Analysis and refinement of pulse rate variability

Loading...
Thumbnail Image

Files

Publication or External Link

Date

2019

Authors

Advisor

Citation

Abstract

Heart rate variability (HRV), calculated from the cardiac intervals of electrocardiogram (ECG), is a promising marker of the cardiovascular system status and fitness. However, ECG signal is not always available and photoplethysmogram (PPG) is easier to obtain, and more widely used in clinical is running HRV analysis on pulse-to-pulse intervals of PPG signal, which is usually referred to as pulse rate variability (PRV). Thus, whether PRV can be used as a substitution of HRV is of substantial interest to researchers.

In this thesis, two issues about PRV are discussed. The first issue is the selection of characteristic point, which determines the length and location of the pulse-to-pulse interval and will affect the agreement between PRV and HRV. Six characteristic points of PPG pulse are extracted and the agreement between HRV and corresponding PRV is calculated and compared, in two situations, subjects with cardiovascular diseases (CVD) and subjects without cardiovascular diseases (non-CVD). The result indicates that pulse peak is most suitable for CVD subjects, and 50% max amplitude point and 75% max amplitude point on pulse slope are most suitable for non-CVD subjects.

The second issue studied in this thesis is the PRV refinement using arterial blood pressure (ABP) information. The relationship between systolic blood pressure extracted from ABP signal and pulse transit time (PTT) is modeled using linear kernel support vector regression (SVR) and RBF kernel SVR, respectively. Estimated PTT is used to adjust the location of PPG pulse-to-pulse intervals. PRV after adjustment is calculated, and its agreement to HRV is compared with the original PRV. For CVD subjects, the improvement to the agreement is limited, and only the agreement for variables representing long-term variability is improved. For non-CVD subjects, there is a relatively large improvement for approximately all variables after refinement and linear kernel outperforms RBF kernel in this situation.

Notes

Rights