Real-time detection of peak cardiac motion signal using one-dimensional dilated convolutional neural networks

Abstract

Doppler radar as a non-contact detection of cardiac motion signals has been widely studied in recent years. However, in the present study, cardiac motion signals are highly susceptible to interference from breathing and other noises, leading to unreliable peak detection of cardiac motion signals. Considering the real-time and robustness of peak detection, this paper proposes a real-time detection algorithm of cardiac motion signals based on one-dimensional dilated convolutional neural network. This algorithm is based on dilated convolution to generate a large receptive field and extract long temporal sequence features with a low number of parameters, thereby achieving higher peak identification performance in cardiac motion signals. Especially for the cardiac movement signals that are interfered by breathing and sub-peak, our algorithm can reduce the missing recognition and false recognition caused by interference with a real-time delay of 50ms. Therefore, the algorithm proposed in this paper achieves the effect of fewer parameters, high robustness and real-time for the peak detection of cardiac motion signals.