Recurrence Quantification Analysis And Identification Of Heart Sounds Based On Phase Space Denoising

Cardiovascular disease is a serious threat to human health. The motion of the heartsend a vibration signal called heart sounds, which contain many information associatedwith the heart, such as atrial, ventricle, blood vessels, valves and so on. So heart soundsanalysis is an important means for non-invasive heart disease diagnosis. Heart beating isnonlinear, which determines the nonlinear characteristics and complexity of the heartsounds. There is a greater use of the linear method to simplify or approximate the heartmodel at present, but it can not fully reveal the regular of the life activities. With thegradual boost of science and technology, the research of nonlinear system has madesignificant progress. This suggested a way to better evaluate cardiac status for us.Heart sound is a kind of weak vibration signal. In the acquisition process, it isinevitable to introduce interference noise, which would pollute the useful signalcomponents, thus affect the subsequent analysis. So heart sounds denoising isparticularly important before further analysis. Traditional denoising method is oftenstart from the angle of linear, which would ignore the nonlinear essence of heart sounds,thus undermining the dynamics characteristic of the heart sound itself. Subsequently, thecharacteristic parameters extracted from the heart sounds may be not accurate, whichfinally affect recognition.Considering the nonlinear characteristics of heart sounds, this paper proposes adenoising method of local projection of adaptive noise level estimation combined withdiscrete wavelet threshold. The method is based on phase space reconstruction, at first,local projection denoising method was used to remove random noise with a bigger embedding dimension, thus the stronger signal and weak signal were retained. Then thesmaller embedding dimension was taken, and local projection denoising method wasused again, at this time, signals with larger characteristic value were called strong, withsmaller eigenvalue were namely weak signal. But the weak signals also contains somenoise, and the discrete wavelet threshold denoising can effectively retain informationsignal, then it was used to remove the noise, thus a complete heart sounds were attained.The actual observed heart sounds were analyzed, and the results show that the methodcan effectively suppress noise. Calculated the largest Lyapunov index before and afterthe signals were denoised, it is concluded that the method can well retain the nonlinearfeatures of the original signal. Considering the basic characteristics of the chaotic dynamics system, normal heartsounds and two types of abnormal heart sounds which were selected from clinicalacquisition as the objects, the denoised heart sounds were analyzed by nonlinearmethods, including quantitative recursive analysis which is based on recurrence plotsand another chaos characteristics—Kolmogorov entropy, from which the fourcharacteristics, namely Recursive rate, Determine rate, the longest diagonalandKolmogorov entropy, are merged into one feature vector which is analyzed andtransformed by the principal component analysis into an orthogonal anddimension-reduced feature vector which is used to establish joint probability modelswith an identifying threshold for the three kinds of heart sounds. At last, the jointprobability density arbiters based on the models are used to identify the type of heartsounds.