Research On Methods Of QRS Complexes Detection And Classification Of ECG Signals Based On Deep Learning

Cardiovascular disease is the number one disease that endangers the health of our people.The number of patients has exceeded 300 million.The automatic electrocardiogram(ECG)analysis technology can greatly improve the efficiency of the examination and diagnosis of cardiovascular diseases,and has significant application value and research significance.QRS complexes detection is the most important step in the automatic electrocardiogram analysis technology.The existing QRS complexes detection and classification methods are sensitive to waveform distortion,are susceptible to noise interference,and rely heavily on the selection of empirical parameters,and the detection accuracy and efficiency are limited.In view of the above problems,the main research contents of this article are as follows:(1)ECG signal preprocessing.Firstly,apply two median filters and a low-pass filter to remove the baseline drift,power frequency interference and electromyography(EMG)noise of the ECG signal.Then,for QRS complexes detection,use binary coding to identify QRS complexes periods and non-QRS complexes periods,and cut the ECG signal every 8seconds to reduce the memory overhead of deep neural network training.Finally,for the heartbeat classification,all R peaks of ECG signal are traversed.Taking the current R peak as the reference point,the ECG signal between the 50 th sampling point after the previous R peak as the starting point and the 100 th sampling point after the current R peak as the ending point is cut as the heartbeat signal,and it is resampled as the heartbeat signal with a length of128 sampling points.(2)QRS complexes detection combining prior knowledge and deep neural network.Firstly,this paper uses the traditional Pan-Tomkins method to initially detect the R peak,and combines the average QR and RS interval to achieve the preliminary positioning of the QRS complexes period.Then,the deep neural networks which include Conv-Tas Net and U-net are used to fine-tune the preliminary positioning results and determine the QRS complexes period.Finally,perform peak detection in each QRS complexes period to relocate the R peak.On the one hand,this method can avoid the heavy dependence of traditional methods on empirical parameters.on the other hand,it can compress the scale of network parameters and reduce the computational load while maintaining the accuracy of model detection.The experimental results on the QT database verify the superiority of the proposed method.(3)Heartbeat classification combining QRS complexes detection and morphological features.Firstly,according to the results of QRS complexes detection,the QRS complexes period and R-R interval sequence are calculated as empirical features.Then,the empirical features and the pre-processed ECG waveform are fused and input into Res Net to classify the heartbeat.This method combines empirical features with good interpretability with deep neural networks,which can effectively improve the generalization ability of the network.The experimental results on the MIT-BIH arrhythmia database verify the superiority of the proposed method.