Research On Arrhythmia Classification Algorithm Based On ResNet-DCA

In modern society,the pace of people’s life and work is speeding up,which brings the increasing pressure,leading to the increasing proportion of patients with arrhythmia year by year.Arrhythmia is a common cardiovascular disease which is difficult to be found.Therefore,the diagnosis of arrhythmia needs to be confirmed by a professional doctor with ECG knowledge to observe the dynamic electrocardiogram for 48 hours.However,the artificial diagnosis of arrhythmia is excessively time-consuming and labor-consuming,and requires doctors who have enormous knowledge of ECG pathology and experience in ECG disease diagnosis.As a result,it is particularly important to design an arrhythmia classification algorithm based on ECG to assist doctors in diagnosis.Many researchers have proposed a large number of arrhythmia classification algorithms based on ECG so far,but these algorithms have some shortcomings:First of all,due to the imbalance and noise of arrhythmia data,the classification algorithm of arrhythmia is not effective.Secondly,most of the arrhythmia classification algorithms only use the features of single lead ECG signals,but they didn’t use the correlation features among the multi lead ECG signals,so the feature extraction is not enough,and the classification accuracy can be much higher.In order to solve the above problems,the following research content is carried out:(1)We build a hybrid netwoek model — ResNet-Bi LSTM-FL,which is comprised of Bi-directional Long Short-Term Memory(Bi LSTM)and Residual Network(ResNet),it could automatically classifies and identifies five different types of ECG signals in the MIT-BIH arrhythmia database.As a kind of convolutional network,ResNet not only has strong noise robustness,but also can effectively extract the local features of the ECG signal,while Bi LSTM can focus on the global features of the ECG signal.The two networks are complementary The ResNet-Bi LSTM-FL proposed in this paper.Experimental results show that our method has well noise robustness.It also shows that the addition of focus loss can enhance the model’s ability to process tilted data and improve classification accuracy.In addition,We contracted ResNet-Bi LSTM-FL with classification algorithms which is widely used in the field of arrhythmia the field of arrhythmia classification,and we got an overall accuracy of 99.31%,which is better than other classification algorithms.(2)We propose a classification algorithm of arrhythmia based on ECG multi-lead,in order to solve the problem of the insufficient features extracted by ResNet-Bi LSTM-FL,the single lead data,and the low classification accuracy of less data.The algorithm flow is as follows: First,we extract each lead deep features through ResNet.Next,the discriminant correlation analysis algorithm is used to maximize the feature correlation of the two-lead ECG signal to realize the two-lead feature fusion.Finally,the support vector machine classifier is used to classify the feature set.For five different types of ECG signals in MIT-BIH arrhythmia database,we peopose ResNet-DCA algorithm that the overall accuracy and F1 value of the proposed dual-lead fusion classification algorithm are higher than those of ResNet-Bi LSTM-FL.In addition,by comparing the classification algorithms of other references,the ResNet-DCA algorithm proposed in this paper has good performance.