Research On ECG Signal Denoising And Heartbeat Retrieval Algorithm Based On Deep Learning

According to the Chinese cardiovascular report in 2017,the number of patients with cardiovascular disease in China is 290 million.With the death rate accounting for more than40% of the deaths,and the prevalence and mortality of cardiovascular disease are still on the rise.Electrocardiogram has become an effective method to diagnose cardiovascular diseases as a non-invasive way of human heart electrical activity.ECG is a weak electrical signal.The signal is highly susceptible to noise interference in the process of signal monitoring,especially during dynamic ECG monitoring.How to effectively filter out noise and retain useful signals to the maximum extent has become the key link in ECG processing.At present,the diagnosis of cardiovascular diseases mainly depends on doctors' discrimination of ECG.The long-term ECG monitoring,especially the rapid development of telemedicine will bring a huge amount of ECG data.It has become an important research field of ECG data intelligent analysis to assist doctors to find the ECG interval of interest in the massive data.In the background of research,the method of deep learning is introduced to study the denoising and heartbeat retrieval algorithm of ECG signals.The main contents of work are as follows:(1)Aiming at the problem of ECG signal denoise.An ECG signal denoising algorithm based on lossless constrained denoising automatic encoder is proposed in this paper.The deep features of the ECG signal are extracted by stacking multiple lossless constrained denoise automatic encoders.Based on the correlation between the ECG signal time series,the network training optimization network parameters are constructed.ECG signal is denoised through the trained network.The denoising algorithm proposed in this paper has been applied to remove three kinds of noise: electrode interference,electromyography interference and baseline drift.The experimental results show that the SNR obtained is all above 19.19 dB,17.27 dB and17.80 dB respectively,which proves that the denoising algorithm proposed in this paper can achieve the denoising of ECG signals well while retaining the information of ECG signals.(2)Aiming at the problem of ECG signal retrieval.An improved dynamic time warping retrieval algorithm based on total Bregman Divergence is proposed in this paper.The ECG signal retrieval has been realized by calculating the similarity of signal amplitude and morphological features.The total Bregman Divergence is used to calculate the similaritybetween signal sequences,which makes full use of its stability.Then the dynamic time warping is used to select the optimal matching path.The improved algorithm can not only calculate the ECG signals with mismatched feature points,but also effectively resist the influence of noise,and has good robustness.The experimental results demonstrate that the proposed algorithm can effectively retrieve ECG signals of arbitrary length without extracting feature waves,and the retrieval accuracy is above 90.91%.