Novel heart rate variability index for wrist-worn wearable devices subject to motion artifacts that complicate measurement of the continuous pulse interval

Objective: Wearable health monitoring devices have recently become popular, but they can still only measure the average heart rate. Heart rate variability (HRV) is known to represent changes in the autonomic nervous system and analysis of HRV has the potential to be used for monitoring various wellness-related parameters such as sleep or stress. HRV analysis requires accurate measurement of the heartbeat interval. In wearable devices, it is difficult to accurately measure the heartbeat interval due to motion noise. In this paper we propose a new method for performing HRV analysis on photoplethysmographic (PPG) signals corrupted by motion artifacts measured at the wrist. Approach: A frequency-tracking algorithm based on the oscillator-based adaptive notch filter was used to measure instantaneous heart rate. The algorithm consists of a time-varying bandpass filter for enhancing the heartbeat signal and an adaptive mechanism for tracking heart rate frequency. By optimizing the filter bandwidth and forgetting factor of the adaptive mechanism, the frequency-tracking algorithm better reflects the variability of instantaneous heart rate. The new HRV index was calculated as the standard deviation of the heartbeat interval data converted using the heart rate estimated by the frequency-tracking algorithm. In order to verify the effectiveness of the proposed index, the new HRV index calculated for each sleep stage was compared with SDNN, the standard deviation of the heartbeat interval, which was calculated using simultaneous electrocardiogram measurements. In addition, changes in SDNN and the new index were compared during a socially evaluated speech task. Finally, the relationship between the new index and SDNN was compared with the data collected during daily activities over a 24 h period. Main results: Experimental results showed that statistically significant changes in HRV could be monitored in different sleep stages using the proposed method. In addition, when subjects were stressed by a socially evaluated speech task, significant reduction in HRV was observed using the proposed method. Finally, HRV values measured during daily activities over a 24 h period showed a high correlation coefficient of 0.812 with reference HRVs. Significance: The new HRV index calculated by the proposed method is expected to be an effective new solution for noisy PPG signals.