Renji endoscopic submucosal dissection video data set for Esophagus

Area of Science:

• Medical Artificial Intelligence
• Surgical Education Technology
• Endoscopic Imaging Analysis

Background:

• Artificial intelligence (AI) significantly enhances surgical training and outcomes.
• Deep learning for endoscopic procedures requires extensive, annotated datasets.
• Existing datasets lack comprehensive phase annotations for esophageal procedures.

Purpose of the Study:

• To introduce the Renji database, the first publicly available collection of esophageal endoscopic submucosal dissection (ESD) videos.
• To provide detailed, phase-specific annotations for 25 ESD procedures.
• To establish a benchmark for future esophageal ESD research and AI development.

Main Methods:

• Collected 25 endoscopic submucosal dissection (ESD) video recordings of esophageal lesions.
• Developed a specialized annotation framework for phase recognition.
• A team of expert endoscopists meticulously annotated 141,909 video phases.
• Made the dataset publicly accessible via the Figshare repository.

Main Results:

• The Renji database comprises 25 esophageal ESD videos with 141,909 phase annotations.
• This is the first publicly available dataset of its kind for esophageal ESD.
• The dataset is freely accessible for research purposes.

Conclusions:

• The Renji database represents a significant contribution to AI-driven surgical research.
• It will facilitate the development and validation of AI algorithms for esophageal ESD.
• This resource aims to improve surgical proficiency and patient outcomes through advanced analytics.