Improving the Accuracy of Volume-Compensation-Based Continuous Blood Pressure Measurement via Windkessel-Guided Multidomain Fusion

Abstract

Hypertension is a major risk factor for cardiovascular diseases, and accurate and continuous blood pressure monitoring is crucial for early warning and personalized intervention. This study proposes a continuous noninvasive blood pressure monitoring method based on the Windkessel model and multifeature-domain fusion. A finger-cuff device was used to acquire photoplethysmography (PPG) signals, from which multidimensional features, including time-domain, time–frequency, and reflection-related features, were extracted to construct a high-dimensional feature space. Gradient boosting regression trees (GBRTs) and a ResNet50 deep network were employed to achieve nonlinear feature mapping and fusion, while the XGBoost algorithm was used to adaptively estimate the peripheral resistance and arterial compliance of the Windkessel model, thereby enabling individualized prediction of systolic blood pressure (SBP) and diastolic blood pressure (DBP). The effectiveness of the proposed method was validated on a dataset collected from 38 healthy subjects. The multifeature-domain fusion model achieved mean absolute errors (MAEs) of 2.0–1.8 mmHg for SBP and DBP prediction, with coefficients of determination ( ${\mathrm {R}}^{2}$ ) reaching 0.96 for both. Under the current controlled experimental conditions, the results met the numerical Grade A thresholds of the British Hypertension Society (BHS) grading table. Compared with single-feature-domain models, the proposed multifeature-domain fusion method reduced the MAE by approximately 42 %–46 % relative to the time-domain-feature baseline. The introduction of the Windkessel model strengthened the relationship between the predicted results and hemodynamic parameters, and improved the stability of model outputs under the current controlled experimental conditions. The results indicate that the combination of data-driven methods and hemodynamic modeling provides an effective technical pathway for long-term wearable cuffless blood pressure monitoring and has important potential for clinical application.