Deep Learning Analysis of Surgical Video Recordings to Assess Nontechnical Skills

Area of Science:

• Surgical Innovation
• Medical Imaging Analysis
• Human Factors in Medicine

Background:

• Assessing nontechnical skills in operating rooms (ORs) is vital for surgical performance and patient safety.
• Automated, real-time evaluation of these skills presents a significant challenge.

Purpose of the Study:

• To investigate the feasibility of using motion features from surgical videos for automatic assessment of nontechnical skills.
• Focus on cardiac surgical procedures.

Main Methods:

• Cross-sectional study using video recordings of 30 cardiac surgeries.
• Extracted body pose estimations and team motion features using OpenPose.
• Rated nontechnical skills using the Non-Technical Skills for Surgeons (NOTSS) tool by 3 expert raters.

Main Results:

• NOTSS scores significantly correlated with team displacement trajectory (r=0.51), acceleration (r=0.48), and entropy (r=-0.52).
• Multiple linear regression indicated team trajectory and displacement entropy were associated with NOTSS scores.
• Adjusted R-squared values were 0.335 for trajectory and 0.304 for entropy.

Conclusions:

• A significant association exists between operating room team movements and nontechnical skills ratings during cardiac surgery.
• Automated surgical video analysis shows potential for enhancing nontechnical skills assessment.
• Further validation across diverse settings and specialties is recommended.