The Fractal Characteristics Analysis And Recognition Research Of Heart Sound Based On Fractal Theory

As a vital physiological signal of human body, heart sound has its own uniquemechanism and characteristic, which contains much physiological and pathologicalinformation of the human cardiac system. It can reflect the mechanical movement of theheart and great vessels, and it is the basic method for clinical assessment of the heartfunction condition with the advantage of noninvasiveness and convenience. Researchshows that heart sound signal is deterministic non-linear dynamical system and hasobvious fractal characteristic. But the existing research is based on the lineartime-varying and time invariant model for the most part; it can not reveal the internalmechanism of heart sound signal, so it need to use the theory of nonlinear to analyze theheart sound. As an active branch of mathematics in nonlinear science, Fractal theory canwell reveal the inherent special property of nonlinear process.This paper analyzes the fractal characteristic of heart sounds signal based on fractaltheory, so as to understand the inherent characteristics of heart sound more deeply fromthe perspective of nonlinear and provides a new method for the analysis of heart soundsignal.This paper extracts the fractal characteristic of heart sound from two aspects,including the correlation dimension and multi-fractal. Single fractal dimension focuseson the irregular degree of fractal objects for the overall, multi-fractal has good ability oflocal analysis, considering the characteristics of heart sound from the whole and localall together, and it can improve the classification accuracy. Firstly, using empirical modedecomposition method, the heart sound signals are decomposed into several intrinsicmode functions (IMFs) changed with the frequency of heart sound signal forself-adaption. The main IMF components are chose by using the criteria of mutualcorrelation coefficient, analyze the instantaneous frequency characteristic of main IMFs(IMF1~IMF4) components, the instantaneous frequency characteristics of main IMFshas obvious difference between normal heart sounds and abnormal heart sounds, and itshows that the details information of heart sound signal can be well reflected by meansof empirical mode decomposition, and then calculate the correlation dimension of mainIMFs (IMF1~IMF4) using G-P algorithm, the correlation dimension of main IMFs(IMF1~IMF4) component constitute the feature vector of heart sound signal recognition.Secondly，this part analyzes the multi-fractal characteristics of heart sound signal, including generalized Hurst index, quality index and the multi-fractal spectrum.Comprehensive analysis on the three multi-fractal characteristics shows that the heartsound signal has multi-fractal properties, and the multi-fractal characteristics of normalheart sound signal is stronger than abnormal heart sound signal. Through the selectiveanalysis of the four characteristic parameters of multi-fractal spectral, select themulti-fractal spectrum width for the vector feature of heart sound recognition as it hassignificant difference between normal heart sound signal and abnormal heart soundsignal compared with the other three multi-fractal characteristic parameters. Finally, thefeature vector, the correlation dimension of main IMFs (IMF1~IMF4) component andmulti-fractal spectrum width, is input into support vector machine (SVM) to realize theautomatic recognition of the heart sound signals.Collecting signals, including the normal heart sound and abnormal kinds such as,heart sound with arrhythmia, mitral stenosis, aortic stenosis, splitting first heart sounds,and ventricular septal defect, and all of those heart sound signals are tested with theproposed method in the paper. The results showed that the features of heart soundcombining with the correlation dimension of main IMFs (IMF1~IMF4) component andmulti-fractal spectrum width could get a higher recognition rate,and the fractal theoryhas good properties for revealing the nonlinear characteristic of heart sound signal,which could lay a foundation for the future research of the essential nonlinearity of theheart sound and the cardiac disease diagnosis.