Arrhythmia Recognition Based On Timed Automata Heart Model

With the change of social environment,cardiovascular disease(Cardiovascular disease,CVD)has become the main disease type of human death.In 2016,the total number of deaths in China was 9.67 million,of which cardiovascular disease accounted for 3.975 million,accounting for 41.1%,becoming the leading cause of death in China.Arrhythmia,as a group of important cardiovascular diseases,is hidden and sudden.It can not only occur alone,but also accompany with other diseases.As the most effective tool for clinical diagnosis of arrhythmia,ECG plays an important role in the prediction of disease.The traditional diagnosis mode of arrhythmia is based on the patient’s medical history and clinical examination,according to a set of medical parameters to classify the disease,the diagnosis efficiency is low.Therefore,intelligent classification through automation technology has become the mainstream direction of ECG development.In the early days of AI research,it was said that "the ability of an intelligent program to perform a task mainly depends on the quantity and quality of its knowledge of the task.".In the research of arrhythmia classification,due to the sensitivity of medical data,access to ECG data is highly restricted,so the sample imbalance problem occurs in the classification problem,which leads to poor classification performance.In view of the above problems,this paper starts the experiment from sample equalization to improve the classification performance of 12 lead data.The change of cardiac electrical signal in the conduction process causes the change of human heart rhythm.The whole conduction process has a certain time sequence relationship.These time sequence relationships are ultimately expressed in ECG.Therefore,this paper proposes a sample amplification method based on timed automata heart model.By introducing the tioa heart model,it simulates the generation of ECG data in the conduction system Process.Then the hidden parameters in the heart model are Gaussian optimized to generate more physiological data,so as to improve the classification accuracy.Compared with the current mainstream data generation method,namely generative countermeasure neural network,the effectiveness of the proposed method is proved.Because the residual neural network(RESNET)has a good performance in arrhythmia classification,this paper first designs the residual neural network as the basic structure of multi lead ECG arrhythmia classification,and the overall architecture of the residual network is similar to resnet-18.Due to the complexity of multi lead data,based on the residual network as the classification network,this paper compares the performance of single lead model and lead fusion model.The residual network is used to experiment on 12 lead data respectively.Finally,the lead model fusion is realized by xgboost algorithm.By comparing with the model with the best performance in 10 times cross validation single lead,it is proved that The results show that the optimal F1 score of single lead model is 0.733,while the F1 score of lead fusion is 0.798.Finally,the final F1 score is 0.811.Finally,the data generated by the proposed method are integrated into the experimental data for arrhythmia classification,and the final F1 score is 0.828.Secondly,the data generated by the proposed method can be used to solve the problem of sample imbalance and improve the classification accuracy.It provides more physiological data support for the classification of multi lead arrhythmias.