Automated artificial intelligence-based phase-recognition system for esophageal endoscopic submucosal dissection...

Area of Science:

• Gastroenterology
• Artificial Intelligence in Medicine
• Medical Imaging Analysis

Background:

• Endoscopic submucosal dissection (ESD) for superficial esophageal cancer involves complex, multi-step endoscopic procedures.
• Analyzing individual ESD phases is crucial but labor-intensive, limiting practical application.

Purpose of the Study:

• To develop an advanced artificial intelligence (AI)-based system for automated phase recognition during endoscopic submucosal dissection (ESD).
• To enable automatic identification of each endoscopic phase using video image analysis.

Main Methods:

• A deep neural network was developed for automated phase recognition.
• The system was trained and validated using 94 videos of ESD procedures for superficial esophageal cancer, annotated by two gastroenterologists.

Main Results:

• The AI model achieved an overall accuracy of 90% for automated phase recognition.
• Average precision, recall, and F-score were 91%, 90%, and 90%, respectively.
• Validation on representative videos demonstrated the AI system's clinical utility.

Conclusions:

• An AI-based automated phase-recognition system for esophageal ESD can be successfully implemented with high accuracy.
• This is the first report on automated recognition of ESD treatment phases.
• The system facilitates detailed phase analysis, potentially identifying quality indicators and unmet needs for novel treatments.