The distribution dimension of delay embedded heart rate variability based on slope threshold primary component analysis
Huang Xiao-Lin,Ning Xin-Bao
DOI: https://doi.org/10.3321/j.issn:0469-5097.2008.04.003
2008-01-01
Abstract:Heart rate variability(HRV),which means the time variability between the beat-to-beat heart rates,is a general situation in the normals even in calm state.It is known that the heart rates are partly driven by the competing forces of parasympathetic versus sympathetic stimuli,and exhibit complex nonlinear temporal structures which are similar to those found in physical systems driven away from an equilibrium state.In recent years,the concepts and methods originated from nonlinear dynamics have been applied to the analysis of HRV and much progress has been made so far in this promising area.Among these methods is the typical class of dimension,e.g.,the correlation dimension D2,the information dimension D1,the fractal dimension D0,which characterize the geometry of attractors in the phase space from different aspects.But all these methods are quite data-intensive,time-consuming and noise-sensitive.In this paper,we also focus on the geometry characteristics of HRV series in the delay embedding reconstructed phase space and propose a parameter named distribution dimension.To get a distribution dimension,we improve the primary component analysis(PCA) in the determination of primary directions by the slope threshold which equals the slope of eigenvalue spectrum for Gaussian random series under the same embedding dimension.The distribution dimensions of various systems under the same embedding dimension reflect the linear free degrees and larger distribution dimension corresponding to greeter complexity.The method involves only eigenvalue decomposition in computation,so it is simple to realize.The study also shows that the method puts loose restricts on the data length,and greatly improves the computing efficiency.In addition,it has good resistance to noise infection,which is a general situation in data acquisition of experiments,so it suits the clinical diagnosis quite well.The successful applications to the model nonlinear time series such as Logistic,Lorenz and Roessler demonstrate the good performance of this method.The improved method is finally applied to the HRV series,and we reach the result that,under the same embedding dimension,the distribution dimension for the young normals' HRV is the largest,and those for the old normals and the congestive heart failure(CHF) sufferers decline greatly.The result implies that the distribution of HRV in phase space for the old normals and the CHF sufferers compacts to less directions and has smaller linear free degrees.