Feasibility of Wearable Armband Bipolar ECG Lead-1 for Long-term HRV Monitoring Using a Combined Signal Averaging and 2-stage Wavelet Denoising Technique

Omar Escalona, Sophie Magwood, Anna Hilton, Niamh McCallan
Ulster University


Introduction Heart rate variability (HRV) is a clinically important and prominent cardiovascular diseases diagnostic factor. Since HRV is a highly individualised measure, long-term continuous ECG and HRV tracking using a non-invasive armband-based wearable monitoring device is an appealing option for HRV trend-based indicator of general health. Therefore, we investigated the correlation between the bipolar arm-ECG Lead-1 (electrodes axis coplanar to chest and at axilla level) HRV measurements and their corresponding standard measurements from the Lead-I ECG (chest). Advanced ECG denoising techniques are required to enable this.

Methods An initial clinical-study with 10 subjects selected as having similar upper-arm circumference (28.5cm±2.5%). HRV metrics were measured independently (standalone basis) on the bipolar arm Lead-1, after a novel 2-stage DB4-wavelet-based denoising process supported by a rolling signal-averaged-ECG optimal-thresholding adaptation algorithm, and correlated with same HRV metrics values measured on the standard chest ECG Lead-I, using the conventional Pearson correlation coefficient. Four clinically common HRV time-domain metrics are: SDNN, RR-rms, RR-median and the interquartile-range value of normal-to-normal heartbeat intervals (IQRNN). These HRV metrics were measured on 8-minute-long ECGs. The conventional Pan-Tompkins algorithm was implemented autonomously and independently from standard chest Lead-I and arm-ECG Lead-1 for QRS-detection.

Results The Pearson correlation between the arm-HRV-metrics measured values and HRV-metrics measured from the standard chest Lead-I results were: p=0.789, p=0.995, p=0.991 and p=0.940, and linear regression model coefficients of determination values (from scatter-plot of arm-Lead-1 versus chest-Lead-I HRV values data-point per-subject) of: R²=0.623, R²=0.991, R²=0.982 and R²=0.884, for SDNN, RR-rms, RR-median and IQRNN respectively, in the 10-subject study.

Conclusion Arm-ECG (Lead-1) HRV long-term monitoring on a standalone basis is a feasible approach, using conventional Pan-Tompkins QRS-detection algorithms and an advanced wavelet-based denoising processes. RR-rms and RR-median HRV metrics from bipolar arm-ECG closely correlated to the values measured from the standard Lead-I and present potential for clinical use.