Real-time multicenter semantic segmentation for laparoscopic gastrectomy: development, external validation and...

Area of Science:

• Surgical Technology
• Artificial Intelligence in Medicine
• Computer Vision for Surgery

Background:

• Laparoscopic gastrectomy is a complex procedure where limited situational awareness can lead to errors.
• Existing artificial intelligence (AI)-based computer vision tools for intraoperative use lack generalizable, real-time deployment evidence.

Purpose of the Study:

• To develop and validate a real-time AI navigation system for laparoscopic gastrectomy.
• To assess the system's performance, feasibility for bedside use, and impact on surgeon cognitive workload.

Main Methods:

• A multicenter study utilized laparoscopic gastrectomy videos from four tertiary hospitals.
• A You Only Look Once version 10 (YOLOv10)-based model was trained to recognize 18 surgical categories (instruments and organs).
• The system was integrated into a user interface for real-time deployment, with performance metrics (mAP, recall, F1) and usability (NASA-TLX, SUS) assessed.

Main Results:

• The AI model achieved strong performance in external validation (box AP@[0.50-0.95] 0.684, mask AP@[0.50-0.95] 0.658).
• The system operated in real-time on a laptop-grade GPU, providing on-demand overlays without workflow disruption.
• Surgeon cognitive workload was low (NASA-TLX score 26.72±6.51) and system usability was good (SUS score 78.33±5.20).

Conclusions:

• A real-time AI navigation system for laparoscopic gastrectomy was successfully developed, externally validated, and deployed across multiple centers.
• The system demonstrated robust instrument and organ recognition and proved feasible for bedside use.
• Further prospective trials are recommended to evaluate the impact on intraoperative safety, efficiency, and surgical training.