Design And Implementation Of ECG-assisted Diagnosis System Based On Deep Transfer Learning

In recent years,cardiovascular diseases have become one of the major diseases threatening human life safety,and timely detection of cardiovascular diseases is urgent and necessary.Automatic ECG signal analysis technology can reduce physician workload and improve diagnostic efficiency,which has a very important research value.Traditional feature extraction and classification methods rely on various algorithms to manually select and extract features,and individual differences lead to uncertainty in feature extraction,which affects the accuracy of classification.Deep learning requires a large amount of labeled as the training sets,and there is data dependency phenomenon.To address the above problems,this thesis combines deep learning and transfer learning to propose a new domain-adaptive ECG signal classification algorithm using the common features implied between data to achieve automatic recognition and diagnosis of ECG diseases.Finally,an ECG-assisted diagnosis system is designed and implemented to assist doctors in the diagnosis of ECG diseases.The main research content and work of this thesis can be summarized as follows:1)A domain-based adaptive ECG signal classification network(BiLSTMMKMMD-NET)model is constructed.First,a pre-training model is built using BiDirectional Long-Short Term Memory(BiLSTM)network to learn the temporal characteristics of ECG signals autonomously,and the classification accuracy of the model reaches 99.26%.To address the problem of insufficient ECG annotation data,the model cannot obtain enough training data and the difference in data distribution between the source and target domains.In this paper,an adaptation layer is added to the pre-training model to migrate the knowledge of the existing labeled source domain ECG data to the unlabeled target domain ECG data;by selecting Multi-kernel Maximum Mean Discrepancy(MKMMD),the features of the source and target domains are mapped to the high-dimensional space,minimizing the distribution distance between the source and target domains,and improve the classification effect on the target domain.Finally,multiple migration comparison experiments are conducted in different domain datasets to verify the effectiveness of the domain adaptation model proposed in this paper2)Design and implementation of ECG-assisted diagnosis system.The system is mainly composed of four parts:system management,data sending and receiving,data statistics management and intelligent diagnosis and processing.The system uses the Qt Designer tool to complete the interface design,combines with SQLite to build the system user and ECG database,uses WebService technology to access the server to complete data sending and receiving operations,and uses PyQt technology to complete business logic operations,including ECG drawing and display,user system interaction,generate diagnostic reports and other functions.By integrating the BiLSTMMKMMD-NET model proposed in this thesis,the system completes the intelligent auxiliary diagnosis processing of ECG data.Finally,the system test is carried out to prove the effectiveness of the system’s ECG auxiliary diagnosis function.