Using mmWave Radar and Deep Learning to Classify Caregiver Activities for Infection Prevention

Area of Science:

• Medical Technology
• Artificial Intelligence
• Infection Control

Background:

• Healthcare-associated infections (HAIs) are a significant clinical challenge.
• Caregiver activities in patient rooms are critical in disease transmission.
• Existing monitoring methods may compromise patient and staff privacy.

Purpose of the Study:

• To develop and evaluate a privacy-preserving activity recognition system for caregivers.
• To classify caregiver actions using 3D point cloud data from mmWave radar.
• To assess the infection risk associated with different caregiver activities.

Main Methods:

• Collected a dataset of 30 distinct caregiver activities in a simulated hospital environment.
• Utilized 3D point cloud data from mmWave radar for activity recognition.
• Evaluated deep learning models (PointNet++, Pointformer, DGCNN) using 5-fold cross-validation.

Main Results:

• PointNet++ achieved 75.16% accuracy and 67.08% F1-score for 30 activities.
• Grouping activities by risk level improved performance to 90.84% accuracy and 88.68% F1-score.
• Demonstrated feasibility of non-intrusive, privacy-preserving activity recognition.

Conclusions:

• mmWave radar-based activity recognition is effective for monitoring caregiver actions and infection risks.
• This technology can enhance infection prevention strategies in healthcare settings.
• The system offers a privacy-preserving solution for clinical environments.