Estimation of Continuous Blood Pressure from PPG via a Federated Learning Approach

Area of Science:

• Biomedical Engineering
• Machine Learning
• Cardiovascular Health

Background:

• Ischemic heart disease is a leading global cause of mortality.
• Current methods for monitoring blood pressure, like continuous arterial blood pressure (ABP) readings, are invasive and costly.
• Electrocardiogram (ECG) and blood pressure (BP) are common but limited diagnostic tools.

Purpose of the Study:

• To develop a non-invasive machine learning framework for inferring continuous arterial blood pressure (ABP) from a single optical photoplethysmogram (PPG) sensor.
• To enable large-scale, collaborative learning across low-cost wearables using federated learning methodologies.
• To provide a cost-effective and less invasive alternative for continuous cardiovascular monitoring.

Main Methods:

• Utilized a time-series-to-time-series generative adversarial network (T2TGAN) for signal generation.
• Trained the framework using distributed models and data sources, mimicking a federated learning approach.
• Validated the framework on a previously unseen, noisy, independent dataset.

Main Results:

• The T2TGAN framework achieved high-quality continuous ABP generation from PPG signals.
• Demonstrated a mean error of 2.95 mmHg and a standard deviation of 19.33 mmHg in estimating mean arterial pressure.
• This represents the first GAN capable of continuous ABP generation from PPG using federated learning.

Conclusions:

• The developed framework offers a novel, non-invasive method for continuous ABP monitoring using PPG signals.
• Federated learning enables robust model training across distributed, low-cost wearable devices.
• This approach has significant potential for remote patient monitoring and managing cardiovascular health.