Research On ECG Signal Classification Based On Deep Neural Network

In recent years,the incidence of cardiovascular diseases has increased year by year,and the mortality rate is relatively high.Cardiovascular disease patients are often accompanied by symptoms of arrhythmia in their early onset.Therefore,the early detection of arrhythmia is import to the early prevention of cardiovascular disease and early interventional treatment.The electrocardiogram(ECG)is the most important and basic tool for doctors to diagnose cardiovascular diseases.It can help doctors analyze the condition of heart disease patients.With the advancement and improvement of the ECG signal classification theory,helping doctors to classify the ECG signal through the medical assistant system has become a research hotspot,and the ECG signal classification has become a hot research topic.Most of the existing ECG signal classification methods use traditional methods for feature extraction,and the categories that can be classified are relatively single,and the classification effect needs to be improved.Moreover,due to the imbalance of the ECG signal,there are still great challenges to achieve effective classification of arrhythmia.Based on this,this article focuses on neural network to improve the classification effect of ECG signals and conducts classification research for imbalanced data sets.The specific research of this article is as follows:(1)This paper put forward an ECG signal classification algorithm based on the pyramid convolution structure deep residual network(PC-DRN).Among them,the pyramidal convolution(PC)layer has multiple convolution branches of different sizes,from which the corresponding features are extracted.These features spliced to obtain cascaded features,and then the cascaded features are input into a small-size convolution kernel to obtain the final fusion feature.Training the fusion features in the deep residual network can achieve better classification of ECG signals.By using the public 2017 Cardiology Challenge(Cin C2017)dataset,the algorithm achieved a better classification of 4 types of ECG data.Experimental results show that the proposed algorithm can improve the accuracy of arrhythmia classification.(2)This paper presents an ECG signal classification algorithm based on convolutional neural network(DSC-FL-CNN)with depth separable convolution and focal loss function.Among them,the focus loss function is designed to solve the imbalance problem of category.In the case of imbalanced ECG data categories,especially for the minority types of arrhythmia categories,the focus loss function can increase the weight of the minority samples to achieve improvement of the classification performance.The depth of the separable convolutional layer can reduce the number of convolutional layer parameters and help to improve the stability of model training.The experimental results show that the proposed algorithm can classify 14 types of unbalanced ECG signals in the MIT-BIH arrhythmia database better,especially in the minority arrhythmia categories.The algorithm in this paper is researched from two perspectives of ECG signals,and two different algorithms are proposed for ECG signal classification based on deep neural networks,which can get a better classification effect.