ECG Classification Based On Recurrent Neural Network

Biomedical data,including video,image and signal data,are widely used in the medical field.Among them,the electrocardiogram(ECG)signals of non-invasive detection is one of the most frequently studied data.Due to the extremely low proportion of doctors and patients and the huge amount of ECG data,the prevention and diagnosis of arrhythmia has become a long-term problem.With the development of machine learning,deep learning and other technologies and the success of their application in the biomedical field,the classification of arrhythmias is moving towards intelligence,fastness,high accuracy and diversity.This paper combines pre-processing and classification of ECG data with convolutional neural networks and recurrent neural networks.The main research contents are as follows:(1)There are three main types of interference in ECG signals:EMG interference,power frequency interference,and baseline drift.In this paper,for the comparison of this phenomenon,three traditional filters are used to filter these three kinds of interference,and a double orthogonal wavelet transform method is used to filter these kinds of interference at one time.After experimental comparison,the method of biorthogonal wavelet has more advantages in ECG noise filtering,so this paper chooses biorthogonal wavelet filter.(2)Aiming at the problem of sampling complete QRS waves in experimental data,due to the problem of incomplete collection of QRS waves in the collection of a single QRS wave,this paper proposes a single complete RR wave data collection method,and uses the experimental model to classifify the collected data.In order to further improve the classification performance,on the basis of a single RR wave,considering the time correlation of the ECG signal sequence,a dual RR wave acquisition is proposed,and the experimental model is used to classify the data.Finally,the classification and comparison experiments of the three collected heartbeat data prove that the double RR wave has a better ability to express features than a single QRS wave and a single RR wave.(3)This paper combines the characteristics of CNN network and BGRU network and combines the characteristics of heart wave data to propose a fusion classification model based on CNN and BGRU to classify all 23 categories of arrhythmia data of MIT-BIH.and compares the experimental results with a single CNN and a single BGRU classification model.The performance superiority of the fusion model is proved.And in order to verify the performance of deep learning algorithms,this paper also proposes two machine learning algorithms:support vector machine and random forest for comparison experiments.The experimental results also show that compared to shallow machine learning methods,deep learning methods have better performance.