| The analysis of cardiac physiological signals is one of the main methods for the early noninvasive diagnosis of cardiovascular diseases.It uses the multimodal features of physiological signals to reflect the condition of cardiac function and structure by the analysis of their pathophysiological information,and builds the corresponding disease diagnosis method based on the early diagnostic clue of cardiovascular disease from those.This is also a key research area in the field of biomedical engineering.Chronic heart failure(CHF)has already been an important issue in the prevention and treatment of cardiovascular diseases,and it brings heavy medical and social burdens as well.CHF is a series of complex clinical syndromes such as the impaired ability of ventricular filling or ejection caused by the abnormality of cardiac function or structure,along with compensatory cardiac enlargement or hypertrophy and other compensatory mechanisms.The early symptoms of CHF develop gradually so as to be ignored.Because the proceeding of CHF is irreversible,the effective methods for the staging diagnosis of CHF are of great significance to improve its prognostic effect,especially early staging diagnosis,but the existing clinical detection and diagnostic methods can not provide the objective basis for the noninvasive early diagnosis and staging diagnosis of CHF.The most important aspect of cardiac dysfunction in CHF is not the depressed cardiac performance noted at basal resting states but rather the loss of cardiac reserve,and it can be observed that cardiac contractility reduces subclinically and the relationship between cardiac contractility and peripheral resistance changes at the early stage of CHF.Hence,according to the relationship among heart sound,cardiac contractility and peripheral resistance,this paper conducts the study on the staging diagnosis method for CHF based on heart sound.By the analysis of heart sound short-time feature,we use the changes of heart sound features in the time domain,frequency domain and nonlinear domain to describe the change of cardiac contractility.By the dynamics analysis of heart sound amplitude characteristic sequence,we use the statistical,power spectrum and fluctuation features of the first heart sound amplitude characteristic sequence(S1 sequence),the second heart sound amplitude characteristic sequence(S2 sequence)and the ratio of the first to second heart sound amplitude characteristic sequence(S1/S2 sequence)to measure the variation trend and variability law of cardiac contractility,peripheral resistance and their relative relation in the development and progression of CHF.Finally,we employ the statistical analysis and machine learning methods for the staging diagnosis of CHF.This paper aims to explore objective,convenient and efficient the way to staging diagnosis of CHF so as to improve its prognosis.It has very important clinical application value and social significance.The main contents of this paper are as follows:(1)A novel denoising method for heart sound is proposed.This paper presents an innovative denoising framework based on a joint combination of multi-level singular value decomposition(SVD)and compressed sensing(CS).The performance of proposed framework is evaluated qualitatively and quantitatively,including the test and verification in terms of several standard metrics,and the comparison with the widely used denoising methods such as wavelet transform(WT)and empirical mode decomposition(EMD)using the heart sound databases in different noise levels.The result indicates that the denoising framework can remove the noise as well as preserve the original morphological characteristics of heart sounds effectively without need for knowing the priori knowledge of noise such as frequency and amplitude when applied in any situation.In summary,the proposed denoising framework has potentially theoretical and applied value for noise reduction in heart sound signal.(2)The study on the short-term feature extraction and analysis of heart sound signals from the patients with CHF has been conducted.The difference of heart sound short-term time domain,frequency domain and non-linear features between the patients with CHF and the healthy has been investigated,and the difference of those among the different CHF stages(Stage A,B,C and D)has been also studied.As a result,the change rule of heart sound multi-modal features has been explored in the development and progression of CHF.The result shows that the low-frequency components of heart sound decrease and its corresponding high-frequency components increase,and the chaotic property and complexity of the activity of cardiac mechanical dynamical system decrease so that the cardiac reserve function decreases in the development and progression of CHF.(3)The study on the dynamics analysis of heart sound amplitude characteristic sequences from the patients with CHF has been conducted.We innovatively propose to build the S1 sequence,S2 sequence and S1/S2 sequence,and then to extract their statistical,frequency-domain and fluctuation features for the dynamics analysis of heart sound amplitude characteristic sequence.Through the studies on the difference of variability and fluctuation of the heart sound characteristic amplitude sequences between the patients with CHF and the healthy,and the difference of those among the different CHF stages(Stage A,B,C and D),it reveals dynamic correlation and long-term trends of the main components of heart sound signals caused by the abnormal motion of cardiac mechanical system in the development and progression of CHF.As a result,the variation trend and variability law of cardiac contractility,peripheral resistance and their relative relation in the development and progression of CHF have been explored.The result shows that the amplitude variability of S1 will increase,when CHF occurs,and the long-range correlation of S1 sequence will decrease.It indicates the regularity of cardiac movement and the stability of cardiac contractility decreases.When CHF occurs,the amplitude variability of S2 will also increase,and the long-range correlation of S2 sequence will decrease.It indicates the stability of peripheral resistance decreases due to the serious congestion of pulmonary circulation or systemic circulation with the progression of CHF.When CHF occurs,the long-range correlation of S1/S2 sequence will decrease.It indicates the relative relationship between cardiac contractility and peripheral resistance has been destroyed so as to enlarge its variability.(4)The diagnosis method of CHF has been established based on statistical analysis.We use the statistical methods to establish the mathematical model for the diagnosis of CHF by analysis the features of heart sound short-term sequences and heart sound amplitude sequences.Firstly,the principal components analysis is used to process the obtained heart sound feature set for its dimensionality reduction.Through eliminating the redundancy among each feature,the optimized feature projection set could be obtained.Then,the Fisher criterion based linear discriminant function and the probability statistics based naive Bayes classifier are employed to establish the diagnosis method of CHF.Finally,the receiver operating characteristic(ROC)curve is used to identify the best discriminant thresholds for distinguishing the patients with CHF from the healthy people.The result shows that the Fisher criterion based linear discriminant function performs better than the naive Bayes based classifier,but their accuracy rate for the diagnosis of CHF is restricted.(5)The staging diagnosis method for CHF has been established based on machine learning.The discriminant function,network model and probability statistic model are constructed using the heart sound feature set based on the theory of machine learning,to extend the binary classification that identify the patients with CHF from the healthy people to the multi-classification problem that identify the stages of CHF.The k-nearest neighbor method,artificial neural network,hidden Markov model and support vector machine(SVM)are applied to the establishment of the staging diagnosis method for CHF,and the assessment criteria such as sensitivity,specificity and accuracy are used to evaluate the performance through 10-fold cross validation.According to the performance comparison of these classifiers and considering the properties of sampling data,the SVM is selected as the classifier for further study.The influence of different kernel functions and model parameters on the diagnostic performance of the method has been analyzed and compared.The relationship between the diagnostic feature set and the diagnostic performance of the method is also explored.Finally,by the above optimization steps,the SVM based staging diagnosis method for CHF has been established.The study shows that the machine learning can be used for the screening of CHF and the computer-aided diagnosis for CHF staging. |
PDF Full Text Request