Spectrum Filtering to Extract Pulse Rate Variability from Signals Recorded by Wearable Devices

Photoplethysmography (PPG) is increasingly becoming a modern alternative to heart rate monitoring, gaining popularity due to its mobility and implementation in wearable devices. Its sensitivity to artifacts poses a challenge to algorithms designed to extract pulse rate variability (PRV) from the PPG signal, thereby introducing significant errors, which has led to research aimed at minimizing artifacts in the PPG. The aim of this study is to propose a method reducing the artifacts affecting the PPG to improve the efficiency of PRV extraction using the PPG signal spectrum filtering together with the non-physiological values correction and autoregressive modeling. The proposed method was tested on signals from laboratory experiments (51 cases). The obtained PRVs were compared to the reference heart rate variabilities (HRVs) extracted from simultaneous electrocardiograms in terms of the mean relative root mean square error (mRRMSE) and the nonparametric Wilcoxon signed-rank test. The spectrum filtering algorithm reduced the medians of mRRMSE in almost all experiments. Nevertheless, no statistically significant difference was observed in the mRRMSEs for spectrum filtering vs. postprocessing only. However, after excluding 3 subjects whose RRMSEs increased threefold after spectrum filtration, a statistically significant difference between these effects of PPG processing was shown. This confirms the validity of using this method to reduce the effects of artifacts when extracting PRV from the PPG.