Research On Heart Sound Segmentation Algorithm Based On EEMD

In recent years,the incidence of cardiovascular disease has remained high,and the mortality rate is extremely high.It is particularly seen in the elderly,which seriously threatens human health.Heart disease is the most common disease in cardiovascular disease.There are many existed methods of heart disease,but each method has its own characteristics.This article combined with the most common heart disease detection method-heart sound auscultation,extracted some of the heart sounds from the collected heart sounds in order to achieve automatic analysis and classification of heart sound signals.The main work of this article is as follows:1.In this paper,an improved empirical mode decomposition(EMD)method—Ensemble Empirical Mode Decomposition(EEMD)—is used to decompose the intrinsic mode function(IMFs)from the cardiac signal,Selecting Heart Sound Signals from Intrinsic Modal Functions(IMFs).In this process,due to the end-point effect of EMD in the process of decomposition,this paper analyzes and compares a variety of methods,proposes a combination of waveform matching and proportional extension to suppress the endpoint effect,And judging the result of decomposition based on orthogonality,Experiments show that the improved waveform matching method is superior to other methods in suppressing the end effect.For the selected IMF from the decomposed IMFs as the basis for segmentation,this paper proposes its own method,which is to select the IMF component with the largest and the second largest correlation coefficient in the 50-250 range.If there are two IMF components,According to the segmentation result,Calculate the average duration of S1 and S2,and select the IMF whose average duration of S1 is close to 122ms and the average duration of S2 is close to 92ms.2.Based on the above selected IMF,then extracted the envelope of the signal.In this paper,analyzed and compared several common methods for extracting the heart sound signal envelope.Finally,the Hilbert transform method is used to extract the heart sound signal.3.In this paper,several attempts have been made to segment the heart sounds.Finally,proposed a method:combining double thresholds and extreme points to segment the heart sound signal,which improves the segmentation accuracy compared to the double threshold method.4.Finally,this paper uses the hidden Markov model(HSMM)to segment the heart sound signal and compares it with the above results.It can be seen from the experimental results that both methods can accurately segment the heart sound signal,and the segmentation point error is within 5 ms,and the error is within ± 1%with respect to an average heart sound period of 800 ms.However,the HSMM training set needs to use the ECG signal to mark the peak of the PCG signal,and the HSMM needs to train the model,so it needs a large amount of data support.The EEMD method is to directly process the waveform and does not require ECG signal marking.