Robust estimation of systolic time intervals using ballistocardiogram and seismocardiogram signals

Abstract

The objective of this dissertation is to Pave the way for non-invasive out-of-clinic monitoring of patients with heart failure using systolic time intervals (STIs) obtained from the ballistocardiogram (BCG) and seismocardiogram (SCG) signals. New hardware to measure BCG was explored using a high bandwidth force plate and signal processing techniques which allowed for significant improvement in absolute measurements of pre-ejection period (PEP), and measurements of changes in stroke volume, over current state-of-the-art instrumentation. Additionally, since SCG signals measure local vibrations, the relationship between sensor placement and the morphology of the signals was investigated. This was done by designing a robust algorithm that distinguishes between changes in morphology resulting from placement and changes resulting from physical activities. Consequently, the algorithm detects misplacements of the SCG sensor allowing for robust PEP monitoring in unsupervised settings. Moreover, different placements and interfaces of SCG sensors, on the upper body, were explored to identify the ideal position/ combination of positions. This showed, for the first time, that improved PEP estimates can be obtained by placements different than what is currently used in research. Finally, a universal ensemble regression model, that uses multiple features to estimate PEP from SCG signals, is presented in this work. This algorithm bypasses the lack of a well-defined standard to detect the aortic valve (AO) opening from SCG, resulting from the signal’s morphology being affected by age, sex and heart condition.