Browser-Based Video Supports Scalable Computer Vision Analysis for Open Surgical Skill Assessment

Browser-based laptop-webcam recordings of open surgical hand motion achieved 85.9% PCK (95% CI 80.3–91.0), comparable to overhead recordings, using an internal keypoint pipeline.

• Why it matters: Expert-dependent assessment of open surgical skills is constrained, and ML hand keypoint detection needs validation in accessible browser-based recording settings.
• What they did: In a retrospective technical validation using 203 annotated frames from 14 PGY-1 interns (6,232 ground-truth keypoints), the internal pipeline and MediaPipe Hands were tested across browser-based and overhead videos.
• The result: The internal pipeline matched performance across environments (PCK 85.9% vs 81.5%; detection 94.9% vs 97.7%), while MediaPipe detection was far lower (5.3% vs 15.4%), supporting scalable browser-based assessment.

AI summary of the abstract · may contain errors · not medical advice · read the paper.