Research On Segmentation Of ECG Feature Waveforms And Classification And Recognition Of Common Arrhythmias

The characteristic wave of the ECG signal can reflect the physiological state information of the heart,which is of great significance for the auxiliary diagnosis of cardiovascular diseases,especially the detection of arrhythmia.In the automatic diagnosis system of arrhythmia,locating the characteristic waves of P,QRS,and T in the electrocardiogram is the primary task.Arrhythmia is one of the most important cardiovascular diseases,and electrocardiogram is often used clinically to judge arrhythmia.However,with the huge amount of Holter data,it is difficult for doctors to maintain a high level of judgment.Missed diagnosis and misdiagnosis will bring great losses to patients.Therefore,the use of computer-assisted diagnosis of arrhythmia has become the focus of current research.When diagnosing arrhythmia diseases,the individual differences between each patient are huge,which is a very big challenge for the automatic identification algorithm of arrhythmia.How can we strengthen the generalization ability while accurately identifying the current patients,and we still need to make new breakthroughs in accurately distinguishing new patients.Deep learning has developed rapidly in recent years,and arrhythmia discrimination algorithms based on deep learning often have stronger generalization capabilities.Therefore,this paper has done the following research work to solve the problem of discriminating abnormal arrhythmia.(1)An ECG segmentation algorithm based on U-net network is designed.The algorithm takes an ECG signal of any length as input,and can obtain segmented images of P wave,QRS wave,and T wave.Then a post-processing algorithm is designed to locate the starting point and ending point of each characteristic wave.This method is trained and tested on LUDB database,and the F1 scores used to detect the starting and ending points of P wave,QRS wave,and T wave are at least 98.96%,99.62%,99.34%,respectively.And the error is within 1ms except for the end point of QRS wave,which has very high accuracy.The generalization ability of the model has also been verified on the QT database.Compared with other feature wave positioning algorithms,the method in this paper can detect all feature waves at the same time and has a better recognition effect.(2)Aiming at the problem of arrhythmia heartbeat classification,a heartbeat classification algorithm based on the residual block network is designed,which innovatively uses the segmentation information of the ECG signature wave as the new network input,and uses the inter-patient data division specified by AAMI Method,the training set and test set are divided according to different patients,which is more suitable for actual application scenarios.The MIT-BIH arrhythmia database was used to train and verify the algorithm points,and the heart beats were divided into 4 categories,and finally achieved a classification accuracy of 94.5%.Experiments have proved the feasibility of this algorithm.