Measuring provider-level differences in perioperative workflow using computer vision-based artificial intelligence

Area of Science:

• Surgical workflow analysis
• Artificial intelligence in healthcare
• Computer vision applications

Background:

• Perioperative workflows are complex and vary significantly between providers.
• Objective measurement of workflow efficiency is crucial for quality improvement.
• Existing methods for workflow analysis lack granularity and scalability.

Purpose of the Study:

• To evaluate provider-level variability in the full perioperative workflow.
• To implement a computer vision-based artificial intelligence (AI) system for automated event detection and timestamping.
• To assess the system's ability to segment surgical cases into distinct workflow phases.

Main Methods:

• A cross-sectional study analyzed 2502 total knee arthroplasty cases using an ambient surgical platform with cameras.
• A YOLO-based model identified patients, staff, and equipment.
• A transformer-based event detector predicted key perioperative events, segmenting cases into eight workflow phases.

Main Results:

• The computer vision system demonstrated high agreement with ground truth annotations.
• Significant provider-level variability was found in all workflow segments except room exit.
• The active procedure phase showed the greatest variability among surgeons, followed by room setup among circulating and scrub nurses.

Conclusions:

• Automated workflow segmentation via computer vision offers a scalable approach to evaluate perioperative efficiency with high granularity.
• This technology can support targeted surgical quality improvement initiatives.
• The findings highlight opportunities for standardizing perioperative processes to enhance patient care.