Research On Classification Algorithm Of Myocardial Infarction Based On ECG Signal

Cardiovascular disease is one of the major threats to human health.The high morbidity and mortality of myocardial infarction have become the number one enemy of human health.After the occurrence of myocardial infarction,most patients will have characteristic changes in ECG(Electrocardiogram),showing a unique evolution rule of electrocardiogram.Despite the advent of a number of new diagnostic techniques in recent years,electrocardiogram examination still occupies the most important position in the diagnosis of myocardial infarction due to its non-invasive,convenient and low-cost characteristics.This article focuses on three key issues:denoising ECG signals,waveform detection,and classification of myocardial infarction.It aims to establish a set of deep learning models focusing on myocardial infarction.Including ECG signal automatic denoising,waveform detection,classification model,Provide technical support for automatic diagnosis of cardiovascular diseases.There are mainly the following aspects:(1)Denoising and waveform detection stage;this paper takes the classic U-net model in two-dimensional image processing as the research basis,fully considers the characteristics of ECG signals,and innovatively proposes a one-dimensional ECG-U-net model.The models were applied to the denoising and waveform positioning of ECG signals,respectively,and all achieved good results.Research on denoising of ECG signals;The clean ECG data used in this article comes from the clinical database of Suzhou kowloon Hospital,and the noise signal comes from the MIT-BIH Noise Stress Test Database.The signal-to-noise ratio SNR and mean square error RMSE are used as evaluation indicators.The method proposed in this paper is compared with the soft threshold wavelet transform method and the FCN model that belongs to the convolutional network class.It is proved that the one-dimensional ECG-U-net model proposed in this paper can remove the three kinds of noises of baseline wander,electrode motion,and muscle artifact while also retaining the detailed characteristics of the waveform.For ECG waveform detection;through special label design and post-processing of network output,the ECG-U-net model can be used to locate the start and end points of the three main bands of the EC signal(P wave,QRS wave,and ST-T band)under a single lead.The three waveform labels used in the experiment were labeled by volunteers and confirmed by the ECG doctor.Verified by the clinical data of Suzhou kowloon Hospital,the method proposed in this paper can achieve good results under the conditions of different leads and different diseases.(2)The stage of classification of myocardial infarction;According to the characteristics of myocardial infarction disease,the ECG classification in the case of 12 leads was studied.The multi-lead multi-scale convolutional neural network model MM-ECG-CNN is proposed,the model combines the position information of the 12-lead ECG and the characteristic waveform,and uses two-dimensional convolution and one-dimensional convolution.Use the clinical database of Suzhou Kowloon Hospital for testing,the model has an accuracy rate of 92.62%on the 7 common ECG signals,and an accuracy rate of 80.59%on the classification of the location of the 5 types of myocardial infarction,which can initially achieve the classification of the infarction location.Compared with the deep learning models proposed by other researchers,the MM-ECG-CNN model proposed in this paper achieves the best results under the comparison of various situations,and has good clinical application potential.The research results of this paper have achieved good results in the three aspects of ECG signal denoising,waveform positioning and classification,and have good clinical application potential.