New Cardiovascular Indices Based on a Nonlinear Spectral Analysis of Arterial Blood Pressure Waveforms
Taous-Meriem Laleg, Claire Médigue, Yves Papelier, Emmanuelle Crépeau, Michel Sorine
aa r X i v : . [ m a t h - ph ] O c t a p p o r t (cid:13)(cid:13) d e r e c h e r c h e (cid:13) I SS N - I S RN I NR I A / RR -- -- F R + E N G Thème BIO
INSTITUT NATIONAL DE RECHERCHE EN INFORMATIQUE ET EN AUTOMATIQUE
New Cardiovascular Indices Based on a NonlinearSpectral Analysis of Arterial Blood PressureWaveforms
Taous-Meriem Laleg — Claire Médigue — Yves Papelier — Emmanuelle Crépeau —Michel Sorine
N° 6312
Octobre 2007 nité de recherche INRIA RocquencourtDomaine de Voluceau, Rocquencourt, BP 105, 78153 Le Chesnay Cedex (France)
Téléphone : +33 1 39 63 55 11 — Télécopie : +33 1 39 63 53 30
New Cardiovascular Indices Based on a NonlinearSpectral Analysis of Arterial Blood Pressure Waveforms
Taous-Meriem Laleg , Claire M´edigue , Yves Papelier , Emmanuelle Cr´epeau, Michel Sorine
Th`eme BIO — Syst`emes biologiquesProjet SISYPHERapport de recherche n ° Abstract:
A new method for analyzing arterial blood pressure is presented in this report.The technique is based on the scattering transform and consists in solving the spectralproblem associated to a one-dimensional Schr¨odinger operator with a potential dependinglinearly upon the pressure. This potential is then expressed with the discrete spectrumwhich includes negative eigenvalues and corresponds to the interacting components of an N-soliton. The approach is similar to a nonlinear Fourier transform where the solitons play therole of sine and cosine components. The method provides new cardiovascular indices thatseem to contain relevant physiological information. We first show how to use this approachto decompose the arterial blood pressure pulse into elementary waves and to reconstruct itor to separate its systolic and diastolic phases. Then we analyse the parameters computedfrom this technique in two physiological conditions, the head-up 60 degrees tilt test andthe isometric handgrip test, widely used for studying short term cardiovascular control.Promising results are obtained.
Key-words:
Blood pressure, scattering transform, solitons, cardiovascular indices ouveaux indices cardio-vasculaires bas´es sur uneanalyse spectrale non lin´eaire de la forme des ondes depression art´erielle
R´esum´e :
Une nouvelle m´ethode d’analyse du signal de pression art´erielle est propos´eedans ce rapport. La technique est bas´ee sur la transformation scattering et consiste `ar´esoudre un probl`eme spectral associ´e `a l’op´erateur de Schrodinger unidimensionnel avec unpotentiel qui d´epend lin´eairement de la pression. Le potentiel est reconstruit `a l’aide duspectre discret uniquement qui comprend des valeurs propres n´egatives correspondant `a Nsolitons en interaction. L’approche est analogue `a une transform´ee de Fourier o`u les solitonsjouent le rˆole des composantes sinus et cosinus. La m´ethode introduit de nouveaux indicescardio-vasculaires qui semblent contenir des informations physiologiques int´eressantes. Dansun premier temps, nous appliquons la m´ethode `a la d´ecomposition du signal de pressionart´erielle en ondes ´el´ementaires et `a sa reconstruction ou `a la s´eparation de ses partiesdiastolique et systolique. Nous analysons par la suite les param`etres calcul´es `a partir decette technique dans deux conditions physiologiques g´en´eralement utilis´ees pour l’´etude ducontrˆole cardio-vasculaire `a court-terme: test d’inclinaison `a 60 degr´es et exercice isom´etrique(handgrip). Des r´esultats prometteurs ont ´et´e obtenus.
Mots-cl´es :
Pression art´erielle, transformation scattering, solitons, indices cardio-vasculaires ew cardiovascular indices based on a nonlinear spectral analysis of arterial blood pressure waveforms Contents
RR n ° Laleg & M´edigue & Papelier & Cr´epeau & Sorine
The analysis of mean values and beat-to-beat variability of cardiovascular (CV) time serieshas been widely used as a noninvasive approach to study the control of the autonomic ner-vous system (ANS) on the CV function [2], [28]. CV time series usually give information inthe frequency and amplitude domains. The frequency (or period) is given by the RR interval(between two R peaks on the ECG ) or the pulse interval (PI) (between two systolic bloodpressure peaks). The amplitude concerns systolic, diastolic and mean pressures (noted SBP,DBP and MBP respectively). Standard measures of these parameters are mean levels andglobal variability, spectral, temporal and time-frequency analysis [33].Instead of the usual decomposition of the arterial blood pressure (ABP) waveform intoa linear superposition of harmonic waves (sine and cosine functions) [40], [46], in this articlewe propose to use nonlinear superpositions of particular travelling waves, the N-solitons-solutions of the Korteweg-de Vries (KdV) equation where N travelling components are in-teracting. These N-solitons play the role here of the harmonic-waves solutions of the linearwave equation. The concept of soliton refers in fact to a solitary wave emerging unchangedin shape and speed from the collision with other solitary waves [38]. They fascinate scien-tists by their very interesting coherent-structure characteristics and are used in many fieldsto model natural phenomena. Solitons are solutions of nonlinear dispersive equations likethe KdV equation arising in a variety of physical problems, for example to describe wavemotion in shallow water canals [39], [49]. The use of solitons to describe the ABP wasalready introduced in [48] and [35] where a KdV equation and a Boussinesq equation wererespectively proposed as a blood flow model. Recently, in [9], [24] a reduced model of theABP cycle was introduced. The latter consists of a sum of a 2 or 3-soliton solution of a KdVequation, describing fast phenomena during the systolic phase and a 2-element windkesselmodel describing slow phenomena during the diastolic phase. We recall that the systolicphase corresponds to the contraction of the heart, driving blood out of the left ventriclewhile the diastolic phase corresponds to the period of relaxation of the heart.The decomposition of the ABP signal into a nonlinear superposition of solitons intro-duced in this article is based on an elegant mathematical transform: the scattering transformfor a one-dimensional Schr¨odinger equation [6], [12], [14]. The main idea in our utilization ofthis transform consists in interpreting the pressure as a potential able to attract or repulse”fluid particles” or equivalently to transmit or reflect waves associated with them. Thissituation is modelled by a one-dimensional Schr¨odinger operator with a potential dependinglinearly upon the pressure wave [25]. The discrete levels of energy or speed of this systemare given by the discrete spectrum of the Schr¨odinger operator. The associated eigenstatesdescribe some coherent structures that are present in the pressure waveform and expressedas N solitons.The scattering-based signal analysis (SBSA) method introduces a new spectral descrip-tion of the pressure waveform leading to new cardiovascular indices. This study aims to
INRIA ew cardiovascular indices based on a nonlinear spectral analysis of arterial blood pressure waveforms
In this section, we introduce a new signal analysis method based on the scattering theory.We start by briefly recalling the basis of the Direct and Inverse Scattering Transforms (DST& IST). Then, we present the main idea in the SBSA technique. For more details aboutDST and IST the reader can refer to the abundant literature and the references given.
Let V be a given real function in the so-called Faddeev class L ( R ) [1]: L ( R ) = { V ∈ L ( R ) , Z + ∞−∞ | V ( x ) | (1 + | x | ) dx < ∞} . (1)We consider the one dimensional Schr¨odinger operator H ( V ) with a potential V : H ( V ) : ψ → H ( V ) ψ = − ∂ ψ∂x + V ψ. (2)The DST of V will be defined as a function of the solution of the spectral problem for H ( V )where λ and ψ are respectively the eigenvalues and the associate eigenfunctions for somenormalization: H ( V ) ψ = λψ. (3)The spectrum of H ( V ) has two components: a continuous spectrum equal to (0 , + ∞ ) anda discrete spectrum with negative eigenvalues [11], [12], [14]. RR n ° Laleg & M´edigue & Papelier & Cr´epeau & Sorine
For the positive eigenvalues denoted here λ = k , there are eigenfunctions, called scatter-ing solutions of (3) consisting of linear combinations of exp ( ikx ) and exp ( − ikx ) as x → ±∞ .Among them, we consider the Jost solutions from the left f l and from the right f r , normal-ized at ±∞ : H ( V ) f j = k f j , k ∈ R \{ } , j = l, r, (4)exp ( − ikx ) f l ( k, x ) = 1 + o (1) , x → + ∞ , (5)exp ( − ikx ) ∂f l ( k, x ) ∂x = ik + o (1) , x → + ∞ , (6)exp (+ ikx ) f r ( k, x ) = 1 + o (1) , x → −∞ , (7)exp (+ ikx ) ∂f r ( k, x ) ∂x = − ik + o (1) , x → −∞ , (8)We recall that, for each fixed x ∈ R the Jost solutions have analytic extensions in k to theupper-half complex plane [1].The transmission coefficient T and the reflection coefficients R l and R r from the left andfrom the right respectively are defined through the relations: f l ( k, x ) = exp ( ikx ) T ( k ) + R l ( k ) exp ( − ikx ) T ( k ) + o (1) , x → −∞ , (9) f r ( k, x ) = exp ( − ikx ) T ( k ) + R r ( k ) exp ( ikx ) T ( k ) + o (1) , x → + ∞ , (10)These coefficients satisfy: | T ( k ) | + | R l ( k ) | = | T ( k ) | + | R r ( k ) | = 1 . (11)On the other hand, for the negative eigenvalues of the discrete spectrum, equation (3)admits solutions called bound states that belong to L ( R ) in the x variable. When V belongsto the Faddeev class, the bound states solutions of (3) decay exponentially as x → ±∞ andtheir number N is finite [1]. Let us denote λ n = − κ n with λ ≤ λ ≤ ... and ψ n the N negative eigenvalues and L -normalized bound states: H ( V ) ψ n = − κ n ψ n , Z + ∞−∞ | ψ n ( x ) | dx = 1 , n = 1 , · · · , N. (12)The eigenspaces being of dimension 1, the bound states and the Jost solutions are propor-tional: ψ n ( x ) = c ln f l ( iκ n , x ) = ( − N − n c rn f r ( iκ n , x ) , (13)where c ln and c rn are called the bound-state norming constants and are defined by: c jn := [ Z + ∞−∞ | f j ( iκ n , x ) | dx ] − , j = l, r. (14) INRIA ew cardiovascular indices based on a nonlinear spectral analysis of arterial blood pressure waveforms V as the sets of scattering data from the left, S l ( V ) or from theright, S r ( V ): S j ( V ) := { R j , κ n , c ¯ jn , n = 1 , · · · , N } , { j, ¯ j } = { l, r } . (15)The potential can be uniquely reconstructed by using any one of these sets. The solutionof this inverse problem, called IST, is the object of many studies concerned with specificclasses of potentials [1], [7], [11], [15], [29]. Two transforms are then available, S and S − (in the sequel we choose j = r and drop the subscripts r and l for simplicity).In this study, we will use the special class of reflectionless potentials for which the left orright reflection coefficients are zero. Such potentials can be constructed as follows: let Π d be the projector zeroing the R -component of S ( V ), then S − ◦ Π d ◦ S ( V ) is reflectionlessfor any V in the Faddeev class. There are useful explicit representations of reflectionlesspotentials using only the discrete spectrum as in the following theorem [14]: Theorem : If V is reflectionless for H ( V ), then: V ( x ) = − N X n =1 κ n ψ n ( x ) , x ∈ R . (16) V can be also written: V ( x ) = − ∂ (log (det( I + A ))) ∂x , x ∈ R , (17)where A is an N × N matrix: A ( x ) = [ c m c n κ m + κ n exp (( κ m + κ n ) x )] , n, m = 1 , · · · , N. (18)Note that in (17) and (18), the potential is entirely defined with 2 N parameters namely κ n and c n , n = 1 , · · · , N .A very close relation between a soliton solution of a KdV equation and a reflectionlesspotentiel of the Schr¨odinger operator was introduced in [14]. In fact these potentials remainreflectionless when evolving in time and space according to a KdV equation. For t → + ∞ , N κ n , c n ) such that 4 κ n gives thespeed of the soliton and c n its position. Therefore each component − κ n ψ n in the sum (16)refers to a single soliton. We now present how to use IST in a seemingly new method to analyse pulse-like signals ofthe Faddeev class.
RR n ° Laleg & M´edigue & Papelier & Cr´epeau & Sorine
The main idea in the SBSA approach is to interpret a positive signal y in the Faddeevclass as a quantum well by changing the sign, and to tune the depth of this well with apositive parameter χ in order to approximate y by a coherent state y χ . For a deeper wellthe trapped energy will be higher and the approximation better, as we will prove. Theestimate is then obtained by filtering out the nonlinear reflections: y χ = − χ S − ◦ Π d ◦ S ( − χy ) . (19)A convenient explicit formula is available, χy χ being a reflectionless potential: y χ = 4 χ N χ X n =1 κ χ,n ψ χ,n , (20)where − κ χ,n and ψ χ,n , n = 1 , · · · , N χ are the negative eigenvalues and the associated L -normalized eigenfunctions for H ( − χy ).Then, we look for a value ˆ χ for the parameter χ such that the signal y is well approxi-mated by y ˆ χ . This is the decomposition of the signal y into the nonlinear superposition ofsolitons announced in [25].It is well-known that the number of negative eigenvalues N χ of H ( − χy ) is a nondecreasingfunction of χ and there is an infinite unbounded sequence ( χ n ) such that N χ n = N χ n − + 1[25], [31]. Determining the parameter ˆ χ determines the number of negative eigenvalues andhence the number of solitons components required for a satisfying approximation of thesignal y . Fig. 1 summarizes the SBSA technique. The scattering transform has an infinite number of invariants which are related to the KdVconserved quantities [14], [32]. Let us denote these invariants I m ( V ), m = 0 , , , · · · . Theyare of the form (we take − V as argument having (19) in mind): I m ( − V ) = ( − m +1 m + 12 m +2 Z + ∞−∞ P m ( V, ∂V∂x , ∂ V∂x , · · · ) dx, (21)where P m , m = 0 , , , · · · are known polynomials in V and its successive derivatives withrespect to x ∈ R [6].A general formula relating I m ( − V χ ), with V χ = − χy , to the scattering data of H ( V χ )can be deduced; see for example [6], [16], [32]: I m ( χy )= N χ X n =1 κ m +1 χ,n +2 m +12 π Z + ∞−∞ ( − k ) m ln ( | T χ ( k ) | ) dk, (22) INRIA ew cardiovascular indices based on a nonlinear spectral analysis of arterial blood pressure waveforms m = 0 , , , · · · .We introduce the Riesz means of the negative eigenvalues λ n of H ( V ) such that λ n ≤ λ ≤ S γ,λ ( − V ) = X λ n ≤ λ | λ n | γ , γ ≥ . (23)Remark that S ,λ ( V ) is the number of eigenvalues of H ( V ) smaller than λ . RR n ° Laleg & M´edigue & Papelier & Cr´epeau & Sorine
For an N χ -soliton, for instance − χy χ of the previous subsection, the invariants onlydepend on the discrete spectrum and they are related to the Riesz means as follows: I m ( χy χ ) = S γ, ( χy χ ) , γ = m + 12 , m = 0 , , , · · · (24)A ”sum rule” is then verified by the invariants of χy and χy χ : I m ( χy ) = I m ( χy χ )+2 m + 12 π Z + ∞−∞ ( − k ) m ln ( | T χ ( k ) | ) dk, (25) m = 0 , , , · · · .In this article we are only interested in the two first invariants ( m = 0 and m = 1)corresponding to the conservation of mass and momentum for the KdV flows. Here it issufficient to see them as invariants of the DST, in the same manner energy is invariant forthe Fourier transform (Plancherel’s theorem). We will show later in the application of theSBSA to the ABP that these two invariants are related to some important cardiovascularparameters.So, for m = 0, P ( V χ , · · · ) = V χ , we get with (21) and (25): Z + ∞−∞ ydx = Z + ∞−∞ y χ dx + 2 πχ Z + ∞−∞ ln ( | T χ ( k ) | ) dk. (26)For m = 1, P ( V χ , · · · ) = V χ , we have with (21) and (25): Z + ∞−∞ y dx = Z + ∞−∞ y χ dx − πχ Z + ∞−∞ k ln ( | T χ ( k ) | ) dk. (27)Equation (27) is known as the Buslaev-Faddeed-Zakharov trace formula. Proposition : Let y : R → R be a continuous non-negative function with a compactsupport, then we have the convergence of the estimates of the first two invariants:lim χ → + ∞ I m ( y χ ) = I m ( y ) , m = 0 , . (28)Proof: We can apply the results on the Lieb-Thirring semiclassical limit of the Riesz means[4], [20], [21], [26]: lim χ → + ∞ S γ, ( χy ) χ + γ = L clγ Z R y ( x ) + γ dx, γ ≥ , (29)where L clγ is the so-called Lieb-Thirring constant given by: L clγ ≡ (4 π ) − Γ( γ + 1)Γ( γ + 32 ) . (30) INRIA ew cardiovascular indices based on a nonlinear spectral analysis of arterial blood pressure waveforms γ = 12 :lim χ → + ∞ S , ( χy ) χ = L cl Z R y ( x ) dx, L cl = 14 . (31)So, we deduce the convergence of the first invariant estimate.For γ = 32 we have an analog of the Plancherel identity for the Fourier transform:lim χ → + ∞ S , ( χy ) χ = L cl Z R y ( x ) dx, L cl = 316 . (32)Therefore, we get the convergence of the second invariant. In the previous subsection, we presented a new signal analysis method based on the scatter-ing transform. Now, we propose to use this method for ABP analysis. For convenience wereplace the space variable x by the time variable t .We note the ABP signal P ( t ) and the estimated pressure with the SBSA technique ˆ P ( t )such that: ˆ P ( t ) = 4 χ N χ X n =1 κ χ,n ψ χ,n ( t ) , (33)where − κ χ,n and ψ χ,n , n = 1 , · · · , N χ are the N χ negative eigenvalues and associated L − normalized eigenfunctions of H ( − χP ). We recall that each component 4 κ χ,n ψ χ,n in(33) refers to a single soliton [14], [25].In Fig. 2 and Fig. 3, measured and reconstructed pressures at the aorta and at the fingerlevels are presented respectively. The aortic pressure was measured using a catheter while aFinapres was used to measure the pressure at finger. Only 5 to 10 components are sufficientfor a good reconstruction of the ABP waveform. Following the work done in [9], [24], we propose here using the SBSA technique to separatethe pressure into fast and slow parts corresponding respectively to the systolic and diastolicphases. Indeed a reduced model of ABP has been proposed in [9], [24]. The latter consists of
RR n ° Laleg & M´edigue & Papelier & Cr´epeau & Sorine
ABP ( mm H g ) Reconstructed pressureMeasured pressure
Figure 2: Measured and reconstructed pressure at the aorta
ABP ( mm H g ) Reconstructed pressureMeasured pressure
Figure 3: Measured and reconstructed pressure at the fingera sum of two terms: a 2 or 3-soliton solution of a KdV equation describing fast phenomenawhich predominate during the systolic phase and a 2-element windkessel model describingslow phenomena during the diastolic phase. As noticed in the previous section, the SBSAtechnique decomposes the ABP signal into a sum of solitons, each one characterized by itsvelocity given by the discrete eigenvalues − κ χ,n . So the largest κ χ,n , n = 1 , · · · , N s describe INRIA ew cardiovascular indices based on a nonlinear spectral analysis of arterial blood pressure waveforms N s = 2 or 3. We note ˆ P s and ˆ P d the estimated systolic and diastolic pressuresrespectively such that:ˆ P s ( t ) = 4 χ N s X n =1 κ χ,n ψ χ,n , ˆ P d ( t ) = 4 χ N χ X n = N s +1 κ χ,n ψ χ,n . (34)We compute the first two invariants of these partial pressures with the Riesz means for thechosen cut-off speed λ : IN V ( λ ) = 4 χ S ,λ ( χP ) , IN V ( λ ) = 163 χ S ,λ ( χP ) , (35)We can now define the proposed invariants for the whole beat and for the systolic anddiastolic phases ( IN V j , IN V S j , IN V D j , j = 1 , IN V j = IN V j (0) ,IN V S j = IN V j ( λ s ) , j = 1 , ,IN V D j = IN V j (0) − IN V S j ( λ s ) , (36)where λ s = λ χ, or λ χ, .In Fig. 4 and Fig. 5, we represent the measured pressure and the estimated systolic anddiastolic parts respectively. We remark that ˆ P s and ˆ P d are respectively localized during thesystole and the diastole, as expected. The SBSA technique provides a new description of the ABP signal using the DST. As seenin the previous section, the reconstruction of the signal by IST from its spectral data givesgood results. Instead of reconstructing the original signal, it is possible to modify the spec-trum leading to some kind of filtering. This is illustrated by the separation of the systolicand diastolic phases. Moreover, the SBSA method introduces new parameters. The firsttwo global invariants
IN V and IN V are respectively, by definition, the usual mean bloodpressure (MBP) and the less usual, but easy to compute directly, integral of the square ofthe pressure. The first systolic invariant IN V S is a new index: we think that it can becorrelated to SV. Remark that if IN V is easy to compute directly, the ”fast part” of thisintegral, the second systolic invariant IN V S is a new less obvious index and might con-tain information on ventricular contractility. SV and contractility are in fact parameters ofgreat interest that are difficult to measure routinely, as they require invasive or sophisticated RR n ° Laleg & M´edigue & Papelier & Cr´epeau & Sorine
ABP ( mm H g ) Reconstructed systolic pressureMeasured pressure
Figure 4: ˆ P s and fast systolic phenomena ABP ( mm H g ) Reconstructed diastolic pressureMeasured pressure
Figure 5: ˆ P d and slow diastolic phenomenatechniques. For instance SV can be estimated by invasive nuclear ventriculography [47], 2Dechocardiography [23], radionuclide monitoring [43], impedance cardiography [8], [10]. Onlyone evaluation of SV from ABP has been proposed [19]. Ventricular contractility is assessedby the mean of the tissue doppler echocardiography [17], [18]. INRIA ew cardiovascular indices based on a nonlinear spectral analysis of arterial blood pressure waveforms χ and the eigenvalues also change. This fact is illustrated in Fig.6. Therefore, the eigenvalues could be used to assess the baroreflex sensitivity (BRS) in acertain way. In fact, the BRS expresses the variation of the heart beat interval in response toeach arterial pressure variation. The BRS concept was first based on drug-induced responsesof ABP and heart period [37]. Then various time [27], [34], [36] and spectral [5], [28], [34],domain methods [27] were compared.In this section we analyse the SBSA parameters in two situations, devoted to the assess-ment of the ANS control of the CV system. We restrict the study to the modulus of thefirst two negative eigenvalues | λ χ, | = | λ | , | λ χ, | = | λ | and the first two invariants (global,systolic and diastolic). We include in the analysis some classical parameters which are PI,SBP and DBP. The head-up tilt test is mainly used for vasovagal syncopes diagnosis, characterized by anautonomic dysfunction. It consists in the orthostatic transition from the supine to thestanding positions. This leads to a redistribution of the venous blood volume, from theintrathoracic region towards the venous volume in the leg and lower abdominal veins. Thisleads to a decrease in SV and PI and an increase in SBP [10], [41], [42], [44].A group of 15 healthy subjects under 0 . Hz paced breathing, already studied [3], wasconsidered. The table was rotated to an upright position at 60 degrees. The continuousABP was measured at the finger using a Finapres device [45]. The two positions, supineand standing, were compared using the Wilcoxon non parametric paired test.Fig. 7 shows the time series of PI, SBP, IN V S and | λ | in the supine and standingpositions. Mean levels of the ABP parameters are presented in Table 1. We notice thatsignificant differences between the supine and standing positions appear for PI, DBP, | λ | , | λ | and IN V S while for the other parameters the differences are not significant. So afterthe tilt test, SBP and MBP recover their prior values (preceding the tilt phase) whereas IN V S is decreased. If IN V S is related to SV then this means that SV remains decreasedafter the tilt test and this can be explained by the decrease in the venous return.The ABP analysis during the transition from the supine to the standing positions waspossible for only nine subjects. For these, we assessed the linear relation between PI of thebeat n + 1 and | λ | computed for the beat n . In Fig. 8, we note that as PI decreases, | λ | increases and that the linear correlation between PI and | λ | is stronger around thetransition than in supine position. Table 2 shows that the correlation coefficient ( R ) andthe slope, compared by a one-way repeated measures analysis of variance, were significantly RR n ° Laleg & M´edigue & Papelier & Cr´epeau & Sorine
Figure 6: (a) First eigenvalue time series of a healthy subjects under 0 . Hz paced-breathingduring 29 beats. The signal exhibits a beat-to-beat variability that is twice as fast as thebreathing rate. (b) A zoom on a few beats shows morphological changes in the blood pressuresignal.stronger around the transition than in supine or standing positions. Moreover, a comparisonwith usual BRS indices, SBP and pulse pressure (PP), shows that | λ | has the strongestcorrelation with PI (Table 3). It is not surprising that | λ | is more informative about therelation between the heart period and the ABP because it reflects the arterial waveform andnot only one (SBP) or two (PP) samples of this waveform. The isometric handgrip exercise is mainly used for the evaluation of a non appropriate ANSbehavior, that mimics actual conditions of professional or domestic exposure, with arterial
INRIA ew cardiovascular indices based on a nonlinear spectral analysis of arterial blood pressure waveforms (a) (b) Figure 7: Time series of ABP parameters in a healthy subject under 0 . Hz paced breathing,in supine (a) and standing (b) positions after a 60 degrees tilt test. PI and IN V S arereduced whereas | λ | is increased in the standing position, more than two minutes after thetilt test. SBP has the same level in both positions.hypertension. Indeed, the ANS acts in the same way as in a dynamic exercise usually char-acterized by an increase in muscle oxygen needs. The isometric exercise is a form of exerciseinvolving the static contraction of a muscle without any visible movement in the angle ofthe joints [13].A group of 13 healthy subjects was considered. The continuous ABP was measured atthe finger with a Finapres device [45]. The two conditions: at rest and during the handgrip RR n ° Laleg & M´edigue & Papelier & Cr´epeau & Sorine
Table 1: ABP parameters during tilt protocol in 15 healthy subjects
ABP parameters supine standing probabilityDirectPI 918 ±
33 778 ±
21 ***SBP 121 ± ± ± ± ±
131 1965 ±
137 ***Second 1205 ±
86 1612 ±
92 ***First invariantsGlobal 77 ± ± ± ± ± ± ±
975 6833 ±
711 NSSystolic 2705 ±
328 2566 ±
247 NSDiastolic 4336 ±
648 4276 ±
465 NSData are expressed as means and SEM; **: P ≤ .
01; ***: P ≤ . Table 2: Beat-to-beat BRS during tilt protocol supine tilt standing probability R . ± . . ± . . ± .
079 ***slope − . ± . − . ± . − . ± .
015 ** R and slope of the linear regression between | λ ( n ) | and P I ( n + 1), over about 60 beats, in 9 healthysubjects. Data are expressed as means and SEM; ** p ≤ .
01; *** p ≤ . R are the strongest during tilt. Table 3: Three indices of beat-to-beat BRS during tilt R tilt | λ | . ± . . ± . . ± . R of the linear regression between ABP parameters ( | λ | , SBP, PP) and P I , over about 60 beats,in 9 healthy subjects of the tilt test. Data are expressed as means and SEM. | λ | is the most stronglycorrelated with PI. test, were compared using the Wilcoxon non parametric paired test.In Fig. 9, the time series of PI, SBP, IN V S and | λ | are presented at rest and duringthe handgrip test. Table 4 illustrates the mean levels of the ABP parameters. We notice thatwhile all the first invariants do not change significantly, the second invariants are sensitively INRIA ew cardiovascular indices based on a nonlinear spectral analysis of arterial blood pressure waveforms P I ( n + ) m s First eigenvalue (n)supinetilt
Figure 8: Beat-to-beat BRS, represented as the relation between | λ ( n ) | and P I ( n + 1).Slope and correlation are stronger during tilt.increased during handgrip.The voluntary central command, involved in the handgrip, synchronously activates themotor and CV systems, leading first to an increase in the heart rate, followed by an increasein the ABP [22], [30]. So, as expected, PI decreases and SBP increases (Table 4). Moreover, IN V S might inform us that SV does not change while IN V S might inform us about theincreasing contractility, as if the heart tries to eject the same quantity of blood during asmaller period. Such a result evokes the treppe effect (or frequency-force relation), where anincrease in heart rate indirectly induces an increase in contractility.As in the case of the tilt test, we study the relationship between PI and | λ | . Fig. 10illustrates the relation between PI of the beat n + 1 and | λ | computed for the beat n at restand during the handgrip test. The strong linear correlation between PI and | λ | is the sameat rest and during the handgrip, but the slope is significantly lower during the handgrip(Table 5). Moreover, a comparison with usual BRS indices, SBP and PP, shows that | λ | has the strongest correlation with PI (Table 6). This result, as obvious as in the case of thetilt test leads us to consider | λ | as a promising index that can be used to study the relationbetween the heart period and ABP. RR n ° Laleg & M´edigue & Papelier & Cr´epeau & Sorine (a) (b) nbbeat10003000300075001302206501100 PI (ms)SBP (mmHg)first eigenvalue 401 1 30second systolic invariant
Figure 9: Time series of ABP parameters in a healthy subject under spontaneous breathing,at rest (a) and during the handgrip test (b). PI is reduced whereas SBP,
IN V S and | λ | are strongly increased during handgrip.Table 4: ABP parameters during handgrip protocol in 13 healthy subjects ABP parameters rest handgrip probabilityDirectPI 953 ±
59 748 ±
42 ***SBP 148 ± ± ± ± ±
160 2325 ±
249 **Second 1186 ±
125 1872 ±
180 ***First invariantsGlobal 96 ± ± ± ± ± ± ±
866 12886 ± ±
344 5013 ±
470 **Diastolic 6283 ±
526 7872 ±
727 **Data are expressed as means and SEM; ** p ≤ .
01; *** p ≤ . ew cardiovascular indices based on a nonlinear spectral analysis of arterial blood pressure waveforms rest handgrip probability R . ± . . ± .
050 NSslope − . ± . − . ± .
024 ** R and slope of the linear regression between | λ ( n ) | and P I ( n + 1), over about 40 beats, in 13 healthysubjects. Data are expressed as means and SEM; ** p ≤ .
01; *** p ≤ . R is not significantly different. Table 6: Three indices of beat-to-beat BRS during handgrip protocol R rest handgrip | λ | . ± . . ± . . ± . . ± . . ± . . ± . R of the linear regression between ABP parameters ( | λ | , SBP, PP) and P I , over about 40 beats,in the 13 healthy subjects of the handgrip test. Data are expressed as means and SEM. | λ | is the moststrongly correlated with PI. First eigenvalue (n)600 P I ( n + ) m s Figure 10: Beat-to-beat BRS, represented as the relation between | λ ( n ) | and P I ( n + 1).Slope is lower during handgrip. RR n ° Laleg & M´edigue & Papelier & Cr´epeau & Sorine
This article deals with a new ABP analysis method based on the scattering theory. ThisSBSA method can be thought of as a nonlinear Fourier analysis for pulse-like signals. Itallows analysis and precise reconstruction as is shown by the very good agreement betweenreal and estimated pressures. We have also presented an application to a filtering problemconsisting in separating the systolic and diastolic phases. Then, we have introduced newcardiovascular indices computed with the SBSA method. These parameters include the firsttwo systolic invariants and we think that they might give information on the variation ofthe stroke volume and the ventricular contractility, that are difficult to measure routinely.Another interesting parameter is the first eigenvalue which seems to reflect the BRS in acertain way. The results obtained from the analysis of two widely used physiological condi-tions are promising and we are now working on the validation of the advanced hypotheses.
INRIA ew cardiovascular indices based on a nonlinear spectral analysis of arterial blood pressure waveforms References [1] T. Aktosun and M. Klaus. Chapter 2.2.4, inverse theory: problem on the line. In
E.R.Pike and P.C. Sabatier, Scattering , pages 770–785. Academic Press, London, 2001.[2] M. Baumert, L. Brechtel, J. Lock, M. Hermsdorf, R. Wolff, V. Baier, and A. Voss. Heartrate variability, blood pressure variability, and baroreflex sensitivity in overtrained ath-letes.
Clin. J. Sport. Med. , 15(5):412–417, September 2006.[3] J. Bestel, L. Mangin, C. M´edigue, A. Monti, and M. Sorine. Proposition of an indexof the autonomic nervous system activity in the arterial blood pressure short termregulation.
Blood Pressure Variablity in Health and Disease, Satellite Symposium of theXVII e Scientific Meeting of the International Society of Hypertension , 12/Suppl. 1 ofthe Fundamental & Clinical Pharmacology journal, Nancy, June 1998.[4] Ph. Blanchard and J. Stubbe. Bound states for Schr¨odinger hamiltonians: phase spacemethods and applications.
Rev. Math. Phys. , 35:504–547, 1996.[5] E. J Bowers and A. Murray. Effects on baroreflex sensitivity measurements when differ-ent protocols are used to induce regular changes in beat-to-beat intervals and systolicpressure.
Physiol. Meas. , 25:523–538, 2004.[6] F. Calogero and A. Degasperis.
Spectral Transform and Solitons . North Holland, 1982.[7] K. Chadan and P. C. Sabatier.
Inverse problems in quantum scattering theory . Springer,New York, 2 nd edition, 1989.[8] A. Charloux, E. Lonsdorfer-Wolf, R. Richard, E. Lampert, M. Oswald-Mammosser,B. Mettauer, B. Geny, and J. Lonsdorfer. A new impedance cardiograph device for thenon-invasive evaluation of cardiac output at rest and during exercise: comparison withthe direct Fick method. Eur. J. Appl. Physiol. , 82(4):313–320, July 2000.[9] E. Cr´epeau and M. Sorine. A reduced model of pulsatile flow in an arterial compartment.
Chaos Solitons & Fractals , 34:594–605, 2007.[10] T. L. Culham and G. K. Savard. Hemodynamic strategies in blood pressure regulationduring orthostatic challenge in women.
Revue canadienne de physiologie appliqu´ee ,22(4):351–367, 1997.[11] P. A. Deift and E. Trubowitz. Inverse scattering on the line.
Communications on Pureand Applied Mathematics , XXXII:121–251, 1979.[12] W. Eckhaus and A. Vanharten.
The Inverse Scattering Transformation and the Theoryof Solitons . North-Holland, 1983.[13] U. Freyschuss. Cardiovascular adjustment to somatomotor activation. The elicitationof increments in heart rate, aortic pressure and venomotor tone with the initiation ofmuscle contraction.
Acta Physiol Scand Suppl , 342:01–63, 1970.
RR n ° Laleg & M´edigue & Papelier & Cr´epeau & Sorine [14] C. S. Gardner, J. M. Greene, M. D. Kruskal, and R. M. Miura. Korteweg-de Vriesequation and generalizations VI. Methods for exact solution. In
Communications onpure and applied mathematics , volume XXVII, pages 97–133. J.Wiley & sons, 1974.[15] I. M. Gel’fand and B. M. Levitan. On the determination of a differential equation fromits spectral function.
Amer. Math. Soc. Transl. , 2(1):253–304, 1955.[16] F. Gesztesy and H. Holden. Trace formulas and conservation laws for nonlinear evolutionequations.
Reviews in Mathematical Physics , 6(1):51–95, 1994.[17] J. Gorcsan, D. P. Strum, W. A. Mandarino, V. K. Gulati, and M. R. Pinsky. Quan-titative assessment of alterations in regional left ventricular contractility with color-coded tissue doppler echocardiography. Comparison with sonomicrometry and pressure-volume relations.
Circulation , 95:2423–2433, 1997.[18] N. L. Greenberg, M. S. Firstenberg, P. L. Castro, M. Main, A. Travaglini, J. A. Od-abashian, J. K. Drinko, L. L. Rodriguez, J. D. Thomas, and M. J. Garcia. Doppler-derived myocardial systolic strain rate is a strong index of left ventricular contractility.
Circulation , 105:99–105, 2002.[19] M. P. M. Harms, K. H. Wesseling, F. Pott, M. Jenstrup, J. Van Goudoever, N. H.Secher, and J. J. Van Lieshout. Continuous stroke volume monitoring by modellingflow from non-invasive measurement of arterial pressure in humans under orthostaticstress.
Clinical Science , 97:291–301, 1999.[20] B. Helffer and D. Robert. Riesz means of bound states and semiclassical limit connectedwith a Lieb-Thirring conjecture II.
Ann. de l’Inst. H. Poincare , 53(2):139–147, 1990.[21] B. Helffer and D. Robert. Riesz means of bound states and semiclassical limit connectedwith a Lieb-Thirring’s conjecture I.
Asymptotic Analysis , 3:91–103, 1990.[22] S. F. Hobbs. Central command during exercise: parallel activation of the cardiovascularand motor systems by descending command signals. In
Circulation, Neurobiology, andBehavior , pages 2176–232. 1982.[23] D. Kozlowski, P. Byrdziak, W. Krupa, M. Gawrysiak, G. Piwko, J. Kubica, andG. Swiatecka. Left ventricle systolic volume in vasovagal syncope patients.
Folia Mor-phol (Warsz) , 62(3):175–178, 2003.[24] T. M. Laleg, E. Cr´epeau, and M. Sorine. Separation of arterial pressure intoa nonlinear superposition of solitary waves and a windkessel flow.
To ap-pear in Biomedical Signal Processing and Control Journal . Available online athttp://dx.doi.org/10.1016/j.bspc.2007.05.004.[25] T. M. Laleg, E. Cr´epeau, and M. Sorine. Travelling-wave analysis and identification. Ascattering theory framework. In
Proc. European Control Conference ECC, Kos, Greece ,July 2007.
INRIA ew cardiovascular indices based on a nonlinear spectral analysis of arterial blood pressure waveforms
ActaMathematica , 184(1):87–111, 2000.[27] R. D. Lipman, J.K. Salisbury, and J. A. Taylor. Spontaneous indices are inconsistentwith arterial baroreflex gain.
Hypertension , 42(4):481–487, 2003.[28] L. Mangin, A. Monti, C. M´edigue, I. Macquin-Mavier, M. E. Lopes, P. Gueret, A. Cas-taigne, B. Swynghedauw, and P. Mansier. Altered baroreflex gain during voluntarybreathing in chronic heart failure.
European Journal of Heart Failure , 3:189–195, 2001.[29] V. A. Marchenko.
Strum-Liouville operators and applications . Birkh¨auser, Basel, 1986.[30] D. I. McCloskey. Centrally-generated commands and cardiovascular control in man.
Clin. Exp. Hypertens. , 3(3):369–378, 1981.[31] P. D. Miller and S. R. Clarke. An exactly solvable model for the interaction of linearwaves with Korteweg-de Vries solitons.
SIAM J. MATH. ANAL. , 33(2):261–285, 2001.[32] S. Molchanov, M. Novitskii, and B. Vainberg. First KdV integrals and absolutelycontinuous spectrum for 1-d Schr¨odinger operator.
Commun. Math. Phys , 216:195–213, 2001.[33] A. Monti, C. M´edigue, and L. Mangin. Instantaneous parameter estimation in cardio-vascular time series by harmonic and time-frequency analysis.
IEEE Transactions onBiomedical Engineering , 49(12):1547–1556, December 2002.[34] A. Monti, C. M´edigue, H. Nedelcoux, and P. Escourrou. Cardiovascular autonomic con-trol during sleep in normal subjects.
European Journal of Applied Physiology , 87(2):174–181, 2002.[35] J. F. Paquerot and M. Remoissenet. Dynamics of nonlinear blood pressure waves inlarge arteries.
Physics Letters A , pages 77–82, October 1994.[36] J. Parlow, J. P. Viale, G. Annat, R. Hughson, and L. Quintin. Spontaneous car-diac baroreflex in humans. Comparison with drug-induced responses.
Hypertension ,25(5):1058–1068, May 1995.[37] R. J. Parmer, J. H. Cervenka, and R. A. Stone. Baroreflex sensitivity and heredity inessential hypertension.
Circulation , 85:497–503, 1992.[38] M. Remoissenet.
Waves called solitons, concepts and experiments . Springer, 3 rd edition,1999.[39] A. C. Scott, F. Y. F. Chu, and D. W. McLaughlin. The soliton: A new concept inapplied science. Proceedings of the IEEE , 61(10):1443–1483, October 1973.[40] P. Segers and P. Verdonck.
Principles of Vascular Physiology , chapter 6, pages 116–137.Panvascular Medecine. Integrated Clinical Managements. Springer Verlag, 2002.
RR n ° Laleg & M´edigue & Papelier & Cr´epeau & Sorine [41] J. K. Shoemaker, C. S. Hogeman1, and L. I. Sinoway. Contributions of msna andstroke volume to orthostatic intolerance following bed rest.
Am. J. Physiol. Regul.Integr. Comp. Physiol. , 277:R1084–R1090, 1999.[42] J. Siebert, P. Drabik, R. Lango, and K. Szyndler. Stroke volume variability and heartrate power spectrum in relation to posture changes in healthy subjects.
Medical ScienceMonitor , 10(2):MT31–MT37, 2004.[43] B. Takase, H. Hosaka, K. Kitamura, A. Uehata, K. Satomura, K. Isojima, S. Kosuda,S. Kusano, A. Kurita, and F. Ohsuzu. The repeatability of left ventricular volumeassessment by a new ambulatory radionuclide monitoring system during head-up tilt.
Jpn. Heart J. , 42(6):749–758, November 2001.[44] I. Taneja, C. Moran, M. S. Medow, J. L. Glover, L. D. Montgomery, and J. M. Stewart.Differential effects of lower body negative pressure and upright tilt on splanchnic bloodvolume. m. J. Physiol. Heart. Circ. Physiol. , 292:H1420–H1426, 2007.[45] K. H. Wesseling. Method and device for controlling the cuff pressure in measuring theblood pressure in a finger by means of photo-electric plethysmograph.
United StatePatent , September 1985.[46] N. Westerhof, P. Sipkema, G. C. Van Den Bos, and G. Elzinga. Forward and backwardwaves in the arterial system.
Cardiovascular Research , 6:648–656, 1972.[47] M. F. Wilson, B. H. Sung, G. A. Pincomb, and W. R. Lovallo. Simultaneous mea-surement of stroke volume by impedance cardiography and nuclear ventriculography:comparisons at rest and exercise.
Ann. Biomed. Eng. , 17(5):475–482, 1989.[48] S. Yomosa. Solitary waves in large vessels.
Journal of the Physical Society of Japan ,50(2):506–520, February 1987.[49] N. J. Zabusky and M. D. Kruskal. Interation of ’soliton’ in a collisionless plasma andthe recurrence of initial states.
Physical Review Letters , 15(6):240–243, 1965.
INRIA nité de recherche INRIA RocquencourtDomaine de Voluceau - Rocquencourt - BP 105 - 78153 Le Chesnay Cedex (France)
Unité de recherche INRIA Futurs : Parc Club Orsay Université - ZAC des Vignes4, rue Jacques Monod - 91893 ORSAY Cedex (France)Unité de recherche INRIA Lorraine : LORIA, Technopôle de Nancy-Brabois - Campus scientifique615, rue du Jardin Botanique - BP 101 - 54602 Villers-lès-Nancy Cedex (France)Unité de recherche INRIA Rennes : IRISA, Campus universitaire de Beaulieu - 35042 Rennes Cedex (France)Unité de recherche INRIA Rhône-Alpes : 655, avenue de l’Europe - 38334 Montbonnot Saint-Ismier (France)Unité de recherche INRIA Sophia Antipolis : 2004, route des Lucioles - BP 93 - 06902 Sophia Antipolis Cedex (France)ÉditeurINRIA - Domaine de Voluceau - Rocquencourt, BP 105 - 78153 Le Chesnay Cedex (France)