Estimating Heart Rate Variability in Challenging Low SNR Regimes Using Wearable Magnetocardiography Sensors

Abstract

We report Beat Estimation, a novel method used to calculate Heart Rate Variability (HRV) from low Signal to Noise Ratio (SNR) data (−7 dB to −4 dB in this work) acquired via wearable magnetocardiography (MCG). MCG activity is first collected using an in-house wearable sensor and filtered to remove noise outside the band of interest. Beat Estimation extracts a single heart beat from the filtered recording and correlates it with a small number of beats individually to average out the remaining noise. The de-noised beat is then correlated with the full recording to identify the location of each of the heart beats. Using these locations, HRV parameters are, finally, calculated. Results show $\sim$99.9% accuracy in estimating HRV metrics using beat-to-beat intervals as opposed to traditional R-to-R-peak intervals. The average accuracy of detecting the true location of beats is shown to increase to 96.43% using Beat Estimation as opposed to 59.98% using our previous method that relied on R-peak detection. In summary, Beat Estimation renders wearable MCG sensors capable of accurately estimating HRV, despite the low SNR levels associated with sensor operation. The approach can be game-changing in assessing heart health, cardiovascular fitness, stress levels, cognitive workload, and more.