ebook img

Multifractal age? Multifractal analysis of cardiac interbeat intervals in assessing of healthy aging PDF

0.34 MB·English
Save to my drive
Quick download
Download
Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.

Preview Multifractal age? Multifractal analysis of cardiac interbeat intervals in assessing of healthy aging

PROCEEDINGSOFTHE7THESGCO2012,APRIL22-25,2012,KAZIMIERZDOLNY,POLAND 1 Multifractal age? Multifractal analysis of cardiac interbeat intervals in assessing of healthy aging Danuta Makowiec, Stanisław Kryszewski, Joanna Wdowczyk-Szulc, Marta Z˙arczyn´ska-Buchowiecka, Marcin Gruchała, Andrzej Rynkiewicz, 3 1 Abstract—24-hour Holter recordings of 124 healthy people at classified as fractal. However, more careful investigation al- 0 different age are studied. The nocturnal signals of young people lows one to distinguish different scaling regimes. One type of 2 revealthepresenceofthemultiplicativestructure.Thisstructure scaling is present for the short-time scales and another one n issignificantlyweakerindiurnalsignalsandbecomeslessevident for the long ones. Such variations of scaling can be related a for elderly people. Multifractal analysis allows us to propose J qualitativeandquantitativemethodstoestimatetheadvancement to changes in the intrinsic properties of the considered time of the aging process for healthy humans. series [5], [6]. 6 Index Terms—autonomic control, heart rate variability, mul- The cardiac interbeat time intervals, so called RR-signals, ] tifractal spectrum are studied in four generally accepted [7] frequency regimes: n a • high-frequency (HF) (0.15,0.4) Hz, a- I. INTRODUCTION • low-frequency (LF) (0.04,0.15) Hz, t • very-low-frequency (VLF) (0.0033,0.04) Hz, a The structure-function-based multifractal analysis relies on • ultra-low-frequency (ULF) <0.0033 Hz. d scaling properties of some statistical measure of the data set . Such a division is associated with different, physiologically cs (see [1]). Namely, if {Xi, i = 1,2,...,N} is a time series distinguishable aspects of cardiac rhythm control. Therefore, for which the multifractality is investigated, and R(i,n) is a i by considering time scales in the scaling procedure corre- s function that measures a certain property of a signal in the y spondingtothelistedfrequencybands,wehopetogetinsights ith box containing n consecutive points of data (boxes do h into these aspects. In practical application, assuming that the p not overlap), then the multifractal analysis considers scaling mean cardiac interbeat is 0.8 s, we have assigned size of [ properties of all real value moments of R(i,n). Then, the ”boxes” for investigation of the scaling properties in signals. question is whether the following power-law dependence on 1 In particular, we assume n = 32,...,420 RR-intervals as box size n exists or not for various real q’s: v corresponding to the VLF band 9 Z(n,q)=<|R(i,n)|q > ∼nτ(q), (1) Multifractal analysis based on the study of structure func- 1 {Xi} tions must be applied with care. If the data exhibits noise 0 where < . > denotes averaging over the corresponding 1 {Xi} properties then the partial summation must be performed data sets. When it is found that such a power-law scaling . before the proper analysis [1]. On the other hand, if the data 1 is present for some q, then we say that the studied process 0 has some features of the random walk, there is no need for has fractal structure. Furthermore, if the scaling exponent 3 partial summation. Therefore, before multifractal analysis one 1 function τ(q) is not linear in q then it is said that the process needs to decide whether the data should be treated as a noise : is multifractal. The multifractal spectrum of such a process: v process or as a random walk. In the following we will show h → D(h) is obtained from the scaling exponent function i what additional information can be gained if the initial data is X τ(q) by the Legendre transformation (q,τ(q))→(h,D(h)). considered both as a noise process and as a random walk. r Different values of q correspond to certain statistical prop- a The method summarized above was applied to 24-hour erties of a signal. For example, when q = 2, the method Holter recordings of cardiac time interbeat signals obtained provides estimates for the Hurst exponent, H =(τ(2)+1)/2. from healthy persons. The obtained results led to the estima- If q → ∞ then we get information about large changes in torswhichallowedustodescribeand(atleasttosomeextent) the signal, while q → −∞ characterizes the smooth parts of quantifytheadvanceoftheagingprocess.Thespecificdetails a series. These properties support and motivate our efforts in are given in [2], [3], [4]. Extensive discussion, numerous developing methods based on the multifractal approach. examples are presented in just cited articles. Here we focus It often appears that the scaling curves, namely logZ(n,q) only on the essentials. vs logn plots are, at least at the first sight, linear. Hence, according to the above given arguments, the signal can be II. METHODS DanutaMakowiecandStanisławKryszewskiarewithGdan´skUniversity, 24-hour Holter monitoring of ECG was performed for 124 Poland,e-mail:fi[email protected]. JoannaWdowczyk-Szulc,MartaZ˙arczyn´ska-Buchowiecka,MarcinGrucha- healthysubjects:youngadults:36persons(age18-26),middle- łaandAndrzejRynkiewiczarewithGdan´skMedicalUniversity,Poland. aged adults: 40 persons (age 45-53), elderly: 48 persons (age PROCEEDINGSOFTHE7THESGCO2012,APRIL22-25,2012,KAZIMIERZDOLNY,POLAND 2 65-94). Thus, we have analyzed RR-signals consisting of timeintervalsbetweenconsecutivenormalcardiaccontractions which originated in the sinus node. Multifractalanalysis,similarlyasinthetime-domainstatis- tics, is found to be robust to data removal (see [9]). However, multifractal statistics requires long consecutive series to have access to large pool of scales. Additionally, studied series shouldhaveastablecharacter.Therefore,wecarefullydivided the obtained signals into two parts corresponding to nocturnal and diurnal (daily) activities. The high levels of meanRR, SDNN and pNN50 were assumed to specify sleep hours. For Fig.2. ExamplesofmultifractalspectraobtainedfromRRseries. each RR-signal the period of the six consecutive hours with high levels of these quantities was extracted and labeled as inasignalplaysaveryimportantrole.Itallowstodistinguish sleep. Then, the part corresponding to the typical afternoon additive from multiplicative processes. We have found that activity was labeled as wake. such a distinction can be made for diurnal and nocturnal parts Multifractal analysis relying upon the study of structure of the signals. The main observation is that the nocturnal functionswasperformedwiththewell-knownapproachcalled signals possesses multiplicative character in the VLF region. Wavelet Transform Modulus Maxima (WTMM) [8]. The Such organization is quite weak in the diurnal signals. More- WTMM-based analysis was performed for each person, both over, it was found that the multiplicative features are less and for wake and sleep parts of his/her signals. This was done less pronounced when the age of the investigated individuals not only for the proper signal, but also for the correspond- grows.Hence,weakeningofmultiplicativitycanbeconsidered ing integrated one. The scaling properties were estimated as an indicator of aging. for scales corresponding to the VLF band. Then, after the necessary Legendre transformation, the multifractal spectra were obtained for wake and sleep parts of every individual B. Multifractal age RR-signal. In our analysis we have tested several parameters to char- acterize typical multifractal spectra. In particular, for each person, we have examined: h , H (Hurst), h , h , max left right ∆ – the width (see Fig.1 for the corresponding illustration) in his/herwake and sleep parts of the signals and their integrated counterparts. Tedious statistical analysis for sleep and wake parts of the signals allowed as to formulate the following five criteria. hsleep <hwake, (2) max max hwake,intis large enough(>1.15), (3) max ∆sleep ≈1, (4) max Hsleep <Hwake, (5) Fig. 1. Assessment of a typical multifractal spectrum: hmax – scaling of the most dense subset of a signal; Hurst – correlation; hleft – extreme, Hint,sleepis large enough. (6) andhright –smootheventsscalings;width=hright−hleft –varietyof subsetswithdifferentscalings The notation was explained above and the additional super- scripts seem to be self-evident. These criteria were applied to In Fig. 1 we present the typical multifractal spectrum and all 124 signals obtained from all age groups. collect the main parameters characterizing such a spectrum. Theresultsaresummarizedinthetablebelow.Thefirstrow However, real spectra obtained from cardiac time series differ tellsushowmanypositiveanswersweregiventothequestion considerably from typical ones. Examples of such spectra are whether each (out of 5) criterion is fulfilled. The main body presented in Fig. (2). The evident irregularities (especially in of the table presents the percentage of the number of positive theintegratedsignals)requirespecialcareandthecorrespond- answers obtained for three investigated age groups. ing analysis is quite time-consuming. The details concerning such spectra (examples and their discussion) can be found in Number of positive answers [4]. group name: 0 1 2 3 4 5 young 53 25 11 11 0 0 III. RESULTS middle 3 33 30 20 7 7 A. ∆ in cardiac signals elderly 0 10 10 21 29 29 max Thedistance∆ betweenh inthespectrumobtained The number of positive answers clearly increases with age max max from the integrated signal and h in a spectrum obtained and the table has a characteristic skew form. The skewness of max PROCEEDINGSOFTHE7THESGCO2012,APRIL22-25,2012,KAZIMIERZDOLNY,POLAND 3 the table contents suggests that we can associate the number of positive answers to criteria (2) – (6) as indicators of the multifractal age. IV. CONCLUSIONS Wehaveappliedthemultifractalanalysisto12424-hourRR signalsdivideintowakeandsleepparts.Themainconclusions can be summarized as follows • The skewness of the presented table stems from the careful multifractal analysis of Holter recordings of RR signal obtained from healthy individuals. The number of positive answers whether the criteria are met, clearly increases with age. Thereby, we suggest that the term multifractal age may be useful in the studies of healthy aging. • The quantity ∆max leads to distinction between additiv- ity and multiplicativity of the investigated RR signals. Hence, we suggest that ∆ should be accepted as an max additional parameter characterizing multifractality of the time series. REFERENCES [1] J.Gao,Y.Cao,WW.TungandJ.HuMultiscaleanalysisofcomplextime series.Integrationofchaosandrandomfractaltheory,andbeyond John Wiley&SonsInc.,2007. [2] D.Makowiec,A.Dudkowska,R.Gała¸skaandA.RynkiewiczMultifractal estimatesofmonofractalityinRR-heartseriesinpowerspectrumranges PhysicaA.388:3486–502,2009. [3] D. Makowiec, A. Rynkiewicz, R. Gała¸ska, J. Wdowczyk-Szulc and M. Z˙arczyn´ska - Buchowiecka Reading multifractal spectra: Aging by multifractalanalysisofheartrate EPL.94:68005-p1–p6,2011. [4] D. Makowiec, A. Rynkiewicz, J. Wdowczyk-Szulc, M. Z˙arczyn´ska- BuchowieckaR.Gała¸skaandS.Kryszewski Aginginautonomiccontrol by multifractal studies of cardiac interbeat intervals in the VLF band Physiol.Meas.32:1–19,2011. [5] S.Stoev,MS.Taqqu,Ch. ParkandJS.MarronOnthewaveletspectrum diagnostic for Hurst parameter estimation in the analysis of Internet traffic Comp.Networks48:423–45,2005. [6] JW.Kantelhardt FractalandmultifractaltimeseriesinEncyclopediaof ComplexityandSystemsScienceRA.Meyers(Ed.) Springer,2009. [7] Heartratevariability.Standardsofmeasurement,physiologicalinterpre- tation,andclinicaluseTaskForceoftheEuropeanSocietyofCardiology theNorthAmericanSocietyofPacingandElectrophysiology. Eur.Heart J.17:354–81,1996. [8] JF. Muzy, E. Bacry and A. Arneodo Multifractal formalism for fractal signals: the structure-function approach versus the wavelet-transform modulus-maximamethod Phys.Rev.E.47:875–951993. [9] GD.Clifford ECGStatistics,Noise,ArtifactsandMissingDatapp.55–92 in Advanced Methods and Tools for ECG Data Analysis GD. Clifford, F.AznajeandPE.McSharr(Eds.) ArtechHouseInc.,2006.

See more

The list of books you might like

Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.