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

Abstract

Highlights What are the main findings? A personalized calibration and hybrid feature fusion framework was developed for continuous BP estimation. The proposed method integrates handcrafted physiological features, self-supervised waveform representations, and subject-specific prior BP information. What are the implications of the main findings? The integration of physiological interpretability and self-supervised learning improves BP estimation under small-sample conditions. This framework shows potential for wearable BP monitoring, although further validation is required before clinical deployment. Abstract Continuous and noninvasive blood pressure (BP) monitoring is important for hypertension screening and long-term cardiovascular management. Continuous BP estimation using photoplethysmography (PPG) and electrocardiography (ECG) has become a promising alternative to conventional cuff-based measurements. However, handcrafted feature-based methods are sensitive to fiducial point detection, while end-to-end deep models often require large labeled datasets and may ignore subject-specific BP baselines. This study proposes a personalized calibration and hybrid feature fusion framework for continuous BP estimation from ECG and PPG signals. The framework integrates physiologically interpretable handcrafted features, self-supervised waveform representations, and subject-specific prior BP information to predict systolic and diastolic BP. The handcrafted branch extracts pulse transit time, heart rate variability, and PPG morphological descriptors, whereas a masked reconstruction-based self-supervised encoder learns latent waveform embeddings without BP labels. Personalized calibration incorporates base BP from the earliest calibration file through an alpha-weighted strategy, and the fused representation is fed into XGBoost regressors. The framework was evaluated on a private dataset and the BP-UCI dataset, achieving SBP/DBP MAEs of 7.25/3.98 mmHg and 7.85/4.02 mmHg, respectively. Comparative and ablation results indicate improved baseline performance in the current experimental setting, suggesting the methodological feasibility of the proposed framework for small-sample continuous BP estimation. Further validation with larger and more diverse cohorts is still required.