Deep Regression Network With Sequential Constraint for Wearable ECG Characteristic Point Location
Zuo Wang, Jinliang Wang, Mingyang Chen, Wei Yang, Rong Fu
Abstract
Accurate location of characteristic points in wearable ECG signals may be a challenge due to the high noise. Taking the time sequence of waveforms and missing waveforms into account, we design a location regression network ECG_SCRNet, combined with the sequential constraints to accurately identify characteristic points of wearable ECGs. We add a classification head to determine whether there is a P-wave or a T-wave missing. This architecture ensures that the network considers both the time sequence of physiological waveform and class information to improve the accuracy in locating characteristic points. The proposed ECG_SCRNet was evaluated on a wearable dataset and the LUDB, achieving highly accurate results compared to other state-of-the-art methods. On the wearable dataset, the average <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Sen, PPV</i> and <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">F1</i> score are 97.13%, 99.96%, and 99.51%. On the LUDB, the average <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Sen, PPV</i> and <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">F1</i> score are 96.86%, 99.83%, and 98.97%. These results demonstrate that the proposed ECG_SCRNet has good flexibility and reliability when applied to signal characteristic point detection, and it is a reliable method for analyzing ECG signals in real time.