Research And Implementation Of Arrhythmia Classification Based On Deep Learning

Arrhythmia is a common cardiovascular disease.Cardiovascular disease is the disease with the highest mortality rate in the world.The number of people who die from cardiovascular disease every year is increasing.Arrhythmia is not always an irregular heart activity.It can occur in a healthy heart and can cause serious problems such as stroke or sudden cardiac death.Arrhythmias are prone to other conditions,such as heart failure,angina pectoris,and systemic vascular embolism.Therefore,the automatic detection and classification of arrhythmias is critical in clinical cardiology.Traditional machine learning methods rely on the artificial design and selection of ECG signal features,which limits the classification accuracy of the model.Deep learning integrates traditional feature extraction and classification,and can automatically extract feature values and learn feature representations.Deep learning has become a research hotspot to automatically identify ECG signals.In order to realize the automatic recognition of different types of arrhythmia signals,this paper proposes a deep learning-based arrhythmia classification method.In this paper,ResNet and DenseNet are used to classify and identify ECG signals,and a compression-excitation(SE)module is embedded to improve the classification accuracy of the model and replace the standard convolution with Depth separable convolution to reduce the amount of parameters and calculations of the model.The research content of this article mainly includes the following parts:1.Aiming at the noises such as power frequency interference,baseline drift and EMG interference in the original ECG signal,this paper uses a combination of 50 Hz FIR notch filter,median filter and Butterworth low-pass filter.Filter out and get high-quality ECG signals that are good for neural network classification.2.The compression-excitation(SE)module and deep separable convolution were applied to the residual network(ResNet)and densely connected network(DenseNet).The SE-ResNet-16 model and the SE-DenseNet model were proposed,which slightly improved the correctness of the network.Rate,and reduces the amount of calculation and parameters of the model.3.The arrhythmia databases of MIT and Boston Beth Israel Hospital(MIT-BIH)were used to verify the SE-ResNet-16 model and the SE-DenseNet model.The experimental results proved the effectiveness of the method proposed in this paper and classified them as arrhythmias.Diagnostics provide a new approach and validate the effectiveness of SE modules and deep separable convolutions in ResNet and DenseNet.It also shows that compared with ResNet,dense connections in DenseNet can effectively utilize features and reduce the amount of parameters.The SE-ResNet-16 model and the SE-DenseNet model can automatically learn the deep features of the input ECG signal,obtain higher classification accuracy,and reduce the network running time,which has certain positive significance for related clinical applications.