Research On Intelligent Detection And Localization Of Myocardial Infarction With ECG Based On Deep Learning

The change of people’s lifestyle and the development of the aging population have increased the morbidity and mortality of cardiovascular diseases,which has become the first killer of national health.According to the 2018 report on cardiovascular disease in China,there are about 2.5 million patients with myocardial infarction in our country.On average,it increased at one patient every 3 seconds,and only 5% of them have received timely and effective treatment.Within the golden treatment time after the occurrence of myocardial infarction,the sooner patient is diagnosed and treated,the smaller infarct area and lower fatality rate.Early diagnosis of myocardial infarction is of great significance for its treatment and prognosis.In recent years,early intelligent detection and rapid and accurate localization of myocardial infarction have become hotspots in the field of cardiovascular disease.At present,there are some problems in the early diagnosis algorithm of myocardial infarction,such as poor generalization ability of manual design features,inconformity of feature classification and positioning accuracy to clinical requirements.In this thesis,the deep learning method is adopted to automatically acquire the key features of disease diagnosis from ECG signals,so as to realize the intelligent detection and localization of myocardial infarction.The specific research contents are as follows:(1)Aiming at the detection problem of myocardial infarction,a single-lead myocardial infarction detection algorithm based on densely connected convolutional networks is proposed,under the background of massive electrocardiogram in telemedicine.A densely connected convolutional network is constructed by stacking multiple dense blocks.The richer description and discrimination of ECG features are enhanced by the tight connection between dense blocks,so that the network can automatically learn the effective features with strong robustness and significant discrimination,so as to achieve rapid and accurate intelligent detection of myocardial infarction.For myocardial infarction detection,the proposed method realizes an accuracy of 99.79%.And in multiple groups with different signal noise ratio,the accuracy of myocardial infarction detection also exceeds 97.88%,which fully proved the significance of the algorithm in this thesis.(2)Considering the priori guidance of medical information,a multi-lead myocardial infarction localization algorithm based on network compression and densely connected convolutional networks is proposed,aiming at the localization problem of myocardial infarction.The network compression technology is introduced to optimize and simplify the myocardial infarction localization model,because of the large dimensions of 12-lead data and information redundancy of intra-lead signal.The compact and efficient model is obtained by channel-level sparse regular training and pruning that effectively reduces the channel dimension,and fine-tuning the pruned network.For myocardial infarction localization,the experiment result proves that the proposed method has achieved high-precision recognition with accuracy,sensitivity and specificity of 99.92%,99.90% and 100%,respectively.At the same time,the model parameters and calculation operations are significantly reduced,which fully proved the effectiveness of the algorithm in this thesis.