Fair non-contact blood pressure estimation using imaging photoplethysmography

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare
• Cardiovascular Monitoring

Background:

• Hypertension diagnosis often requires specialized equipment, complicating long-term monitoring.
• Existing non-contact blood pressure estimation methods have limitations in dataset diversity and applicability.
• There is a significant need for portable, non-contact solutions for continuous blood pressure assessment.

Purpose of the Study:

• To develop a practical and accurate non-contact blood pressure estimation method.
• To create a system applicable across diverse demographics (age, ethnicity, skin tone) and real-life scenarios.
• To improve upon the accuracy and generalizability of current non-contact blood pressure estimation techniques.

Main Methods:

• Facial videos were used to extract the imaging photoplethysmogram (IPPG) signal.
• A four-layer filtering process was applied to refine the IPPG signal for pulse wave analysis.
• A hybrid deep learning model (CNN+BiLSTM+GRU) was developed for blood pressure prediction.

Main Results:

• The proposed method demonstrated improved accuracy in estimating blood pressure compared to existing approaches.
• The mean absolute error (MAE) for systolic blood pressure (SBP) was reduced by 13.6%.
• The MAE for diastolic blood pressure (DBP) was reduced by 16.4%.

Conclusions:

• The developed non-contact method offers a promising advancement for hypertension monitoring.
• The CNN+BiLSTM+GRU model enhances the accuracy and applicability of facial-based blood pressure estimation.
• This technology has the potential to facilitate convenient, long-term blood pressure monitoring in diverse populations and settings.