Enabling R-peak detection in wearable ECG: Combining matched filtering and Hilbert transform

Precise detection of R-peaks is a prerequisite in real-world ECG applications - this is particularly critical for wearable ECG where sensors are typically low resolution and embedded. Such recorded ECG data are typically contaminated by noise, motion artefacts, unbalanced skin-electrode impedance and other physiological signals. These affect the quality of R-peak detection and can consequently lead to failure in the evaluation of physiological functions or a misinterpretation of the state of the body, such as in monitoring stress. While numerous methods for R-peak detection are available for stationary and comparably noise-free ECG, robust DSP software for wearable devices is still emerging. To this end, a new approach which combines matched filtering and Hilbert transform is proposed. The RR-intervals and cross-correlation are used in conjunction to not only automatically locate the R-peaks but also to display the candidate ambiguous peaks via an interactive graphical user interface. The performance of the proposed approach is compared to the well-known Pan-Tompkins algorithm and is evaluated for two types of ECG databases: standard stationary data and low-SNR ECG data obtained from wearable ECG. The proposed method results in a distinctly higher positive predictivity and leads to more satisfying overall outcomes, especially for the critical call of low-SNR data.