Research And Implementation Of Arrhythmia Classification Based On Mixed T-F Features And DL

With the acceleration of population aging,China will face great pressure on population health,and the elderly are also the biggest potential threat of cardiovascular diseases.Data show that one of the highest mortality rates in the world is cardiovascular disease,the main manifestation of cardiovascular disease is arrhythmia.The most effective method for medical detection of arrhythmias is electrocardiogram(ECG),which can identify arrhythmias by observing the characteristics of ECG.Therefore,long-term ECG monitoring has become a hot research issue.In recent years,deep learning has been used in various fields.At present,deep learning has been explored in ECG classification field.However,some further research is needed to achieve long-term daily ECG monitoring and ensure the accuracy of arrhythmia diagnosis.In this paper,an intelligent diagnosis platform was developed for arrhythmias connected by using wearable ECG acquisition equipment.Moreover,the deep learning diagnosis algorithm was proposed for intelligent diagnosis platform,which can meet the characteristics of simple,effective and stable,and is convenient for stable and efficient operation on the platform.The main research work of this paper is divided into the following parts:(1)In this paper,a hybrid time-frequency domain feature extraction method is proposed for arrhythmia classification by using convolution neural network method.The fused features include the time domain characteristics came from the RR period,frequency domain characteristics from hilbert-huang transform,and joint timefrequency domain features extracted from continuous wavelet transform.Then the fused features are used as an input to the convolution neural network for training classification model,and the focal loss is used to replace cross entropy as the loss function of the training model,so as to realize the arrhythmias classification.In addition,the MIT-BIH arrhythmia database was used to verify the performances of the proposed method for arrhythmias classification of four types of ECG data.Experimental results show that compared with the existing classification algorithms,the proposed method improves the F1 of class obviously.(2)In this paper,a simple and efficient method based on Squeeze-and-Excitation network(SENet)was proposed for arrhythmia classification.In this method,ECG features were extracted by continuous wavelet Transform.And Lightweight Context Transform(LCT)combined with normalized,linear Transform and SENet was used to classify arrhythmia data.The performance of the proposed method was verified on the MIT-BIH arrhythmia database.ECG samples with four different labels were classified,and the effect of four wavelet families on the classification results was discussed.The experimental results show that the proposed method can maintain high accuracy of arrhythmia classification with simple,effective,and low computational cost.In addition,the proposed method can meet the requirements of simple and efficient automatic diagnosis of arrhythmia intelligent monitoring platform.(3)Based on Django Rest Framework,a Restful Application Programming Interface(API)was developed to implement the arrhythmia classification.Combined with ECG QRS recognition,ECG denoising and arrhythmia diagnosis algorithm based on deep learning,a real-time monitoring and intelligent diagnosis platform was designed for arrhythmia diagnosis.According to the requirements of ECG classification,the platform can change the deep learning model flexibly,so as to diagnose corresponding ECG classification accurately.The My SQL database and reverse proxy were used to deploy on Aliyun ECS cloud server.This paper mainly studies a method for arrhythmia classification based on hybrid time-frequency domain features and deep learning.The developed arrhythmia classification system is verified by ECG data of normal heart rate,and it can operate normally,and the arrhythmia classification algorithm also has good recognition performance.