Researchers Automated Surgical Skill Assessments
A new AI framework uses video foundation models to evaluate trainee performance in standard surgical tasks.
Updated on Sept. 18, 2026 in Artificial Intelligence
Researchers have introduced VBA-Net+, a research-stage framework that automates the assessment of surgical skills by analyzing procedural video. The system provides continuous score prediction and pass-fail classification for standardized FLS tasks.
Why it matters
Automated assessment reduces the need for manual review in surgical education by providing objective, consistent grading. This approach aims to accelerate feedback loops for trainees by using foundation models to quantify performance on unseen surgical tasks.
The framework utilizes frozen feature extractors including VideoPrism, V-JEPA2, and VideoMAE v2. Performance metrics surpassed a frame-level SimCLR baseline, with AUC for pass-fail classification reaching as high as 0.9906 for pattern cutting tasks.
The players
VBA-Net+
A research-stage AI framework designed to perform automated surgical skill assessment and FLS task scoring.
The details
VBA-Net+ operates by training a lightweight fully convolutional head—a neural network layer that processes input data through sliding filters—on embeddings generated by frozen video-encoder pipelines. By using these embeddings offline, the system effectively translates raw video input into standardized FLS (Fundamentals of Laparoscopic Surgery) scores. The model relies on participant-level leave-one-user-out cross-validation to ensure the evaluation metrics hold for unseen trainees.
Timeline
September 18, 2026: Publication of research paper.
The Tech Race
This development marks a shift from manual expert evaluation toward algorithmic scoring in surgical training. It builds on the broader trend of applying video foundation models to automate high-stakes performance benchmarks.
This system remains in the research stage and is currently limited to standardized suturing and pattern-cutting tasks. Its future application depends on further validation for complex, non-standardized clinical procedures in surgical residency programs.
The takeaway
The research demonstrates that foundation models can achieve high accuracy in objective surgical grading through frozen embedding extraction. Future developments will focus on moving from controlled FLS tasks to evaluating performance in unpredictable, real-world surgical settings.
Further reading
For more on the development of machine learning in professional training, visit Artificial Intelligence.
Source note: This article includes information reported by Nature.
