A Physiological-Model-Based Neural Network Framework for Blood Pressure Estimation from Photoplethysmography Signals

Abstract

Continuous blood pressure (BP) estimation via photoplethysmography (PPG) remains a significant challenge, particularly in providing comprehensive cardiovascular insights for hypertensive complications. This study presents a novel physiological model-based neural network (PMB-NN) framework for BP estimation from PPG signals, incorporating the identification of total peripheral resistance (TPR) and arterial compliance (AC) to enhance physiological interpretability. Preliminary experimental results, obtained from a single healthy participant under varying activity intensities, demonstrated promising accuracy, with a median standard deviation of 6.88 mmHg for systolic BP and 3.72 mmHg for diastolic BP. The median error for TPR and AC was 0.048 mmHg*s/ml and -0.521 ml/mmHg, respectively. Consistent with expectations, both estimated TPR and AC exhibited a reduction as activity intensity increased.