Research On Electrocardiogram Classification Based On Deep Neural Network

Cardiovascular diseases have become the main cause of death in many countries in recent years.Electrocardiogram is a means of detecting the level of cardiovascular health in human body,and widely used to prevention and control of cardiovascular diseases.However,due to the large daily business of hospitals,the workload of professional electrocardiogram doctors in diagnosis has increased.In recent years,deep neural networks have been widely used in image classification,object detection,and computer-aided electrocardiogram analysis.By using computer-aided electrocardiogram diagnosis technology,the diagnostic efficiency of cardiovascular diseases can be improved,and the workload of clinical doctors can be reduced.There are many types of Electrocardiogram,and how to accurately use deep neural networks for electrocardiogram assisted diagnosis is still a research hotspot.Therefore,this proposal focuses on the electrocardiogram manifestations of arrhythmia diseases and the main problems in the classification process.Based on deep neural network technology,we propose electrocardiogram normal and abnormal classification models and atrial fibrillation disease classification models.The main work of this article includes:(1)The proposed method proposes a deep learning model that combines residual neural network,squeeze and excitation attention mechanism and Bi-directional long short-term memory network to classify the normal and abnormal of multi-lead electrocardiogram.In response to the complex performance of arrhythmia in electrocardiogram signals,squeeze and excitation attention mechanism is used to give weight to important features from the channel dimension,and suppress redundant features;The time series characteristics of complex electrocardiogram signals are acquired by adding bidirectional short-term memory network;In addition,Poly-focal loss function is used to alleviate the problem of data imbalance.In the MIT-BIH-AR database,the specificity,sensitivity,accuracy and F1 score of the normal and abnormal heartbeats classification obtained by this method are 99.54%,99.46%,and 99.52% and 81.66%,respectively.Through testing more than 150,000 electrocardiogram records on the Chinese cardiovascular disease database,the specificity,sensitivity,accuracy and F1 score of the normal and abnormal classification performance of the proposed method are 87.57%,80.50%,84.48% and 81.92%,respectively.The experimental results show that the classification algorithm has achieved good classification performance on both databases.(2)The proposed method proposes a classification method of atrial fibrillation that combines Conv Ne Xt and bidirectional long short-term memory network.In response to the characteristics and positional representation of atrial fibrillation disease on the electrocardiogram,convolutional block attention mechanism is added to the model to effectively extract the channel and spatial features of the electrocardiogram.Through testing more than 150,000 electrocardiogram records on the Chinese cardiovascular disease database.The specificity,sensitivity,accuracy,and F1 values obtained were 98.89%,97.74%,98.87%,and 74.32%,respectively.By compared to other methods,this paper achieved good classification performance,and providing effective method for the prevention and auxiliary diagnosis of atrial fibrillation disease.According to the diagnosis process of doctors,the proposed designs electrocardiogram normal and abnormal classification and atrial fibrillation disease classification based on deep neural networks,and solves the problem of data imbalance.It provides an effective method to assist doctors in diagnosis and has certain clinical application prospects.