Chronic Heart Failure Diagnosis Based On The Extraction Of Heart Sound Characteristics

Chronic heart failure(CHF)is the terminal state of various cardiovascular diseases.The cause of the disease is complicated,and the prevalence and mortality are high,which makes CHF become a major public health problem.Early diagnosis and treatment can effectively delay its progress.At present,the clinical diagnosis methods for CHF are inconvenient to popularize and the problem of diagnostic lag is widespread,leading to a low diagnosis rate of CHF.As one of human physiological signals,heart sound signal contains a large number of physiological and pathological information about the function of various parts of the heart and large blood vessels.Analysis of heart sound signals has become an important means of clinical auxiliary diagnosis of heart diseases.However,the study of CHF based on heart sound signals is not deep enough and lacks effective diagnostic indicators,which limits the application of heart sound signal analysis in heart failure auxiliary diagnosis.In order to find more diagnostic indicators that are specific to heart failure,this paper has studied the relationship between heart sounds and heart failure,and extracted the features of heart sound signals from three perspectives: time-frequency domain,energy,and complexity.Statistical analysis was used to screen characteristic parameters in order to achieve heart failure auxiliary diagnosis by classification of heart sounds.The paper first elaborated the physiological mechanism and basic characteristics of heart sounds,and lays a theoretical foundation for the preprocessing and feature extraction of heart sound signals.Due to the high requirement for heart sound preprocessing,this paper proposed an improved heart sound denoising algorithm based on Complementary Ensemble Empirical Mode Decomposition(CEEMD)and Wavelet Packet Transform(WPT).The denoising of heart sounds is compared with the results of the other three commonly used denoising algorithms,and SNR and RMSE are used to quantitatively evaluate the superiority of the denoising algorithm under different SNR conditions.Hilbert Huang Transform(HHT)and Shannon Energy(SE)are used to extract the feature envelope of the denoised heart sound,then the threshold segmentation of the heart sound signal is achieved.For the preprocessed heart sound signal,features are extracted from three aspects: First,the time-domain characteristic heart reserve indexes has extracted: the ratio of the first heart sound to the second heart sound amplitude(S1/S2)and the ratio of diastolic to systolic duration(D/S).The statistical analysis results show that this indicator can reveal significant differences in heart reserve ability,myocardial contractility,peripheral resistance,and ventricular blood perfusion time in healthy patients.Secondly,the wavelet packet decomposition sub-band energy scores EF1~EF8 of the heart sound signal has extracted to characterize the energy distribution of the heart and the audio domain.The analysis results show that the heart sound energy of the heart failure has a tendency to migrate to the high frequency compared with the normal heart sound,and can reflect the heart sound heart sound Changes in composition,including EF1 ~ EF4,EF7,EF8 six indicators can be used to describe the difference between the two types of heart sounds.In addition,on the basis of murmur separation,the heart sound energy efficiency indicators has extracted: the first heart sound energy fraction(S1_EF)and the murmur energy score(HM_EF).The results show that there is a significant difference in heart sound heart sounds on S1_EF,reflecting the heart failure heart energy The use of obstacles leads to a lower energy efficiency of the heart.Thirdly,we extracted the sample entropy(SampEn)of the complexity of heart sound time series.The results showed that heart sound time series entropy is higher and the distribution range is smaller,which reflects the heart sounds of patients with CHF is more irregular and the cardiac adaptive adjustment ability of them is poor.In this paper,the Support Vector Machine(SVM)was selected as a classifier.Principal Component Analysis(PCA)was used to reduce the dimension of initially obtained 10 characteristics,including S1/S2,D/S,and EF1~EF4.EF7,EF8,S1_EF,and SampEn.Then optimize the classifier parameters.Training recognition was performed on 120 normal heart sounds and 108 heart failure heart sounds.The experimental results show that the highest recognition rate is 95.59% when the kernel function is Gaussian radial basis,(δ,C)is(2,2),and the characteristic parameter contribution rate is 88.2%.The comparison with the traditional BP neural network classifier shows that the heart failure classification method proposed in this paper has advantages in both effect and speed,so that provides a fast and efficient method for clinical diagnosis of CHF.