Accelerometer body sensor network improves systolic time interval assessment with wearable ballistocardiography

Abstract

Systolic time intervals (STI) are non-invasive measures of cardiac function. Due to the fact that STI can be measured noninvasively outside the clinic, STI are a promising method for long-term monitoring of patients with cardiovascular disease. In particular, the pre-ejection period (PEP) has been measured successfully from body vibrations of the beating heart, a technique called ballistocardiography (BCG), using a weighing scale. Similar measurements can be made with on-body accelerometers, however these wearable BCG signals are typically more challenging to interpret than whole-body BCG. In this paper, we conducted a small pilot study with four subjects to investigate whether a body sensor network of four accelerometers positioned on the wrist, arm, sternum, and head could improve beat-by-beat PEP prediction beyond that of each sensor alone. Linear models were fitted from the R-J and R-I intervals of the four BCG signals to PEP measured with impedance cardiography from 5-minute recordings after isometric lower-body exercise. Specifically, we found that (i) the RMS error of PEP estimation from the wearable BCG sensors can be reduced by using double integration, (ii) the standard deviation of PEP estimates from R-I intervals was smaller than from R-J intervals, and (iii) linear models combining both R-J and R-I measurements from all sensors resulted in the best average correlation (r(2) = 0.96 ± 0.01) and lowest average root mean square error (2.5 ± 0.8 ms) from 5×2-fold cross validation.