Generative AI for synthetic surgical training videos
Abstract
Recent advancements in artificial intelligence (AI) have primarily been driven by the introduction of generative artificial intelligence (GenAI). Despite significant interest in applying AI to clinical data for current and future generations of surgeons1, its actual implementation continues to be hampered by certain challenges2. However, the application of GenAI has demonstrated an increasing capacity to overcome these boundaries and enhance its interpretability within the clinical context3. The aim of this study was to develop synthetic videos for surgical training utilizing GenAI and assess their realism and educational value. Three videos of real robot-assisted simulation tasks (suturing, knot tying, and needle handling, Fig. 1) were selected from the publicly and free available JIGSAWS dataset4. The CogVideo X5 tool is an AI-powered video Generation tool that enables creating video from text, images or video. CogVideo X was used to generate 15-s synthetic videos (Video S1, Fig. 2, resolution: 720 × 480 pixels; frame interpolation: 16 fps) based on real videos (see screenshot in Fig. 2). A text prompt and the actual video of each surgical task was the input for the generation of synthetic videos. To assess the realism and training capacity of the synthetic videos, the six videos (both real and synthetic) were evaluated in random order by 11 healthcare professionals from various surgical specialties (endoscopy, laparoscopy, and microsurgery). These professionals evaluated whether the video they were watching was real or synthetic and assessed its usefulness as a training material for minimally invasive surgery using a 5-point Likert scale. For analysis, true positive and true negative status represented videos that were correctly assigned as belonging or not belonging to a real or synthetic class. False positive and false negative status was assigned to videos that were wrongly assigned. Accuracy was calculated as a summary measure from the above classifications6.
Fields of Study & Subjects
Surgical Simulation and Training
Artificial Intelligence in Healthcare and Education
Anatomy and Medical Technology
Publication details
- Journal / source
- British journal of surgery
- Volume
- 113
- Issue
- 3
- Publication date
- 19 February 2026
- Language
- EN