The Ear-ECG provides a continuous Lead I like electrocardiogram (ECG) by measuring the potential difference related to heart activity with electrodes which are embedded within earphones. However, the significant increase in wearability and comfort enabled by Ear-ECG is often accompanied by a degradation in signal quality. We aim to resolve this issue by introducing a Deep Matched Filter (Deep-MF) for the highly accurate detection of R-peaks in wearable ECG, thus enhancing the utility of Ear-ECG in real-world scenarios.
Matched Filtering is the process of finding the position of a known pattern in noise through cross-correlation. This shares many similarities with the operation of convolutional neural networks (CNN). Through the ear-ECG, we have a noisy signal which contains a known pattern (the P-QRS-T waveform), and thus this is an ideal case for the application of matched filtering. To this end, we implement a Deep Matched Filter, whereby the input convolutional layer is initialised with an ECG template as weights to search for the P-QRS-T pattern in the noisy input signal. The subsequent layers of the Deep-MF then refine these matches into the true R-peak position. In this way, the input convolutional layer acts as a Matched Filter, and the subsequent layers act as an “ideal observer” of the Matched Filter output.
The proposed Deep Matched Filter is evaluated using leave-one-subject-out cross-validation over 36 subjects with an age range of 18-75, with the Deep-MF outperforming existing algorithms for R-peak detection in noisy ear-ECG. Overall, this Deep-Match framework serves as a valuable step forward for the real-world functionality of Ear-ECG and, through its interpretable operation, the acceptance of deep learning models in e-Health.