Research On ECG Signal Processing Method Based On Machine Learning

Electrocardiography is an important bioelectrical signal which can reflect a great deal of physiological information.It is an important basis for doctors to diagnose myocardial diseases.In order to improve the work efficiency of doctors in diagnosing cardiac health status through ECG signals,the researches on ECG signal processing mainly focuses on:(1)how to get more clean and less distorted ECG signals;(2)how to identify all kinds of ECG features more accurately and effectively;(3)how to automatically determine the patient's health status through ECG signals.In order to obtain cleaner ECG signals,adaptive filter is used to process the original ECG signals and filter out the power-line noise and EMG noise signals mixed in them.In this paper,an adaptive filter models without external reference is used to avoid the shortcomings of introducing reference signals from the outside of the adaptive filter.The single lead ECG signal is used as a reference signal,and obtaining desired signal by shifting the ECG signal or ECG signal is used as the desired signal and obtaining reference signal through the mean value operation.The experimental results show that three adaptive noise suppressor models without external interference can achieve good result for suppressing the power frequency interference and EMG interference,and the model 2 is the best(the least mean square error is minimum among them).The baseline drift can easily cause the amplitude of baseline of ECG to fluctuate unperiodically,which makes the signal deform and affects the analysis and diagnosis of ECG signals by doctors.In this paper,a median filter is used to fit the ECG base drift signal,and then the fitted base drift signal is subtracted from the original signal to achieve the effect of suppressing the base drift signal.In this paper,the characteristics of the time domain signal of ECG are taken as the characteristic signal of Support Vector Machine(SVM),and the R wave peak,extracted from many ECG sampling points through SVM,is used as the reference point to locate other waveforms.In this paper,in order to verify the accuracy of the algorithm in identifying the peak of R wares,the arrhythmia signal in open ECG database provided by MIT is shown as an example.From the experimental results show that in the case of serious noise interferencein the original ECG signal,the algorithm in this paper can recognize the peak position of R wave well(AUC:0.9992),and on this basis,other characteristic waveforms can be located.Automated diagnosis of cardiac diseases is one of the focuses of ECG signal research at present.In this paper,for the first time,the chaotic features of ECG is used as the feature set of random forest to diagnose four kinds of common arrhythmia signals and normal ECG with time-domain and frequency-domain characteristic of ECG signals.The experimental results show that the classification accuracy of the proposed algorithm is more than 90% for cardiac arrhythmia signals,and the diagnosis results have certain reference value.