Validation of an AI-Based Anatomical Landmark Recognition System for Pelvic Lymph Node Dissection in Gynecologic Surgery

Study Objective

To evaluate whether an AI-based anatomical landmark recognition support system for pelvic lymph node dissection (PLND) can improve the anatomical recognition abilities of gynecologic surgeons with varying levels of surgical expertise.

Design

Prospective, multi-arm observer performance study evaluating organ recognition with and without AI support.

Setting

A total of 640 laparoscopic video clips were prepared from 10 hysterectomy cases: 320 without AI overlay and 320 with AI. Each set included clips with and without the ureter, obturator nerve, external iliac artery, and vein.

Patients or Participants

Twelve gynecologic surgeons were enrolled and stratified into three experience-based groups: Group A (4 laparoscopic board-certified experts), Group B (4 obstetrics and gynecology specialists without laparoscopic certification), and Group C (4 trainees in residency). Each participant provided consent. The study was conducted over a one-month period.

Interventions

Participants were first asked to identify key pelvic structures (ureter, obturator nerve, external iliac vessels) in selected video clips without AI assistance. Subsequently, they reviewed the clips with AI-based overlay highlighting the anatomical targets and repeated the identification task.

Measurements and Primary Results

Accuracy of anatomical recognition (sensitivity and specificity) was calculated for each group with and without AI support. Across all groups, AI support improved identification of the ureter (mean sensitivity from 47.1% to 67.9%), obturator nerve (65.2% to 78.8%), external iliac artery (83.5% to 91.9%) and vein (71.3% to 90.0%) with all p-values < 0.05. The greatest improvement was observed in the trainee group (Group C), suggesting AI assistance is particularly beneficial for less experienced surgeons.

Conclusion

The AI-based anatomical landmark recognition support system for PLND significantly enhanced surgeon’s organ recognition across all experience levels, with the most pronounced benefit observed in trainees. These findings support integrating AI systems into surgical education and real-time intraoperative guidance to improve anatomical understanding and reduce the risk of injury. Further studies in live surgical settings are warranted to assess real-world impact on clinical outcomes.