Improved Pre-Ejection Period Estimation From Ballistocardiogram and Electrocardiogram Signals by Fusing Multiple Timing Interval Features

Sympathetic nervous system (SNS) activity plays a significant role in cardiovascular control. Preejection period (PEP) is a noninvasive biomarker that reflects SNS activity. In this paper, unobtrusive estimation of PEP of the heart using ballistocardiogram (BCG) and electrocardiogram (ECG) signals is investigated. Although previous work has shown that the time intervals from ECG R-peak to BCG I and J peaks are correlated with PEP, relying on a single BCG beat can be prone to errors. An approach is proposed based on multiple regression and use of initial training data sets with a reference standard, impedance cardiography (ICG). For evaluation, healthy subjects were asked to stand on a force plate to record BCG and ECG signals. Regression coefficients were obtained using leave-one-out cross-validation and the true PEP values were obtained using ECG and ICG. Regression coefficients were averaged over two different recordings from the same subjects. The estimation performance was evaluated based on the data, via leave-one-out cross-validation. Multiple regression is shown to reduce the mean absolute error and the root mean square error, and has a reduced confidence interval compared with the models based on only a single feature. This paper shows that the fusion of multiple timing intervals can be useful for improved PEP estimation.