Research On Recognition Of Valvular Heart Disease Heart Sound Based On Complexity

Valvular Heart Disease (VHD) has increasingly become one of the most harmfuldiseases to human health. With the growing of the life and work rhythm and theaggravation of the aging of population, its incidence and prevalence is graduallyincreasing, so early check is extremely important to the diagnosis. As one of the mostimportant physiological signals of human body, heart sound contains a large number ofthe physiology and pathology information. Its murmurs are appeared containing thepathological diagnosis information which is reliable to heart valve before abnormalsymptoms such as the obvious electrocardiogram abnormalities or pain have appeared.For a long time, researchers have simplified and abstracted the complex cardiac systemin order to establish an ideal linear model, and the time domain and frequency domainmethod are studied to analysis the cardiac system. Although good results have beenachieved，but people also found that，using linear method is not effective enough toanalysis nonlinear life system. To study the heart sound of Valvular Heart Disease,nonlinear complexity methods in this paper are used to extract the essential dynamiccharacteristics of normal and Valvular Heart Disease’s heart sound. Results show thatthe algorithm are deeply from essence and realize the computer aided diagnosis ofcardiac disease based on heart sound signal with a view from a new angle.This thesis mainly expounds generation mechanism of normal and abnormal heartsounds of VHD and the feature extraction and recognition algorithm based oncomplexity and Support vector machine. On the basis of analyzing the relation betweenheart sounds and murmurs, this paper firstly put forward to the adaptive filteringde-noising method in use of Ensemble Empirical Mode Decomposition (EnsembleEmpirical Mode Decomposition, EEMD), and a automatic segmentation algorithm areproposed based on Lemple-Ziv complexity. Secondly, the normal and abnormal heartsounds are analyzed by EEMD, the algorithm based on approximate entropy andLemple-Ziv complexity of combining the nonlinear dynamics are proposed to analyzethe clinical heart sounds of VHD patients, and acquire effectively some Characteristicvalues reflecting physiological and pathologic characteristics of heart Valve. Finally,Support Vector Machine (Support Vector Machine, SVM) is proposed to identify heartsound signals of several common VHDs.In this paper, the empirical mode decomposition algorithm is proposed, and based on EEMD threshold an adaptive threshold de-noising method is studied. Throughanalyzing the clinical data heart sounds signal, the noise can be effectively filtered andmodal aliasing problems can be solved. A new automatic segmentation method as thesame time based on Lemple-Ziv complexity and EEMD are proposed to segment theheart sounds. When segment the clinical heart sound signals, the experimental resultsshow the algorithm has high accuracy and no extra additional physiological signal toassist.In order to improve the recognition precision and classification accuracy, themethod combining EEMD and Approximate Entropy (Approximate Entropy, ApEn) isproposed to extract the complexity features. Firstly, the heart sound signals aredecomposed into a finite number of Intrinsic Mode Function (IMF). Then, theApproximate Entropy of five intrinsic mode functions using correlation several criteriacan be quantitatively evaluated. After analysis heart murmurs of clinical heart sounds ofVHD, the experimental results shows the nonlinear dynamics features effectively reflectthe information of VHD and effectively describe the characteristics of the pathologicaland pathologic change information hidden in the murmurs.This paper study heart sounds automatic identification on support vector machine.The binary tree of support vector machine method is proposed based on kernel functionchoice to analyze the subclass classification problem. Six common clinical heart soundof VHD such as mitral valve stenosis(MS)、mitral valve insufficiency(MI)、aortic valvestenosis(AS)、aortic valve insufficiency(AI)、ventricular septal defect(VSD) and so on,experimental results verify the feasibility of the proposed method for pathological heartsounds pattern recognition, and the recognition rate can reach to93.23%，which canprovide a strong basis for heart diseases diagnosis and clinical application.