Research On QRS Wave Detection And Classification Method For Dynamic ECG Signal Based On Deep Learning

Heart diseases have become common diseases posing major threat to human health,causing the increasing of mortality rate year by year.Therefore,it is urgent to prevent and actively intervene it in advance.At present,monitoring the dynamic ECG signal is one of the routine clinical diagnosis and treatment techniques for cardiac diseases.However,automatic analysis of dynamic ECG signal is still the nodus limted by noise,the complexity of symptoms and some other factors,Therefore,it is significant to automatically detecte and classify the abnormal ECG signals with high acuracy in advance.This study reasearched the automatic detection algorithm of the dynamic ECG signal QRS wave based on the generative countermeasure network,which also integrating the R wave detectig-result as an auxiliary feature.The premature ventricular beats(PVC)and supraventricular premature beats(SPB)are classified to obtain good results and lay the foundation of future research,used the improved deep residual network.The main research work of this paper includes:(1)QRS Wave detection algorithm of dynamic ECG signal based on Generative Adversarial Networks.In order to solve the problem of poor signal quality and abnormal rhythm wave forms of the dynamic ECG signals,a new QRS wave detection algorithm based on Generative Adversarial Networks was proposed.This algorithm is based on the Pix2 Pix network.The Generative network adopts the U-Net structure.The Discriminant network uses the Patch-GAN idea.The loss function is the Wassertein distance(EM distance)of WGAN.And jump connection structure of the U-Net structure is used to map ECG data to R wave peak position data.To improve the signal-to-noise ratio,the original signal has been denoised and reorganized.2000 groups of single-lead ECG signals in ICBEB database were used for algorithm verification.And results compared with P&T algorithm and CNN algorithm.The accuracy of QRS wave detection is 99.13%,which is significantly better than P&T algorithm and CNN algorithm.The QRS wave detection algorithm based on Generative Adversarial Networks obtains the best results,which shows the effectiveness of this paper.Simultaneously,the heart rate is calculated,reaching the accuracy of 98.32%,which is also significantly better than P&T algorithm and Bi LSTM algorithm.(2)In order to solve the problem of low accuracy in the classification of abnormal heart rhythms for dynamic ECG signals,a classification algorithm for PVC and SPB based on improved Res Net was proposed.The algorithm mentioned above is used to detecte R wave and calculate heart rate of ECG signals after the preprocessing.The heart rate results are integrated into ECG signals to enhance the data.For the network,,an improved Res Net is designed,which change some parameters of the standard Res Net50 network like changing the size of the convolution kernel,introducing asymmetric bottleneck residual blocks and increasing the pooling layer.To verify the effectiveness of the prosed algorithm,PVC and SPB classification were carried out under with and without data enhancement resectively.And the data with data-enhancement also used to contrast the proposed algorithm,the standard Res Net50 network and the classic classification network.The results show that the recognition accuracy of the proposed algorithm has reached 0.86,which indicates the enhanced data set fused with heart rate features can effectively improve the classification accuracy,and the improved network classification results are optimal.