Research Of Chronic Heart Failure Typing Based On Heart Sounds

In recent years,cardiovascular disease has always been a major threat to human health,heart failure is the final stage of the disease,its hospitalization rate and fatality rate are all high.According to the left ventricular ejection fraction,heart failure is divided into heart failure with reduced ejection fraction?HFrEF,LVEF?27?50%?and heart failure with preserved ejection fraction?HFpEF,LVEF?50%?.The mortality of HFrEF is significantly higher than that of HFpEF,and HFpEF can be converted into HFrEF,so it is of great clinical significance to diagnose the two kinds of heart failure.Heart sounds come from the vibration of the heart,which can reflect the structure and function of the heart.There is different ventricular reconstruction between HFrEF and HFpEF,so the change of heart sounds is different either.Give this,the relationship between the characteristics of HFrEF and HFpEF were analyzed in this paper,explore the method of chronic heart failure typing?Firstly,according to?2014 ESC Guidelines for the diagnosis and treatment of acute and chronic heart failure?,this study focused on three groups of people,group 1 was made up of HFrEF patients?n?28?72?,group 2 was comprised of adult HFpEF patients?n?28?172?,and group 3 contained 98 healthy people,the heart sound signal of every person lasting 20 minutes was recorded.After analyzing the main types and characteristics of noise in the heart sounds,fast independent component analysis?FastICA?combined with wavelet decomposition was used to denoising,compared with the denoising effect of FastICA and wavelet decomposition respectively,this method can remove not only the respiratory sounds but also gaussian white noise in the heart sounds.Secondly,the characteristic envelopment of the heart sound signal was extracted by the viola integral,and the first order Shannon energy was used to smooth,on this basis,a double threshold is set to determine the location of the endpoint for automatic segmentation of heart sounds.Then the time ratio of the first to second heart sound?TS1/TS2?,the amplitude ratio of the first to second heart sound?S1/S2?,the population standard deviation?SDDS?of the ratio of diastolic to systolic duration?D/S?,the population standard deviation of the first heart sound interval?SDSSI?were extracted as characteristics in time domain.Then S transformation was performed on heart sound signal to analyze its characteristics in time-frequency domain,and the energy ratio of the first to second heart sound?ES1/ES2?,the energy fraction of heart sound signal with low frequency?EF-LF?,the energy fraction of heart sound signal with high frequency?EF-HF?,the energy fraction of heart sound signal with low and high frequency in cardiac systole respectively?EF-SLF,EF-SHF?,the energy fraction of heart sound signal with low and high frequency in cardiac diastole respectively?EF-DLF,EF-DHF?were extracted too.Since the heart system is a chaotic system,the heart sounds have obvious chaotic characteristics,and the maximum Lyapunov exponent(Lyapunov_max)was extracted for nonlinear analysis.Lastly,principal component analysis?PCA?was performed on eight characteristic values which were relatively independent,and the cumulative variance contribution rate of the first four eigenvalues achieved 94.07%??85%?.The new eigenvector was composed of the first four eigenvalues for classification recognition,comparing with sensitivity?specificity and accuracy among Fisher linear classification?back-propagation neural network and Iterative Self—Organizing Data Analysis Techniques Algorithm?ISODATA?,ISODATA was selected as the classifier,its sensitivity?specificity and accuracy for discriminating HFrEF patients and HFpEF patients reached 86.11%?93.02%and90.98%,which were high.It's demonstrated that the feature extracted from heart sound signal described the significant difference among two groups,that provideds the basis of aided typing diagnosis for chronic heart failure.