Quality of heart rate variability features obtained from ballistocardiograms

Abstract

Heartbeat intervals (HBIs) vary over time, and that variance can be quantified as heart rate variability (HRV). HRV has several health-related applications including long-term health monitoring and sleep quality assessment.

The focus of this research is obtaining HRV from ballistocardiograms (BCGs), force signals caused by micro-movements of the human body in response to blood ejections. This method of HRV estimation is attractive because it does not require direct attachment of any sensor to the body. However, the HBIs and corresponding HRV measured with BCGs are different than those obtained via electrocardiograms (ECGs), signals obtained by attaching electrodes to the body to detect electrical heart activity. Because ECG-based HRV is typically considered ground truth, differences in BCG-based versus ECG-based parameters are referred to as HBI and HRV errors.

This research investigates the effects of HBI error on HRV feature quality. While a few studies have used BCG-based HBIs to estimate HRV features for sleep staging, the effects of HBI error on the quality of the resulting HRV features seem to have been overlooked. As a result, an acceptable HBI error range has not been defined. One contribution of this work is the development of such an acceptable error range.

This dissertation work (i) develops a hardware and software system necessary to record BCGs and to perform BCG peak detection to obtain HBIs with the least possible error, (ii) determines an allowable range for HBI error by studying the effects of this error on HRV quality in the context of HRV-based sleep staging, and (iii) compares the determined acceptable HBI error range to the HBI error of our final system. The inherent error in BCG-based HBI determination due to physiological and platform effects is also taken into account in this comparison.

A minimum HBI error of 20 ms was obtained from the system developed in (i), and the allowable error range was determined to be 30 ms based on the investigations conducted in (ii). The combined physiological and platform effects led to an error of 8.8 ms on average. Based on the comparisons conducted in (iii), the developed system is suitable for long-term sleep quality assessment. In addition, the effects of the HBI errors introduced by this system on the resulting HRV features are negligible in the sleep staging context.