RPAA

Abstract

Detection of R waves from ECG signals is of great importance yet challenging for the diagnosis of cardiovascular diseases due to the noise. In this paper, an anti-noise R Peak Annotation Algorithm (RPAA) is proposed. In RPAA, the detection of the R peak is firstly transformed into a distance optimization problem, with the goal of learning the distance between all data points in the ECG recording and the nearest R peak. Then we design a U-Net-based neural network which is a set of symmetrical encoder and decoder to predict the distance of each point. The encoder extracts the deep feature representation of the signal through down-sampling. And the decoder fuses the encoder's same-dimensional features and performs up-sampling to predict the distance between each data point and the nearest R peak. Four parallel convolutions are employed to extract features at different scales, and the data flows across layers through a short-cut connected residual structure. The Squeeze-and-Excitation module is incorporated to strengthen the features extracted by the previous layer to improve the performance of the annotation algorithm. For the detection of R peaks in abnormal ECG signals with high noise, the RPAA annotation algorithm obtains a precision rate of 99.56%, a recall rate of 98.29%, and a F1 score of 0.9892. For the detection of R peaks from ECG signals with a signal-to-noise ratio of 0, the RPAA annotation algorithm has a F1 score of 0.8946. Experiments conducted on cross-database also verify that the RPAA algorithm has a high generalization ability.