Patient-Specific ECG Classification Algorithem Based On Transfer Learning

In recent years,the rapid economic development has led to the accelerating pace of people’s lives.Increasing pressure of life and irregular lifestyles have led to a sharp rise in the incidence of heart disease,showing a younger trend.The heart disease has become a major threat to human health.Almost all heart attacks are accompanied by arrhythmias,so real-time monitoring and identification of arrhythmias can help people detect and treat heart disease as early as possible.Electrocardingraph(ECG)is an important basis for the analysis and identification of arrhythmia.The classification of ECG signals can identify the existence of arrhythmia and its corresponding type.At present,many scholars have proposed machine learning based ECG signal classification algorithms,but these algorithms generally neglect the individual differences among different patients.In response to this problem,this paper presents a ST-based ECG classification algorithm and a patient-specific ECG classification algorithm based on transfer learning,to achieve differential detection and identification of ECG signals from different patients:(1)Use S transform to extract the time-frequency features of ECG signal.S transform is an effective tool for time-frequency analysis.Compared with the short-time Fourier transform,the window function of S transform is variable.Compared with the wavelet transform,the S-transform avoids the selection process of the complex mother wavelet and solves the phase localization of the wavelet transform.Moreover,the phase spectrum of each frequency component in the time-frequency representation of S transform maintains a direct relationship with the original signal,so that it has good time-frequency characteristics.The experimenta,based on the MIT-BIH arrhythmia database,refers to recommendations of AAMI and divides all ECG signals into five categories.The results show that ECG classification model based on S-transform and SVM classifier performs better than other two contrast models,with the sensitivity and positive rate increasing significantly.The experimental results show that the features based on S transform can reflect the characteristics of ECG signals better,especially the normal and abnormal heartbeats can be distinguished better,which signifies that the classification accuracy is obviously improved.(2)At present,most of the ECG signal classification algorithms do not consider the individual differences between different patients,but in fact the difference is very significant and can not be generalized.It is unreasonable to use uniform standards to judge arrhythmias in different individuals.In order to solve the above problems,this paper presents a patient-specific ECG classification algorithm based on Transfer learning.Transfer learning algorithm uses a joint distribution adaptation algorithm based on feature level.And the algorithm uses the maximum mean difference to measure the distance between the distribution of source domain and the target domain and obtain the characteristic subspace corresponding to the minimum distance based on principal component analysis.The two comparison experiments verify the validity of the classification model proposed in this paper,compared to other personalized EGG signal classification models and the effectiveness of transfer learning on abnormal heartbeat recognition.