Study On Classification Algorithm Of ECG Arrhythmia Based On Deep Learning

Cardiovascular disease is one of the most important diseases that threaten human life,and its prevalence and mortality rate are increasing year by year.Most cardiovascular diseases are accompanied by arrhythmias,an important cause of heart disease and sudden cardiac death.Electrocardiogram(ECG)can reflect the physiological status and working status of various parts of the heart,and is an important means and main basis for diagnosing arrhythmia diseases.The deep learning network combined with the automatic classification of ECG signals can greatly improve the ECG.The accuracy of the signal classification algorithm,which has important research significance for the diagnosis of cardiovascular disease,the main research work of this paper is divided into the following parts:(1)ECG signal preprocessingThis paper uses MIT-BIH ECG database to study.In the data pre-processing stage,the left 70 points and the right 70 points of ECG labeling points are taken as samples.In order to solve the problem of a few arrhythmia types in MIT-BIH database,the ECG signals are re-sampled and expanded to increase the sample size.(2)Feature Multi-Convergence Convolutional Neural NetworkThe confinement neural network based on automatic feature extraction and classification is selected as the research object to limit the traditional ECG arrhythmia detection classification method.Class 6 ECG arrhythmia classification based on LeNet-5 was implemented.Based on this improvement,a Feature Multiple Convergence-Convolutional Neural Network(FMC-CNN)is proposed.The effects of different feature extraction blocks and different convolution kernel sizes on classification performance are compared.The feature extraction block is used as the structural unit of the neural network to construct a multi-cascaded convolutional neural network,which can effectively replace the traditional method of artificially extracting features,and achieve the purpose of automatic extraction of features and efficient classification.The visualization of the convolutional layer is realized,and the information implicit in the feature extraction process is known.The experimental results show that the overall classification of the FMC-CNN built in this paper reaches 98.55%.(3)Feature-Multiple Cascaded Fully Connected Neural NetworkConsidering that in the actual application,the original ECG is lost,or the ECG cannot be obtained.Starting from the original ECG data,the ECG data is directly input into the multi-layer fully connected layer for network construction.In order to avoid the over-fitting problem,the multi-layer fully connected network is improved,and the random loss layer is introduced to construct the feature extraction block.A Feature Multiple Cascaded-Fully Connected Neural Network(FMC-FCNN)was built.The effects of different feature extraction blocks,stochastic inactivation layers and stochastic loss probability on classification performance are analyzed.Automatic feature extraction and automatic classification of the original ECG signals are realized.And the visualization of the middle layer is realized,and the overall accuracy of 99.21%is achieved on the MIT-BIH arrhythmia test data set.