Evaluating Visual Photoplethysmography Method

Area of Science:

• Biomedical Engineering
• Physiological Monitoring
• Signal Processing

Background:

• Cardiovascular diseases (CVDs) are a leading cause of global mortality, necessitating effective monitoring strategies.
• Current physiological monitoring often relies on uncomfortable contact devices, limiting user adherence and data quality.
• Stress-related illnesses incur significant healthcare costs, highlighting the need for accessible stress detection methods.

Purpose of the Study:

• To benchmark remote photoplethysmography (rPPG) against contact photoplethysmography (PPG) sensors.
• To evaluate the efficacy of rPPG in recognizing stress states and differentiating task difficulty.
• To establish a new multimodal database for rPPG research.

Main Methods:

• Developed a multimodal database with 56 subjects undergoing distinct tasks.
• Simultaneously recorded physiological signals using wearable contact PPG sensors and camera-based rPPG.
• Analyzed signal quality, motion artifact resilience, and applied statistical methods and machine learning for stress recognition.

Main Results:

• rPPG signals demonstrated superior quality and better motion artifact handling compared to contact PPG sensors.
• Heart-rate variability (HRV) features, specifically NNi 20 and SAMPEN, effectively differentiated between stress and non-stress states.
• Inter-beat interval (IBI), NNi 20, and SAMPEN distinguished between tasks of varying difficulty.
• Machine learning models achieved 83.11% accuracy in classifying stressed versus unstressed states.

Conclusions:

• Contactless rPPG is a viable and effective alternative to traditional contact sensors for physiological monitoring.
• rPPG-based HRV analysis holds significant potential for objective stress detection and assessment of cognitive load.
• The developed multimodal database provides a valuable resource for advancing rPPG research and applications.