Prospective, FAIR-Compliant Multimodal Video Databank for AI Development and Clinical Outcomes Research in Gastrointestinal Endoscopy

Poster Presenter

Authors

Affiliations

Part of Topic