Myocardial Infarction Detection And Localization Based On Multi-Feature Branch Network

There are many clinical methods to diagnose myocardial infarction,and electrocardiography has always occupied the most important position in the field of myocardial infarction detection.With the rapid development of computer technology in recent years,the use of deep learning methods to automatically "interpret" ECGs has become more sophisticated.The ability of doctors to diagnose myocardial infarction by observing the ECG is the theoretical basis for automatic diagnosis of myocardial infarction based on multilead ECG signals.This diagnostic process can be effectively simulated by building a deep neural network.In order to make full use of the multidimensional features of ECG signals to achieve end-to-end myocardial infarction localization,the research in this paper is implemented based on a deep learning approach with the following main research elements:Firstly,for the task of denoising the ECG signal,this paper decomposes the original ECG signal in multiple levels by wavelet transform,then performs noise filtering of the ECG signal by soft thresholding method,and finally obtains the pure signal by signal reconstruction.MATLAB simulation results show that this method can effectively filter the noise such as baseline drift,myoelectric interference,and industrial frequency interference in the ECG signal.Secondly,for the waveform localization task of ECG signals,it is found that the R-wave waveform is the most obvious after the morphological changes of myocardial infarction occurrence,so the waveform localization work is mainly focused on the R-wave.The localization algorithm consists of preprocessing and double thresholding.The preprocessing operation ensures that the signal is not disturbed by irrelevant waveforms and reduces the probability of false detection,while the double thresholding setting ensures the R-wave detection performance and further reduces the probability of missed detection.According to the experiments,the localization algorithm has an average sensitivity of 99.67% on the public data set,which is better than the classical Pan-Tompkins algorithm overall.Finally,for the task of myocardial infarction detection and localization,this paper proposes a specialized domain knowledge combined with a multi-feature branching network model,which automatically detects and localizes myocardial infarction using twelve-lead ECG signals.The model extracts features at different levels from multiple perspectives of the heart,and further extracts intra-lead detail features and inter-lead implicit features based on generic features.By adding a priori knowledge from the medical field to improve the model accuracy,the accuracy of myocardial infarction localization is improved by 1.76% based on the branching model.The accuracy of the model for myocardial infarction detection in the real ECG signal dataset was 99.95%,and the accuracy of myocardial infarction localization was 99.53%.