Research On ECG Classification Based On Machine Learning

Cardiovascular disease is now a serious threat to human health and its prevalence is increasing year by year,so early diagnosis and treatment of cardiovascular disease are extremely important.Electrocardiograph(ECG)records the changes in body surface potential due to the beating of the human heart,which can well reflect the real-time status of the human heart and is one of the common means of diagnosing cardiovascular diseases.However,hospitals currently generate a large amount of ECG data every day,and relying only on traditional diagnostic methods to manually analyze these data would make ECG diagnosis time-consuming and inefficient Benefiting from the rapid development of science and technology,more and more scholars are now focusing on the automatic classification of ECG signals,aiming at reducing the burden of medical staff and improving the efficiency and accuracy of diagnosis.In this paper,we study how to automatically classify ECG signals by using deep learning and transfer learning techniques in machine learning as follows.(1)The wavelet threshold denoising method is used to denoise the ECG signals in order to address the problem that noise is easily introduced during the acquisition of ECG signals,which is not conducive to the accurate classification of ECG signals.Firstly,the daubechies5(db5)wavelet function is used to scale decompose the ECG signal,then a reasonable threshold function is set to filter the wavelet coefficients of each scale to remove the noise components,and finally,the ECG signal is reconstructed by combining the wavelet coefficients of each scale to obtain the denoised signal.(2)To solve the complexity and inefficiency of manual feature extraction methods in traditional machine learning algorithms,this paper designs an MI-Net network for automatic classification of ECG signals of myocardial infarction based on the convolutional neural network model in deep learning,with reference to the classical convolutional neural network model Le Net-5 and comprehensive analysis of the characteristics of ECG signals.The acquired ECG signals are fed into the designed MI-Net for automatic feature extraction and classification to assist physicians in making final decisions.The experimental results on the PTB(Physikalisch-Technische Bundesanstalt)ECG database show that the proposed algorithm has a higher classification accuracy for myocardial infarction ECG signals.(3)To address the problem that the neural network model trained with the source domain data perform poorly on the target domain data due to the domain discrepancy between the source and target domains,this paper introduces the domain adaptation method in transfer learning to ECG classification and develops a deep domain adaptation network based on convolutional neural network(CNN)for arrhythmia classification.By minimizing the multi-kernel maximum mean discrepancy(MK-MMD)between the feature distributions of the source and target domain data,the domain discrepancy between the source and target domains is reduced.Then,through minimizing the conditional-entropy of the class distribution on the target domain data,the low-density separation between classes is realized.The final experimental validation was performed on the MIT-BIH(Massachusetts Institute of Technology-Beth Israel Hospital)arrhythmia database,and the experimental results shows that the proposed algorithm has good classification performance improvement on the target domain data.