Comparative analysis of personal protective equipment nonadherence detection: computer vision versus human observers

Area of Science:

• Healthcare technology
• Computer vision
• Infection control

Background:

• Human monitoring of personal protective equipment (PPE) adherence is limited by personnel needs and observer fatigue.
• Challenges in maintaining consistent PPE adherence monitoring in healthcare settings.

Purpose of the Study:

• To develop and assess an automated computer vision system for monitoring PPE adherence.
• To compare the system's performance against human observers in detecting PPE nonadherence.

Main Methods:

• An object detection and tracking system was trained to identify 15 classes of PPE (eyewear, masks, gloves, gowns).
• A video surveillance experiment compared system performance to 12 human observers under varying video durations and numbers of individuals.
• Performance was evaluated based on nonadherence detection rates.

Main Results:

• The automated system detected significantly more instances of nonadherence than human observers.
• Human observer performance decreased with longer video durations, while system performance remained consistent.
• The system achieved high accuracy with a sensitivity of 0.86, specificity of 1, and Matthew's correlation coefficient of 0.82.

Conclusions:

• Automated computer vision offers a scalable solution for monitoring hospital-wide infection control practices.
• The system's independence from observation duration surpasses limitations of human monitoring.
• This technology has the potential to significantly improve PPE usage and patient safety in healthcare settings.