Estimating systolic time intervals during walking using wearable ballistocardiography

Ballistocardiography (BCG), the measure of the reactionary forces of the body to cardiac ejection of blood into the vasculature, has recently re-emerged as a viable tool for estimation of mechanical parameters related to cardiovascular health in non-clinical settings. BCG measurements can be taken using a variety of sensors: some modalities require the person to be stationary, with sensors embedded in the environment that the person interacts with; others can be placed on the body, and potentially measure the signals throughout the day in naturalistic settings. The latter modality in particular can enhance the existing "quantified-self" ecosystem, providing complementary cardiovascular information regarding hemodynamics and systolic time intervals that otherwise cannot be measured currently. However, prior research in wearable BCG measurements has focused on taking measurements from stationary subjects only, thus limiting the potential impact of wearable sensing, and precluding any potential analysis of cardiovascular responses to exercise stressors. The primary technical obstacle towards wearable measurements from ambulatory subjects is that motion artifacts can render the BCG signals unreadable. In this paper, we present a framework for robust estimation of systolic time intervals from wearable BCG measurements during walking at different speeds. We demonstrate, using pattern matching techniques and multi-sensor data fusion, a good correlation value between the pre-ejection period (PEP) from the gold standard impedance cardiogram (ICG) and the BCG features and also between left ventricle ejection time (LVET) estimated from both modalities. Based on the results, we anticipate that wearable ballistocardiography can be used for continuously monitoring patients outside clinics and titrating care.