Research On Intelligent Detection And Localization Of Myocardial Infarction

Myocardial infarction is a typical cardiovascular disease caused by obstruction of the coronary arteries.These blockages can cause a lack of blood flow to the heart muscle.The lack of oxygenated blood can cause damage to occur to the heart muscle and can lead to a heart attack.The electrocardiogram is an inexpensive,non-invasive,and convenient tool to check the activity of the heart and has been used extensively in the prevention and diagnosis of myocardial infarction.Professional doctors can detect and treat myocardial infarction in time through electrocardiogram.When dealing with a large number of patients,repeatedly reading an ECG can be a time-consuming and laborious task for physicians.In recent years,with the technological development in the field of computers and medicine,the use of advanced computer technology to assist doctors in medical diagnosis has become a popular research area,and the use of computer aided diagnosis technology can express and effectively detect and locate myocardial infarction.Aiming at the problem of automatic detection and localization of myocardial infarction,in this paper,we propose different algorithms for automatic detection and localization of myocardial infarction on PTB database using single-lead and multi-lead ECG data using deep learning techniques.The main research of the paper is as follows:(1)For single-lead myocardial infarction data,a one-dimensional CNN and LSTM fusion model is proposed to detect and locate myocardial infarction.One-dimensional CNN extracts the morphological features of the ECG signal,and LSTM extracts the temporal features of the signal,and then fuses the features extracted by the two networks to detect myocardial infarction,and locate eleven different types of myocardial infarction.At the same time,the interpretability of the proposed model is studied.The CAM technology is used to visualize the weights of the convolutional network,and the features extracted by LSTM are activated by using the activation function.By overlapping these weights to the original ECG,the characteristics learned from the model can be interpreted and analyzed more intuitively.(2)For the 12-lead ECG data,we propose a multilead fusion CBAM module of Res Net,combined with the a bidirectional GRU network and attention mechanism,is an end-to-end deep network model to detect and locate healthy and five types of myocardial infarction.At the same time,the effects of the model in both intra-patient and interpatient modes are studied,and the robust performance of the model is explored,and studied the robustness of the model.The efficacy of the proposed network model in different modes was verified using the PTB dataset.