A Research Of Arrhythmia Classification Algorithm Based On Deep Residual Shrinkage Network

Electrocardiogram is an important reference for diagnosing and preventing serious cardiovascular diseases such as arrhythmia and myocardial infarction.By observing the changes in the waveform of ECG,the relevant cardiovascular diseases can be diagnosed in conjunction with the clinic,and Chinese medical resources are tight and the volume of ECG data to be diagnosed is huge.With the growing maturity of research in combining deep learning algorithms and medical data applications,the algorithmic model of deep learning can classify and analyse medical data such as ECG signals after pre-processing,which is an auxiliary diagnostic method that can reduce the pressure on doctors to diagnose.Aiming at the arrhythmia classification of deep learning algorithm model,this paper studies the model based on deep residual shrinkage network,and the MIT-BIH dataset as the data base to pre-process and detect and classify arrhythmia ECG signals and compare and analyse with other classical networks,mainly in the following three aspects.(1)Building a one-dimensional modified residual shrinkage networks model for the classification of arrhythmia ECG signals.The model of the residual shrinkage network is improved in order to achieve better classification results.By reducing the two-dimensional convolutional layer in the residual shrinkage network model to a one-dimensional convolutional layer,the computational effort of the network is reduced to fit the one-dimensional ECG signal input.The computational layers of the attention mechanism sub-network are adjusted to use a more appropriate pooling layer and activation function for arrhythmia ECG signals.The residual shrinkage network model uses a larger number of filter feature channels to enhance the network’s ability to extract data features.(2)Algorithmic model based on a one-dimensional modified residual shrinkage network to accomplish the classification and train of arrhythmia ECG signals.The residual shrinkage network combine attention mechanism and soft thresholding sub-network,which is a neural network that can filter noise.The acquisition process of ECG is prone to fusion of many other noise disturbances,industrial frequency disturbances,inotropic disturbances,etc.A neural network with better noise immunity is more capable of identifying arrhythmic ECG signals.Based on the MIT-BIH dataset,pre-processed ECG signals were input for detection and classification.The pre-processed ECG signals were not subjected to any denoising process,and the most original signal features were retained.By designing the model and parameter tuning,the network model completed the effective classification of the four types of heart beats.(3)The improved network model is studied for the analysis of experimental results.The evaluation metrics of the network model classifier are introduced and two other networks,convolutional neural network as well as residual network,are built for experimental comparison.The comparative analysis study is carried out in terms of accuracy,sensitivity,F1 score,confusion matrix,network loss and accuracy comparison curve.The one-dimensional modified residual shrinkage network achieves 99.51% classification accuracy and 97.96% F1 score without data denoising,and the experimental technical indicators of the network noise resistance show better stability and better noise resistance performance.