Intelligent surgical workflow recognition for endoscopic submucosal dissection with real-time animal study.

Area of Science:

• Artificial Intelligence in Medicine
• Surgical Technology
• Gastroenterology

Background:

• AI has achieved human-level performance in various domains, but its application in therapeutic procedures like endoscopic submucosal dissection (ESD) remains underexplored.
• Cognitive assistance systems for surgical procedures require robust pre-clinical validation.

Purpose of the Study:

• To develop and validate an AI-enabled cognitive assistance system, named AI-Endo, for surgical workflow recognition in endoscopic submucosal dissection (ESD).
• To assess the AI-Endo system's performance across diverse datasets and integrate it into a live training environment.

Main Methods:

• AI-Endo was trained on a large dataset of 201,026 labeled frames from expert-performed ESD procedures over a decade.
• The model's performance was validated on diverse cases, including those from endoscopists with varying skill levels, different equipment, and international multi-center cohorts.
• AI-Endo was integrated with an Olympus endoscopic system and tested in animal studies for live ESD training.

Main Results:

• The AI-Endo model demonstrated outstanding performance on validation data, proving its generalizability across different endoscopists, systems, and international centers.
• The integrated system successfully provided AI-enabled cognitive assistance during live animal ESD training sessions.
• Automated reports generated from surgical phase recognition results facilitated skill assessment.

Conclusions:

• AI-Endo represents a significant advancement in AI-enabled cognitive assistance for therapeutic endoscopic procedures.
• The system's validated performance and integration capabilities highlight its potential to enhance surgical training and performance in ESD.
• AI-Endo offers a promising tool for objective skill assessment and improving procedural outcomes in gastrointestinal endoscopy.