Heart Rate Estimation And Arrhythmia Classification Of Ecg Signal Based On NN

Heart rate and arrhythmia are important indicators to describe the health status of the human heart.Through the analysis of ECG signal,heart rate can be estimated and arrhythmia can be diagnosed.Although researchers have been studying the methods of heart rate estimation and arrhythmia diagnosis for many years,there are still some deficiencies in the related methods proposed.The heart rate estimation methods generally have insufficient generalization ability,because it cannot achieve good results on ECG signals that are severely affected by noise or those which are arrhythmias.Besides,there is a general lack of accuracy in the diagnosis of arrhythmia.In view of the great clinical significance of heart rate estimation and arrhythmia diagnosis and the deficiency of current methods,we devote to study a heart rate estimation method with higher generalization ability and an arrhythmia classification method with higher accuracy.To solve the problem of heart rate estimation,a three-step method is proposed.Firstly,each interval which contains a QRS peak and has limited width is segmented,then a peak is determined on each segmented interval,and finally the heart rate is calculated.The method of segmenting intervals which contain QRS peaks and have limited width is based on neural network.In order to achieve such interval segmentation,a method based on piecewise linear representation and according to QRS peaks is proposed to generate such intervals annotations.A wave enhancement method is proposed for the preprocessing of ECG data.This method reduces the time dimension of signal data,and enhances the difference of QRS wave,other waves and non-waves.The proposed heart rate estimation method is tested on the CPSC2019 data set.The test result show that this method has a high accuracy in QRS peak location and heart rate estimation.The test result also confirms that the application of the wave enhancement method in ECG data preprocessing can effectively improve the accuracy of the segmentation of intervals which contain QRS peaks and have limited width and heart rate estimation.According to the classification method of arrhythmia,we proposed an improved method based on the existing classification method of arrhythmia.Firstly,the best loss function for arrhythmia classification problem is selected by a comparing experiment.There are two problems in arrhythmia classification: unbalanced distribution of sample types and unbalanced difficulty of sample classification.Researchers propose a variety of loss functions to overcome the impact of these two types of problems on classification accuracy.Experiments show that the GHM-C loss is the best loss function.Then,an improved neural network model is proposed,and the performance difference between the model and the prototype is compared.The experimental results show that our proposed neural network model is superior to the original model.Finally,we compare the classification accuracy of different inputs by taking the band-pass filtered ECG signal,wave enhanced signal and the combination of band-pass filtered ECG signal and wave enhanced signal as inputs.The results show that the accuracy of classification can be effectively improved by using wave enhancement signal and band-pass filtered ECG signal as input.