Research On ECG Signal Quality Assessment And Classification Algorithm Based On Deep Learning

Cardiovascular Disease(CVD)is one of the major diseases that threaten human health.The number of CVD deaths worldwide is still increasing year by year,and the number of CVD patients in my country ranks first among various diseases.Therefore,realizing the prevention and automatic diagnosis of CVD and improving the health protection level of the people is of great significance to the social and ability of our country.Electrocardiogram(ECG)signal is a one-dimensional time series signal collected according to human heart activity.Due to limited medical resources and lack of experts in related fields,these data cannot be processed in a timely and effective manner.Therefore,there is an urgent need for an automatic classification and diagnosis method based on collected human ECG data.At the same time,the signal quality of these massive ECG data is uneven.Poor quality ECG signals not only pollute the data but also interfere with feature extraction during subsequent automatic classification.The main work content of this paper is to study the ECG signal quality evaluation and classification algorithm based on deep learning.The specific research content is as follows:(1)Aiming at the problem that a large number of complex features such as time domain,frequency domain and nonlinear domain need to be manually extracted by the traditional methods when classifying the quality of ECG signals,the methods for evaluating the quality of ECG based on convolutional neural network(CNN)is proposed.Firstly,two types of processing are performed on the segmented ECG fragments: one is to directly use CNN to perform feature extraction and classification on one-dimensional time series ECG,and the other is to perform continuous wavelet transform(Continuous Wavelet Transform,CWT)first.It is converted into a time-frequency image,and then the feature extraction of the converted time-frequency image is performed through CNN,and finally the binary classification of good and bad signal quality is performed through a three-layer fully connected network.Noise class,part of the atrial fibrillation class and normal class in the PhysioNet/Computing in Cardiology Challenge 2017(PICC 2017)competition database are used in the experiment.4800 training samples and 1200 test samples are selected from the segmented ECG data.The results show that the proposed first processing method outperforms the second processing method,achieving good performance metrics on the test set.Experiments show that the CNN-based ECG signal quality assessment method proposed in this paper effectively solves the problem of single-lead ECG signal quality assessment,while decreasing human errors caused by manual features.(2)In order to solve the problem that the characteristics of the ECG signal in time series are not conducive to be extracted by CNN,this paper proposes an ECG signal quality assessment method based on Inception-BiLSTM network model.Firstly,The ECG signal is divided into heart beat segments with 1000 sampling points(3.33s),then the Inception module is used to extract the ECG features of three different scales,and the ECG features are further extracted through the 2-layer BiLSTM.Finally,the 1-layer fully connected network and softmax function are used to reduce dimension of ECG signal and achieve the good and bad classification.284 noise classes,some atrial fibrillation and normal classes provided in the PICC 2017 database are used in the experiment.8000 training samples and 2000 test samples are selected from the segmented ECG data.Experimental results show that the proposed method achieves high performance metrics on the test set.The proposed method is applied to the MIT-BIH Noise Stress Test Database datasets(NSTDB),and the scores of overall classification accuracy was 95.3%.The research shows that the ECG signal quality assessment method proposed in this paper has good classification performance and generalization ability.(3)Aiming at the complex and low accuracy of traditional ECG classification technology,this paper proposes a CNN-BiLSTM heart beat classification algorithm based on Inception module.Firstly,the ECG signal is segmented into 1000 timestamps heartbeat segments,and then 3 different scales heartbeats are extracted by using the Inception module.The ECG features are further extracted through a 4-layer one-dimensional CNN and a 2-layer BiLSTM,respectively.Finally,a 1-layer fully connected network and a softmax function are used to reduce the dimension of feature and classify the heartbeat.In order to further improve the classification accuracy,a wavelet threshold noise reduction is used to reduce the noise of the raw data.The performance of the proposed model is evaluated on the PICC 2017 database and the classification accuracy and the F1 score are selected as the evaluation indexes.Results show that the established model has an accuracy rate of 91.38 % for the three types of heartbeats(normal,atrial fibrillation,and others)and F1-score is 91.27%,which is 4.77% and4.59% higher than that of the combined model using only CNN-BiLSTM(accuracy rate of86.61%,F1-score of 86.68%),respectively.Therefore the proposed CNN-BiLSTM ECG classification algorithm based on the Inception module has a better classification effect than the CNN-BiLSTM combined model.