Behavior-Based Skill Assessment for Open Surgery from Multi-View and Egocentric Videos
Jun 1, 2026·
,,,·
0 min read
Qiaomu Miao
Jonathan Price
Apostolos Tassiopoulos
Dimitris Samaras
Abstract
Automated surgical skill assessment has the potential to improve medical training and objective evaluation of technical proficiency. While significant progress has been made for minimally invasive surgery using instrument tracking and stable endoscopic views, automated assessment for open surgery remains challenging due to complex visual environments and the lack of reliable tool telemetry. In this work, we propose an interpretable behavior-based framework that models surgical expertise through the holistic kinematic patterns of the surgeon. We first introduce a synchronized multi-view dataset for open surgical simulation containing 14 hours of recordings from 26 participants across three expertise levels, captured using four fixed cameras and a head-mounted egocentric camera with expert OSATS annotations. From the reconstructed pose trajectories, we derive interpretable behavioral metrics that characterize postural stability, motion efficiency, trajectory smoothness, and cycle consistency. These features are used to train a lightweight neural network to predict surgical skill scores. Our approach achieves a Spearman correlation of 0.96 with expert ratings, substantially outperforming RGB-video and raw pose-sequence baselines. The proposed method also remains effective in an egocentric-only setting using 2D pose trajectories and head-motion estimation, demonstrating robustness across camera settings. These results highlight the effectiveness of explicit and interpretable behavior modeling for surgical skill assessment.
Type
Publication
IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2026
