| As an important application of computer in clinical medicine,the automatic classification of ECG signals can assist doctors to diagnose heart diseases and grasp the health status of the heart timely and accurately.Due to the complexity and poor anti-interference ability of ECG signal itself,traditional classification methods rely on prior knowledge and need to manually extract features,so it is difficult to dig out the deeper feature expression of ECG signal.In this context,this thesis proposes an effective ECG signal classification method based on deep learning technology and combining with the characteristics of ECG signals,aiming at the shortcomings in the existing ECG signal classification algorithms.Waveform detection and cardiac beat classification in the process of ECG signal classification are studied.The main research work of this thesis is described as follows:1.At present most of the R wave detection algorithm uses the MIT-BIH database,although there are abundant data resources,the database but its data only from 47 patients,patients’ sample size is not big enough and acquisition environment is relatively single,lead to test set of algorithms in this database,there are excellent performance,but the actual clinical application effect is not ideal.Therefore,this thesis adopts the latest CPSC2019 data set for model training,and in order to further evaluate the effectiveness and generalization ability of the model,the MIT-BIH database is used to evaluate the performance of the model.2.The R-wave detection is realized by directly inputting the original ECG signal without any signal preprocessing.In the traditional R wave detection algorithm,the ECG signal will be filtered and other pre-processing to obtain the "pure" ECG signal,but in this process,there will be the loss of effective information,resulting in the detection effect is not ideal,poor anti-interference ability,especially when facing the dynamic ECG signal.Therefore,for short-term single-lead ECG signals with poor signal quality and abnormal rhythm,a U-Net neural network based R wave detection method is proposed in this thesis by using the characteristics of convolutional neural network.Without any signal preprocessing,the original ECG signals are directly input to achieve automatic R wave detection.The accuracy of the database of MIT-BIH is 99.85%.3.A classification method of ECG signals based on Bi CNN-LSTM network model was proposed,which realized automatic classification of five common arrhythmias,namely N,L,R,A and V.In the process of experiment,two different schemes were designed and LSTM networks of different lengths were used to optimize the model,and the feasibility of the optimized scheme was verified.The overall accuracy of the model is 99.79%,which is verified by MIT-BIH database. |
