Personalized Calibration and Hybrid Feature Fusion for Continuous Blood Pressure Estimation Using PPG and ECG Signals

Area of Science:

• Biomedical Engineering
• Cardiovascular Technology
• Signal Processing

Background:

• Continuous blood pressure (BP) monitoring is crucial for managing hypertension and cardiovascular health.
• Photoplethysmography (PPG) and electrocardiography (ECG) offer noninvasive BP estimation alternatives to traditional cuff methods.
• Existing methods face challenges like sensitivity to feature detection and large data requirements.

Purpose of the Study:

• To develop a personalized calibration and hybrid feature fusion framework for continuous BP estimation.
• To integrate handcrafted features, self-supervised learning, and subject-specific BP data.
• To enhance the accuracy and applicability of noninvasive BP monitoring, particularly in small-sample scenarios.

Main Methods:

• A hybrid framework combining handcrafted features (pulse transit time, heart rate variability, PPG morphology) and self-supervised waveform representations.
• A masked reconstruction-based self-supervised encoder for learning latent waveform embeddings.
• Personalized calibration using an alpha-weighted strategy with initial BP data, feeding into XGBoost regressors.

Main Results:

• Achieved Mean Absolute Errors (MAEs) of 7.25/3.98 mmHg for systolic/diastolic BP on a private dataset.
• Attained MAEs of 7.85/4.02 mmHg for SBP/DBP on the BP-UCI dataset.
• Demonstrated improved baseline performance and methodological feasibility for small-sample continuous BP estimation.

Conclusions:

• The proposed framework offers a feasible approach for personalized continuous BP estimation from ECG and PPG signals.
• Integration of diverse features and personalized calibration enhances BP monitoring accuracy.
• Further validation on larger, diverse cohorts is recommended to confirm generalizability.