Research On Discrimination Of Abnormal Heart Rhythm Based On Deep Learning

Cardiovascular disease has become the leading cause of human death.Every year,15 million people in the world die from cardiovascular disease,which has become a serious threat to human life and health.The onset of cardiovascular disease is often accompanied by arrhythmia.Arrhythmia is a kind of cardiovascular syndrome,it is a relatively common pathological phenomenon,but fatal arrhythmia is not common.In order to capture these uncommon fatal phenomena,real-time detection of arrhythmia is particularly important.The electrocardiogram can be used to determine whether the heartbeat is normal,and to prevent and diagnose cardiovascular diseases.However,as people’s health awareness increases,more and more attention is paid to cardiovascular diseases,and more and more electrocardiograms are also increasing.Relying only on the doctor’s experience to make a diagnosis will lead to inefficiency and even misdiagnosis or missed diagnosis.Therefore,the realization of automatic classification of arrhythmia has very important practical significance.It can enable people to detect and diagnose heart diseases in time,and effectively protect their heart health.In recent years,deep learning has continued to develop and has been applied to various fields with good results.Combining the advantages of deep learning,this paper makes an in-depth study on ECG signal denoising,ECG feature extraction,and ECG signal automatic classification.The main research contents are as follows:(1)The collected ECG signals are often accompanied by noise such as baseline drift,electrode interference,and EMG interference.To solve the influence of noise on the later ECG signal classification results,this paper proposes a method combining wavelet hard threshold and denoising auto-encoding technology to realize the denoising processing of ECG signals.Compared with traditional noise reduction methods,this method has strong advantages in noise feature learning and reconstruction.First,use the wavelet hard threshold to filter out the noise of the original ECG data to obtain a pure signal;then add corresponding noises to the pure signal,namely baseline drift,electrode interference,and electromyographic interference.Input the noise signal into the denoising auto-encoding network model for training,learn its noise characteristics,and finally reconstruct the pure ECG signal.Among them,the original ECG signal comes from the MIT-BIH abnormal heart rhythm database,and the noise comes from the noise stress test database in MIT-BIH.(2)Traditional ECG features need to be extracted manually,which consumes a lot of manpower,time,and other costs,and the feature shape is fixed,it is difficult to find important features,resulting in a decrease in classification accuracy.The deep learning algorithms have deep network that can automatically learn complex data features.For this reason,a convolutional neural network and Attention-based bidirectional gated recurrent unit neural network combination of arrhythmia discrimination method is proposed.This method first uses a convolutional neural network to extract features of ECG signals.Secondly,input the extracted ECG features into the bidirectional gated recurrent unit neural network for learning and training,and then input the output of this step into the attention mechanism network for further training to obtain more critical ECG features.Finally,the softmax classifier is used to classify the ECG signal.The method is verified by computer simulation on the MIT-BIH arrhythmia database.The experimental results prove that the method proposed in this paper has higher classification accuracy,sensitivity,and positive predictive value.