Key Feature Selection and Model Analysis for Blood Pressure Estimation From Electrocardiogram, Ballistocardiogram and Photoplethysmogram

Continuous cuff-less blood pressure (BP) monitoring has become a research hotspot in recent years. Researches have studied the impact of pulse transit time (PTT) and ballistocardiogram (BCG) signals on BP. However, the accuracy of these methods are not high enough to put them into practice on a large scale. In this paper, we propose a new BP estimation model which combines features extracted from electrocardiogram (ECG), BCG and photoplethysmogram (PPG). We calculate several features containing amplitude, time and energy from these three signals and use stepwise regression to select key ones for this combination model. The combination model was examined in 20 young healthy subjects and it presents good results: a correlation coefficient (R) of 0.84 (systolic blood pressure, SBP) and 0.7 (diastolic blood pressure, DBP), a root-mean-squared error (RMSE) of 8.16 mmHg (SBP) and 6.63 mmHg (DBP), and a mean absolute error (MAE) of 6.84 mmHg (SBP) and 5.46 mmHg (SBP). Besides, The PTT-based BP estimation model and BCG-based estimation model are also established in this paper. The comparison of these three models shows that the PPG-ECG-BCG-based model has better performance