Motion-Based Confidence Score to Support the Practical Application of rPPG Methods in Health Monitoring.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Health Informatics

Background:

• Remote photoplethysmography (rPPG) is a non-invasive technique for remote physiological monitoring.
• rPPG accuracy is challenged by motion artifacts, skin tone variations, and ambient light.
• Improving rPPG reliability is crucial for real-world health monitoring and reducing false alarms.

Purpose of the Study:

• To develop a confidence score for rPPG measurements to enhance reliability in real-world settings.
• To identify motion-related variables that correlate with rPPG accuracy.
• To improve the robustness, reproducibility, and generalizability of rPPG models.

Main Methods:

• A confidence score was developed based on motion-related features.
• A classifier was trained using data from three distinct datasets.
• The model focused on identifying variables strongly correlating with measurement accuracy.

Main Results:

• The developed confidence score demonstrated high performance in detecting inaccurate heart rate measurements.
• Area Under the Curve (AUC) values consistently exceeded 0.93 across all datasets.
• Motion-related features proved effective in predicting rPPG measurement quality.

Conclusions:

• The proposed confidence score significantly enhances the reliability of rPPG measurements.
• The model's high performance, even with limited features, highlights its potential for real-world applications.
• This approach can reduce false alarms and improve the accuracy of remote health monitoring systems.