This article was downloaded by: [California State University of Fresno]

On: 29 November 2012, At: 16:12

Publisher: Taylor & Francis
Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office:
Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK

Contemporary Physics
Publication details, including instructions for authors and subscription
information:
http://www.tandfonline.com/loi/tcph20

Physics of the human cardiovascular system

Aneta Stefanovska
Version of record first published: 08 Nov 2010.

To cite this article: Aneta Stefanovska (1999): Physics of the human cardiovascular system, Contemporary
Physics, 40:1, 31-55

To link to this article: http://dx.doi.org/10.1080/001075199181693

PLEASE SCROLL DOWN FOR ARTICLE

Full terms and conditions of use: http://www.tandfonline.com/page/terms-and-conditions

This article may be used for research, teaching, and private study purposes. Any substantial
or systematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or
distribution in any form to anyone is expressly forbidden.

The publisher does not give any warranty express or implied or make any representation that the
contents will be complete or accurate or up to date. The accuracy of any instructions, formulae, and
drug doses should be independently verified with primary sources. The publisher shall not be liable
for any loss, actions, claims, proceedings, demand, or costs or damages whatsoever or howsoever
caused arising directly or indirectly in connection with or arising out of the use of this material.
 Contem porary Physics, 1999, volume 40, number 1, pages 31 ± 55

Physics of the human cardiovascular system

A NETA S TEFANOV SKA and M AJA B RACÆICÆ

C ontemporary m easurement techniques permit the non-invasive observation of several

cardiovascular functions, both from the central and peripheral points of view . W e show that,
Downloaded by [California State University of Fresno] at 16:12 29 November 2012

w ithin one cycle of blood through the cardiovascular system, the sam e dynam ics
characterizes heart function as well as blood ¯ ow in the capillary bed where cells exchang e
energy and matter. Analyses of several quite diŒerent signals derived from respiration,
cardiac function and blood ¯ ow, all reveal the existence of ® ve alm ost periodic frequency
com ponents. T his result is interpreted as evidence that cardiovascular dyna mics is governed
by ® ve coupled oscillators. T he couplings provide co-ordination am ong the physiological
processes involved, and are essential for e cient cardiovascular function. Understandin g the
dynam ics of a system of ® ve coupled oscillators not only represents a theoretical challenge,
but also carries practical implications for diagnos is and for predicting the future behaviour
of this life giving system .

1. Basic role and structure
In the course of evolution, individual cells organized into
cellular system s of increasing com plexity and, as animals
evolved, they further diŒerentiated into specialized tissues
and organs. At this level of organization cells w ere no
longer capable of individually sustaining autonom ous life.
A collective system that provides and distributes oxygen
and nutrient m aterials to each cell and takes away the
products of their m etabolism becam e essential, and also
evolved. It is the cardiovascular system , a closed circuit of
vessels, that enables the life of each individual cell in the
hum an organism , as w ell as in all m am mals.
To enable it to take care of the nutritional and
im munological needs of individual cells, the blood is kept
in continuous motion from the left heart, via the aorta,
arteries, arterioles, capillaries, venules, veins, vena cava, to
the right heart, through the pulmonary artery to the lungs,
and ® nally, through the pulm onary vein, back to the left
heart (® gure 1). The total volum e of blood (4± 6 l, or 7± 8%
of the body w eight) circulates along this path in one
minute, on average in a relaxed, healthy subject in repose
[1,2]. W ith 60 beats per minute the heart of a m an outputs
5.5 litres on average in a m inute. T he process of respiration,
by which the blood exchanges gases with the atmosphere, is
Authors’ address: G roup of N onlinear D ynam ics and Synergetics, Faculty
of Electrical Engineering, U niversity of Ljubljana, Ljubljana, Slo venia.
e-m ail: m aja@ osc.fe.uni-lj.si and aneta@ osc.fe.uni-lj.si
0010-7514 /99 $12.00
also involved in the regulation of pressure and ¯ ow, and it
dominates in the venous ¯ ow . A long the vessels the ¯ ow is
also regulated by myogenic and neurogenic processes [1,2].
B oth are involved in vasom otionÐ continuous oscillatory
m ovement of the vessels. The myogenic process results
from the continuous contraction and relaxation of smooth
m uscle in the vessels’ w alls. This process aŒects the radial
com ponent of the vessel movement and is based on the
concentration diŒerence of ions inside and outside the
m uscle mem brane. The neurogenic process is controlled by
the autonom ous nervous system . H aving its origin in some
centres in the brainstem that are connected to other parts of
the central nervous system, and sensors throughout the
w hole network of vessels, it provides synchronization of the
function of the entire system. It m ainly aŒects the
longitudinal component of the vessels’ m ovem ents.
The place w here the cells of the human body have direct
access to the blood is nam ed the capillary bed (® gure 2). It
serves both transport and exchange functions. The capillary
bed is the netw ork of capillaries, feeding arteriole and
draining venule, that act collectively as a functional
m odule. T he average length of a capillary is about 200 ±
250 ¹ m, and its diam eter ranges from 8 ¹ m to 10 ¹ m
depending on the site, but for the m ost part is com parable
w ith or even sm aller than the diam eter of the red blood
cells. T he ¯ ow of blood in vessels whose diameters are
com parable with that of red blood cells is term ed
m icrocirculation [3]. W hile som e of the blood passes
directly to the venules, thus m aking circulation continuous,
som e stays in the capillaries closed by pre-capillary
Ó 1999 Taylor & Francis Ltd
 32 A. Stefanovska and M. BracÏ icÏ
Downloaded by [California State University of Fresno] at 16:12 29 November 2012
Figure 1. Blood circulates through the cardiovascular system, a
closed system of vessels. In one minute on average the whole
volume of blood passes through the heart and the lungs, and is
then distributed to the diŒerent parts of the body according to
their individual needs. Oxygenated blood is portrayed as red, de-
oxygenated as blue. Modi® ed from [2], with permission.
sphincters. The latter are rings of sm ooth m uscles that
rhythmically switch between the open and closed condition.
The exchange of matter and energy between the blood
and the tissues occurs across the capillary w all, so that the
system is thermodynam ically open. The capillary bed can
be conceived of as two concentric tubular barriersÐ an
inner tube, a layer of endothelial cells, and an outer tube,
the basement membrane. T he latter is directly continuous
with the tissue ground substance [4]. The endothelial cells
are interdigitated to form the interior ® brillar layer, som e
500Ð 600 A thick. T he tw o layers, although continuous in
their structure [5], are capable of selection am ong the
substances that await to enter the intracellular space. From
there, they becom e involved in all form s of tissue
metabolism, both physical and chem ical. T hese processes
occur on tim e scales longer than one minute, however, and
will not be considered in the analysis presented below.
In what follows, we shall summarize current under-
standing of the system responsible for the circulation of the
blood and for controlling the concentrations of dissolved
gases and nutrients within itÐ the cardiovascular systemÐ
restricting our attention to those processes that occur
within one cycle of blood through the system . The rhythmic
metabolic process that in¯ uences blood transport by
facilitating exchange across the capillary wall is thus the
slowest dynam ic process that will be considered. W e w ill
Figure 2. A typical capillary bed, where the cells of human
body have direct access to the blood. An exchange of energy and
matter occurs along the capillary walls so that the cardiovascular
system is thermodynamically open. Modi® ed from [2], with
permission.
® nd that physics has a major role to play in accounting for
the operation of this wonderful self-regulating biological
mechanism. W e w ill also see that, contrary to popular
perception, the heart of a healthy person at rest does not
beat at a constant rate. Indeed, unevenness of the heart
rhythm seem s to be absolutely essential to physical well-
being.
2. B ackground
From the earliest times, blood has been recognized as the
life-giving ¯ uid. U ntil the beginning of the seventeenth
century, however, it had been believed that the blood w as
prepared in the liver and then m oved through veins into
organs, w here it w as consum ed. From veins it came to the
right heart where it divided into two streams, one supplying
the lungs and the other, through `interseptal pores’, going
to the left heart. This was seen as a place where the blood
mixed with the air (pneum a), became heated and then
passed to the aorta. The ® rst criticism to this view, which
was form ulated by Galen of Pergamum in the second
century, w as proposed by Ibn el-Na® s not earlier than in
thirteenth-century. He proposed that the blood from the
right heart continued to the lungs, where spread in the
pulmonary substance to m ix with the air and then returned
to the left heart. The role of valves in the heart w as ® rst
described by Andrea Casalpino in 1571 , who then
introduced the term circulatio [6].
It is to W illiam Harvey that w e owe the conception and
proof of the idea that blood does indeed circulate. He w as
able to show that the valves in the heart are so arranged as
to allow the passage of blood in only one direction.
Further, by watching the m otion of the heart in living
animals he concluded that in the phase of em ptying the
ventricles, known as systole, the blood is expelled to the
 Physics of the human cardiovascular system
lungs via the pulm onary artery and to the rest of the body
via the aorta. In the phase when the atria are ® lling, known
as diastole, he observed that the blood enters the heart
through the vena cava and the pulm onary vein (® gure 1).
He calculated that if only a drachma (3.55 ´ 10
Ð 6 3
m ) of
blood w ere expelled at each beat, in half an hour the heart
would use up all the blood in the body, thus com pletely
emptying the veins and distending the arteries. A fter m ore
than two decades of system atic work, Harvey form ulated
his results in 1628 , when he concluded that the circulation
of the blood is `the sole and only end of the m otion and
contraction of the heart’ [6]. T he fact that blood moves in a
Downloaded by [California State University of Fresno] at 16:12 29 November 2012
circle, entering the veins from arteries, w as supported by
M arcello M alpighi, w ho in 1661 , using a microscope,
discovered the capillaries [7].
H arvey’ s discovery of blood circulation opened the doors
to modern physiology, but at least tw o centuries were to
elapse before science had developed su ciently to pass
through them . Even today, the fact that the lungs and the
heart are the only organs through w hich the entire am ount
of blood passes on average in each cycle is often
overlooked. T heir interplay in m aintaining the ¯ ow and
pressure levels still needs to be clari® ed. The blood ¯ ow
through those organs is usually treated separately, mainly
analysing ¯ ow in the heart and / or veins (see [8] and the
references therein), and gas exchange in the lungs.
In the follow ing subsections we will trace the develop-
ment of our understanding of physics of blood circulation,
grounded on Harvey’ s observations. First, w e will consider
the mechanical function of the heart and the ¯ ow of blood
through vessels. Then, secondly, w e will discuss the
regulatory mechanisms that are involved in maintaining
rhythm ic ¯ ow of blood through the cardiovascular system.
2.1. M echanics of the blood ¯ ow
The study of mechanics of the blood ¯ ow is marked by the
work of Jean Poiseuille. A fter completing his doctoral
research on The force of the aortic heart in 182 8 [9], he
turned his attention to circulation through small vessels. To
be able to control all param eters involved, he built a model
device w here he studied the liquid ¯ ow in sm all diameter
glass capillaries. Poiseuille set out to ® nd a functional
relationship am ong four variables: the volumetric e‚ ux
rate of distilled water from a tube Q , the driving pressure
diŒerential P, the tube length L, and the tube diam eter D.
The diameter of his glass tubes ranged from 15 ¹ m to
600 ¹ m, how ever larger than the size of human capillaries.
He varied the other param eters, too, and by careful
measurem ents established the relation am ong the above
parameters as
K ¢ ¢ PD4
Q 5 , ( 1)
L
33
w here K ¢ ¢ is a function of tem perature and the type of
liquid ¯ ow ing. Later, Eduard Hagenbach solved the
problem of Poiseuille ¯ ow by application of the N avier±
Stokes equations. He show ed that K ¢ ¢ 5 p /128¹ , where ¹ is
the viscosity of the ¯ uid. A lthough viscosity was de® ned by
N avier in 1823 , Poiseuille him self did not use this term, so
that the present-day form of the Poiseuille’ s relation w as
com pleted by Hagenbach who, in 1860 , proposed calling it
Poiseuille’ s law. The value of ¹ derived from K ¢ ¢ , which
Poiseuille obtained, agrees with currently accepted values
to w ithin 0.1% , which illustrates the remarkable precision
of his experiments. H is quest for the utm ost possible
precision w as m otivated in part by the fact that accepted
opinion, including that of the authorities of his tim e,
T hom as Young and Claude Navier, held that Q is
approximately proportional to D 3.
At about the sam e tim e a G erm an hydraulic engineer,
G otthilf Hagen, published in 1839 a paper on the ¯ ow of
w ater in cylindrical tubes. A lthough it is general opinion
that the careful and precise experim ents of Poiseuille were
fully convincing, the law governing ¯ uid ¯ ow through a
tube is also nam ed the H agen± Poiseuille law.
It w as George Stokes himself w ho, in 1845 , solved the
problem of Poiseuille ¯ ow as an application of the N avier±
Stokes equations, but he did not publish his results because
he believed that they con¯ icted with experiment: he w as
evidently unaware of Poiseuille’ s w ork.
The Navier± Stokes equations are used universally today
to describe ¯ uid ¯ ow , including velocity pro® les in large
arteries. T he equations are derived from the basic principles
of conservation of m ass and m omentum. The conservation
of mass is expressed by the continuity equation
­ q
­ t
1 Ñ . [q ]5 0, ( 2)
w here q 5 q ( , t) is the density of the ¯ uid, t is the tim e,
5 ( , t) the velocity vector, and Ñ is the gradient vector
operator. F or a scalar function q , Ñ . q = grad q , and for
the vector , Ñ . = div .
Conservation of momentum leads to the equation of
m otion
­
q 1 [ .Ñ ] 5 Ñ .r 1 q , ( 3)
­ t
w here r is a stress tensor and 5 ( , t) denotes the body
force per unit m ass. F or a ¯ uid, the stress tensor is
r 5 2 p 1 with 5 unit tensor . ( 4)
H ere, 2 p represents an isotropic part, having the sam e
form as the stress tensor for a ¯ uid at rest with a
hydrostatic pressure p, and tensor represents the non-
isotropic part, caused by the ¯ uid motion. However, to
deduce the dependence of on the local velocity gradients,
it is assum ed that d ij is a linear function of the various
 34
components of the velocity gradients and that the ¯ uid is
isotropic, called a New tonian ¯ uid. Under the above
assum ptions is expressed as
1 1
5 2¹ 2 [Ñ . ] with e ij 5
3 2
where is the sym metrical part of the velocity gradient
tensor, know n as the rate of strain tensor, and ¹ is the ¯ uid
viscosity depending on the temperature. B y substituting (5)
into (3), the expression for the velocity is obtained
­ 1
q 1 .Ñ 5 q 2 Ñ p1 Ñ . 2¹
Downloaded by [California State University of Fresno] at 16:12 29 November 2012
­ t
Equations (2) and (6) are known as the N avier± Stokes
equations for ¯ uid motion. For the case w here tem perature
diŒerences are small enough for the temperature to be
taken as uniform over the ¯ uid, and for a ¯ uid that is
incompressible, they reduce to
Ñ . 5 0,
­ the vag us nerve can inhibit heart action, which had already
q 1 .Ñ 5 q 2 Ñ p1
­ t
A complete analytic solution of the Navier± Stokes equa-
tions is in general extrem ely di cult. Although for m ost
regim es of interest numerical solutions may be obtained,
® nding them is still a matter of intensive research for m any
types of ¯ uid. Readers interested in a complete derivation
and num erical solution for velocity pro® les in large arteries
are referred to [10] and the references therein. Based on the
Navier± Stokes equations, both laminar and turbulent types
of ¯ ow are studied. L am inar arises in linear regimes, and
turbulent ¯ ow results from nonlinear characteristics of the
system. Linear ¯ ow usually has a Poiseuille velocity pro® le,
while a variety of velocity pro® les can be obtained in the
case of turbulence. T urbulent ¯ ow is observed mainly in
large arteries, where pressure gradients and velocities are
high, whereas laminar ¯ ow is characteristic for most of the
small vessels [10,11].
The boundary conditions for the N avier± Stokes equa-
tions are complicated, and the initial conditions are di cult
to de® ne. A simpli® ed description of the cardiovascular
system as a w hole m ight serve to obtain the values of initial
conditions. The equations (7) and (8) are obtained on the
assum ptions that the ¯ uid is incompressible, Newtonian
and isothermalÐ which might seem reasonable approxima-
tions, but the vessels themselves undergo dynam ic changes
in geometry which are not taken into account.
The m echanical approach is based on the assum ption
that the system is conservative and that it can be
characterized by the equations of mass, momentum and
energy conservation. The physics of the system is studied
A. Stefanovska and M. BracÏ icÏ
locally, with separate parts of the system being considered
in isolation. T he whole approach needs a very detailed prior
understanding of the system in order to be able to provide a
good description of the observed behaviour. But the more
­ vi ­ vj
1 , ( 5) precisely w e try to understand the m echanics of the system,
­ xj ­ xi
the more detail we need, and the less we are able to consider
its global characteristics and to understand the m echanisms
involved in one cycle of the blood through the system.
2.2. Regulation of the heart function
2 [Ñ . ] . 2.2.1. Autorhythm icity of the heart. A s already indicated,
3 the heart function consists of two phases: systole, the period
( 6) of the cycle when the heart is emptying; and diastole, the
period when it is ® lling. In order to function e ciently, the
cardiac pum ping action m ust proceed in a coordinated
fashion. The coordination is maintained by excitatory
signals generated w ithin the heart itself. It was noticed as
early as 177 7 that arteries contracted and dilated in phase
with heart action, but it w as not until 183 1 that E rnst W eber
( 7)
show ed that these were controlled by the same type of nerves
[6]. In 184 5 A lfred and E rnst W eber obtained evidence that
¹ Ñ 2
. ( 8)
been suggested by A lfred Volkmann in 1837 . Shortly
afterwards, the other type of nerves that innervate the
heartÐ the sympathetic nervesÐ were described by C laud
Bernard. W hen they are severed, the m uscles of the heart and
vessels become less stiŒ, causing the vessels to dilate.
External stimuli of sympathetic nerves provoke m uscle
stiŒening, or contraction. Those tw o states are known in
physiology as vasodilatation and vasoconstriction, while the
continuous constriction and dilation, by which the blood is
pushed forward, is known as vasom otion.
Today it is know n that the cardiac centres in the pons
and in the m edulla oblongata , w hich are both part of the
brainstem ² , exert a direct in¯ uence on the activity of the
heart, by w ay of sympathetic and parasym pathetic nerves.
The action of the heart is therefore primarily controlled by
the autonomous nervous system .
The rhythmic pulsation of the heart is how ever main-
tained by excitatory signals ³ generated within the heart
Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð
² Th e brainstem consists of the m edulla oblongata, pons and m id brain.
The m edulla oblo ngata is an enlarged continuation of the spin al cord
extending up in to the pons. A ll nerve fibres linking the brain to the spinal
cord, ascending and descendin g, as w ell as nerve fibres linking cerebrum
to cerebellum pass through the brainstem, thus m akin g it a crossroads of
the nerve pathw ays. C entres of the autonom ic nervous system that provide
basic control m echanism s for blood pressure, heart and respiratory
function are also located in the brainstem .
³ N erve excitatio n is the state of autom atic, progressive breakdown of its
m em bran e charg e, producin g a propaga ting potentia l, the a ction
potential, along the nerve. C ells in w hich an action potential can be
elicited are called excitable. Excitability is a typical property of nerve and
m uscle cells.
 Physics of the human cardiovascular system
itself. Under suitable conditions a heart rem oved from the
body will continue to beat at a constant frequency [2]. This
occurs through the action of specialized cells of the
pacemaker and conducting system (® gure 3). T he sinoatrial
node is the primary pacemaker of the heart, having the
highest discharge rate, i.e. the frequency of generating an
action potential, based on ion concentration diŒerences
across the membrane. The mechanism by which the action
potential is generated and propagated through the tissue is
however beyond the scope of this article. Readers interested
in it are referred to [2,7].
The parasym pathetic nerves branch oΠfrom the vag us
Downloaded by [California State University of Fresno] at 16:12 29 November 2012
nerves on both sides of the cervical region in the spinal
cord. The right branch of the vag us nerve goes to the right
atrium, w here it is concentrated at the sinoatrial node (SA),
while the left branch goes to the atrioventricular node
(® gure 3). A ccordingly, stimulation of the right vag us
predom inantly eŒects the heart rate. Externally applied
electrical stim uli to the cardiac vag i does slow the heart. It
may reduce the cardiac output, or even stop the heart, but
these eŒects are due to vagi slowing or stopping the
generation of the stimuli by the SA node. W hen the body is
at rest, the SA node drives the heart at a rate of about 60
im pulses/ min. The left vag us eŒects the velocity of
transmission of electrical im pulses from atria to ventricles
and thus the time between the atrial and ventricular
contraction. In this way the left vag us also in¯ uences the
heart rate. There is only sparse parasym pathetic innerva-
tion of ventricles. Its role there is indirect and serves to
inhibit the sym pathetic action.
The sympathetic nerves come from the upper thoracic
segments of the spinal cord. They are uniform ly distributed
Figure 3. The pacemaker and conducting system of the heart as
seen in the frontal section. Modi® ed from [2], with permission.
35
to all parts of the heart. T heir activity increases the heart
rate by increasing the rhythmicity of the SA node and,
m ore importantly, they increase the strength and speed of
contraction of muscles in both atria and ventricles. In
biological em ergencies, such as ¯ ight, fright or ® ght, an
increase in sym pathetic activity is of vital im portance to the
organism in securing maxim al mobilization of the pum ping
m echanism s of the heart [1].
2.2.2. Electrical and mechanical action of the heart. After
the electrical nature of the excitation of smooth m uscles in
the heart and the vessels was appreciated, attempts were
initiated to record the corresponding electrical signals.
U sing a galva nometer, Carlo M atteucci show ed in 183 8
that the heart muscle generated a m easurable electric
charge. In 190 3 W illiam Einthoven m odi® ed the string
galva nom eter to record continuously the electrical activity
of the heart, creating what was eŒectively an early
electrocardiograph (E CG). Both the characteristics of the
electrical activity in the heart, and their mechanisms, have
been intensively studied since then and standards have been
established to m ake the ECG universally useful [2].
The electrical potentials can be detected by electrodes
placed at various points of the body. The basic bipolar limb
leads, nam ed I, II and III, as proposed by Einthoven, are
based on the triangle formed by the shoulders and the
crotch. The electrical signal is obtained as a diŒerence of
potentials sensed by plus and m inus electrode with respect
to the third electrode, giving a ground potential. A typical
signal obtained by electrodes on both shoulders and one
below the heart, from one of the so-called precardial leads
[2], is presented in ® gure 4. Each portion of the ECG
represents electrical activity in a particular part of the
heart, and the speci® c peaks w ere denoted as P-Q-R -S-T,
the R -peak being the maxim al value in most of the
electrode con® gurations.
The standard analysis considers the time interval
between the characteristic peaks, obtained after averaging
over some num ber of beats. By convention, P and T are
nam ed w aves, and the distance betw een two waves is called
a segment, w hile an interval com prises both waves and
segments. F rom the EC G curve an atrial and a ventricular
part can be distinguished, as w ell as the states when valves
are open or closed. D uring the P wave the excitation
spreads over both atria, and w ithin the PQ segment the
atria as a whole continue to be excited. The QRS complex
corresponds to the excitation of both ventricles, and the T
w ave re¯ ects recovery from the excitation of the ventricles.
The states of mitral valve, the valve between left
atrium and ventricle, and the aortic valve, the valve
between left ventricle and aorta, are presented in ® gure
4. The mitral valve closes at the R -peak and at the end
of QR S com plex the aortic valve opens. The blood is
then ejected into the aorta. T his phase is nam ed systole.
 36 A. Stefanovska and M. BracÏ icÏ

in the aorta and ~ 1.2 kPa in the pulmonary artery, both

arterial valves open and blood begins to be expelled. The
intraventricular pressure continues to rise, until it reaches
its m aximum of 16 kPa in the left and 2.7 kPa in the right
ventricle. Towards the end of systole it falls again. The ¯ ow
of blood through the system is driven by the pressure
diŒerence. Its value depends on the side of the system,
being higher in the arterial side, and varies with respect to
the size of the vessels. On the capillary side it is m odulated
by both the central and the peripheral m echanism s. In
® gure 4 one cycle of the electrical activity of the heart
together with the arterial pressure and the blood ¯ ow
Downloaded by [California State University of Fresno] at 16:12 29 November 2012

through a capillary bed in the skin of a human hand is

presented. The electrical activity results in a pressure wave
that drives the ¯ ow of blood through the system .

2.2.3. Heart rate variability. Thus far, w e have discussed

the m echanism of a single heart cycle. The tim e between
two successive R-peaks, i.e. the period of a complete heart
cycle, is however not constant, but rather varies in time. In
their pioneering w ork H yndm an et al. [12], Sayers [13] and
Chess et al. [14], in the 1970s , focused attention on the
existence of physiological rhythm s im bedded in the beat-to-
beat heart rate signal.
By plotting the inverse value of the time between two
successive R -peaks, the instantaneous heart rate (IHR ), as
a function of time, a new time series is generated (® gure 5).
Figure 4. A typical signal from the electric activity of the heart
together with the corresponding states of mitral and aortic valves
(top), the blood pressure (middle) and the peripheral blood ¯ ow
(bottom) during one pumping cycle of the heart.

At the end of the T wave the aortic valve closes and

shortly after the m itral valve opens to allow the passage
of blood from ventricles to atria. This is the phase of
® lling or diastole. Sim ilar phases occur also in the right
heart w ith the ejection period being initiated slightly
earlier and the systole slightly later. T hese tim e
diŒerences are relatively small (of the order of 10±
30 ms) and have no particularly eŒect on the dynam ics
of blood ¯ ow. The corresponding valves in the right
heart are named tricuspid and pulm onary valves. T he
tricuspid and m itral valves are also nam ed atrioventri-
cular (AV) valves, w hile the aortic and pulm onary
valves are nam ed arterial valves.
The excitatory events described above govern the
mechanical activity of the heart by causing the contraction
of muscle cells in the heart, known as myocardial cells.
Opening and closing of the valves is brought about by
pressure changes in the adjacent heart cavities or vessels.
The motion of the valves in turn aŒects the mode of
contraction of the myocardium . W hen the intraventricular Figure 5. Steps in derivation of the heart rate variability (HRV)
pressure exceeds the arterial diastolic pressure of ~ 10 kPa signal.
 Physics of the human cardiovascular system
However, the signal that we derive is discrete and not
continuous and has a variable sam pling tim e (® gure 5).
After interpolation the signal nam ed heart rate variability
(H R V) is obtained. Num erous methods of interpolation
have been proposed, but they all in¯ uence spectral
components in various ways. The analyses that we present
below are obtained by linear interpolation between two
IHR values (® gure 5). The signal is than re-sampled w ith a
constant sam pling tim e of 0.1 s.
A mong all physiological signals, HRV is the one least
in¯ uenced by movement artefacts, or by instrum ental noise.
It has also an additional advantage. W hile the measure-
Downloaded by [California State University of Fresno] at 16:12 29 November 2012
ments of most physiological signals are inevitably in¯ u-
enced by interference from other physiological processes,
the value of the R-peak corresponds to the pumping cycle
of the heart and determines precisely a distinct m oment of
the heart beat.
A kselrod et al. [15] were the ® rst to introduce spectral
analysis of the H RV signal to evaluate quantitatively the
beat-to-beat cardiovascular control. Since then, HR V has
been analysed extensively using the F ourier transform of
the signal itself, or its autoregressive model, i.e. non-
parametric and parametric m ethods (see [16] and the
references therein). T he linear frequency analysis is based
on the assum ption that it is possible to decom pose a time
series into the ® nite num ber of periodic sinusoidal functions
with diŒerent frequencies and phases. The F ourier spec-
trum of the EC G, recorded for 20 m in on a healthy, resting
subject is presented in ® gure 6a.
W hen the shape of the original time-series is non-
sinusoidal, higher harmonic com ponents of the fundam en-
tal frequency are necessary to reconstruct the function.
Thus the spectrum of an EC G consists of a fundamental
frequency of ~ 1 H z (60 beats per minute), and components
at higher integer multiples of this basic frequency, i.e. at
Figure 6. Fourier (top) and wavelet (bottom) transforms of (a) the ECG and (b) HRV signals. The Fourier spectra are obtained as an
average of spectra calculated for 200 s time segments, shifted along the signal for 100 s. The wavelet transform is also averaged in time,
obtained with (a) f0 = 3 and (b) f 0 = 1.
37
2 Hz, 3 Hz, etc, as can be seen in ® gure 6a. As already
m entioned, the heart rate in a healthy subject varies, which
is w hy its basic frequency is not sharp. T he nature of its
variability can be studied by analysing the H RV signal. Its
F ourier spectrum is presented in ® gure 6 b. B ecause of the
discrete nature of the events the basic sam pling rate of the
H RV is inherently the heart rate itself, and consequently
this provides maximum frequency in the H RV spectrum.
T he highest frequency in the spectrum, f max is determined
by the sam pling frequency, f s 5 1 /t s , and the relation
f max 5 f s /2. For exam ple, if the heart rate is ~ 1 Hz, then
f max in the H RV spectrum is ~ 0.5 Hz.
The peaks in the HRV signal correspond to the periodic
processes that m odify the basic heart frequency. It has
already been recognized that the information contained in
the H RV signal is of great clinical and physiological
importance. D iŒerent diseases were shown to m anifest
themselves in a speci® c w ay in the HRV spectrum.
R ecently, standards of measurement, physiological inter-
pretation and clinical use of the HR V w ere proposed [17]. It
has been suggested that a sampling rate betw een 250 and
500 Hz is optim al to record the shape of the ECG curve and
de® ne the R-peak. The importance of the time of
observation is also pointed out. The recording is recom -
m ended to last at least 10 tim es the period of the lower-
frequency bound of the investigated component, but should
not be substantially extended beyond this in order to ensure
the stationarity of the signal. Two diŒerent types of
m easurem ents were suggested: short-term (5 min) and
long-term (24 h) recordings. Three main frequency domains
are distinguished in the short-tim e recordings: (i) high
frequency (H F) range, 0.15± 0.4 H z, (ii) low frequency (LF )
range, 0.04± 0.15 H z, and (iii) very low frequency (VLF )
range, £ 0.04 Hz. From the long-term recordings the VLF
range is de® ned from 0.003 to 0.04 H z, and for the
 38 A. Stefanovska and M. BracÏ icÏ
frequencies £ 0.003 Hz an ultra low range (U LF ) was
de® ned. These frequency domains in the H RV spectrum,
calculated using F ourier transform , are presented in the top
section of ® gure 6b.
The physiological origin of the periodic processes
involved in the modulation of the heart rate is not well
understood. It has been proposed that vaga l activity is the
major contributor to the HF component, w hile disagree-
ment exists in respect to the LF component. It is considered
either as a marker of sym pathetic modulation or, alter-
natively, that it includes both sym pathetic and vagal
in¯ uence. T he physiological interpretation of the lower-
Downloaded by [California State University of Fresno] at 16:12 29 November 2012
frequency components warrants further elucidation (see
[17] and the references therein). This interpretation is
however based on considering only the electrical activity of
the heart m uscles and is focused on understanding the
nerve function that is involved in the observed ¯ uctuations
of the heart rate.
As early as 1733, Hales observed that changes in blood
pressure and heart rate were related in a regular m anner to
the respiratory pattern in the horse [18]. Ludwig’ s invention
of the kym ograph allow ed his observation in 184 7 that the
dog heart rate increased on inspiration and decreased on
expiration [19]. This phenomenon is known as `respiratory
sinus arrhythmia’ and is interpreted as a dynam ic response
of the heart achieved by central nervous modulation of the
input to the sino-atrial node. Here, w e w ould like to stress
the im portance of distinguishing the quantities that are
regulated, w hich are blood pressure and ¯ ow, as well as a
vessel’ s resistance and conductance, and the way that
information about their values is transmitted. It is the
nervous system that conveys the inform ation about these
quantities, derived from mechanical and chem ical sensors at
various points w ithin the cardiovascular system . In general,
the function of all nerve cells in the body is to receive
information, to carry it to other parts of the system , to
compare it to other information, and ® nally to control the
function of other cells. That is w hy they operate on smaller
tim e scales, than for example the scale on which the blood is
distributed through the body [20].
In attempting to understand the function of physio-
logical systems, one of the greatest di culties is to
recognize the distinction betw een cause and eŒect. T he
functioning of the cardiovascular system results from an
interplay of a variety of m echanism s that serve to keep
the values of pressure and ¯ ow within certain limits.
Therefore, in the pages that follow we will present our
understanding of the physics of the cardiovascular
system on the assumption that the quantities character-
izing its function are m echanical, w hile the electrical
activity of the nerves serves for com munication within
the system. Because an understanding of the physiolo-
gical origin of the periodic events that contribute to
heart rate variability is of central importance for a
physical as well as a mathematical representation of the
system, we will turn once again to this question.
3. Frequency analysis of cardiovascular signals
The heart rate variability signal allows us to observe the
function of the heart in time. By extracting the inform ation
that it contains, the basic principles of heart function may
be studied under normal conditions as well as in perturbed
states. However, we cannot apply classical perturbation
theory directly, but rather must observe its function after
reversible perturbations, such as physical exercises and
application of vasodilator substances, or after irreversible
perturbations resulting from diseases.
The H RV signal allow s for observations of the function
of the cardiovascular system from a central viewpoint (at
the heart itself). L ater in this section w e shall introduce
another signal derived from ¯ ow through the vascular bed,
namely the peripheral blood ¯ ow. This signal provides a
peripheral viewÐ from one of the places w here blood
exchanges substances with the cellsÐ of the dynam ics of the
cardiovascular system.
The tim e of observation is of crucial importance in
determ ining the type and am ount of inform ation that can
be extracted from a signal. As we have already stated, our
goal is to understand the dynam ics of one cycle of the
blood through the cardiovascular system. A s it takes
approxim ately one m inute, and its dynamics is not strictly
periodic, we have chosen the observation tim e to be 20 m in.
3.1. M ethods of analysis
The existence of rhythmic activity in the HR V signal has
already been pointed out, so we shall start with an analysis
of the cardiovascular system in the frequency domain. Let
us ® rst discuss the F ourier transform in more detail.
3.1.1. Fourier transform. A physical signal (with ® nite
energy) m ay be presented in either time or frequency
dom ains. The Fourier transform
¥
G (f ) 5 g(t) exp ( 2 i 2p f t) d t ( 9)
2 ¥
and its inverse
1 ¥
g(t) 5 G (f ) exp ( i 2p f t) d f , ( 10)
2p 2 ¥
are the m athem atical tools which connect the two dom ains
[21]. T he representation in the frequency dom ain G (f )
consists of an am plitude and corresponding phase for each
frequency f . The power spectrum of the function
2 2
P ( f ) 5 | G (f ) | 1 | G ( 2 f ) | gives the power (energy density
in the frequency domain) in a given frequency interval
between f and f 1 d f .

W hen a tim e series of ® nite length T 5 N t s sam pled at
discrete points n t s is considered, the calculation of the
Fourier transform reduces to a ® nite sum over all m easured
values. The resultant discrete Fourier transform (D F T)
N2 1
G ( f k) 5 g( j t s ) exp ( 2 i 2p j k /N )
j5 0
is O . . . de® ned only for discrete frequencies f k 5 k /T ,
k 5 N 2 1. The ® nite length of the signal bounds the
frequency resolution ( D f 5 1 /T ) and the low est detected
frequency w hilst ® nite sam pling time determ ines the upper
frequency lim it (f max 5 2 /t s ). Since the sum is taken over
Downloaded by [California State University of Fresno] at 16:12 29 November 2012
® nite time, the signal is assumed to be periodic with period
T . A frequency between any f k and f k1 1 will contribute to
all G ( f k ), known as leakage [22]. T o reduce leakage, the
original signal is windowed. The choice of the transform
window is a com prom ise between making the central peaks
as narrow as possible versus m aking the tails fall rapidly.
Thus it is the w indow length and shape that then
determ ines the frequency resolution.
Spectral properties of m easured signals are com monly
estimated by the periodogram
1 u2
P (f k ) 5 | G (f k ) | 2 , k5 0, k 5 N /2 ,
N2
1 2
P (f k ) 5 ( | G (f k ) | 1 | G (f N 2 k ) | 2) ,
N2
The normalizing factor in the periodogram estim ate is
chosen in such a m anner that the sum of all P (f k ) equals
the average square value of the original tim e series.
Fourier transform s of the ECG and HR V signals
presented in ® gure 6, are obtained as average period-
ograms, calculated from 160 s tim e segm ents, taken at a
regular intervals of 80 s.
3.1.2. W avelet transform. T he representation of time
series in the frequency domain bears no inform ation about
the tim e. Nam ely, the F ourier transform (9) and its inverse
(10) reproduce the time series as a superposition of periodic
functions. These have sharp peaks in the frequency domain,
but are spread over all tim e. If a characteristic frequency
varies with tim e, the corresponding peak will be broader than
its instrum ental width as determined by the data window .
The time-varying nature of characteristic frequencies in
the cardiovascular signals dem ands an analysis in the
tim e-frequency domain. Yet, the relatively broad fre-
quency band within which characteristic peaks are
expected raises a problem in relation to tim e and
frequency resolution. In the time-frequency analysis, a
window of ® xed length is shifted along the signal to
achieve tim e localization and the frequency content of
each window is evaluated. The window length introduces
a scale into the analysis and determines the time and
Physics of the human cardiovascular system 39
frequency resolution. B y the uncertainty principle, sharp
localization in time and frequency are mutually exclusive.
T herefore, the choice of w indow length is in practice a
trade-oΠbetw een time and frequency resolution. If both
low and high frequencies w ith diŒerent tim e spans are to
( 11) be detected simultaneously in a signal, a suitable choice is
very di cult. T his is the problem M orlet was facing
w hile analysing seism ic data, w hich comprised diŒerent
features in tim e and frequency. T o overcome the short-
com ings of the F ourier method he cam e up w ith the
basic idea of w avelet analysis in 198 3 [23]. Later,
G rossm an and M orlet laid the m athematical foundations
of the wavelet transform technique [24].
W avelet analysis is a scale-independent method. As in
the window ed Fourier transform, one begins with a
w indow function, called a mother w avelet w (u ). This
function introduces a scale (its width) into the analysis.
C ommitm ent to any particular scale is avoided by using not
only w ( u ), but all possible scaling of w ( u ). T he m other
w avelet is also translated along the signal to achieve tim e
localization. Thus, a family of generally non-orthogonal
basis functions
5 |s |2 pw
t
. ( 13)
s ,t
s
k5 1, . .. , N /2 2 1 . is obtained. The parameter p is an arbitrary non-negative
number. The prevailing choice in the literature is p 5 1 /2
( 12) [25]. In this case, the norm of the wavelet || w || and thus its
energy is unaŒected by the scaling operator.
The continuous wavelet transform of a signal g( u ) is
de® ned as
¥ ±
g~(s , t) 5 s ,t (u )g(u ) du . ( 14)
2 ¥
T he wavelet transform g~(s , t) is a mapping of the function
g(u ) onto the time scale plane. The interpretation of g~(s , t)
depends on the m other wavelet being used.
Not every function can be used as the m other wavelet.
O nly those that enable us to reconstruct the original function
g(u ) from its wavelet transform g~(s , t) are adm issible.
T he su cient condition for the reconstruction [25] is
¥
w (u ) d u 5 0. ( 15)
2 ¥
T he total energy of the signal g(u ) can be calculated as
|| g|| 2 5 C2 1 | s | 2p2 3
| g~( s , t) | 2 d s dt , ( 16)
2
.
w here the constant C is determ ined by the shape of the
m other wavelet. The function
2p2 3 ~
q 5 C 2 1| s | | g(s , t) | 2 ( 17)
can therefore be interpreted as the energy density of the
signal in the time scale plane [26]. It is often called a
scalogram .
 40 A. Stefanovska and M. BracÏ icÏ

In numerical applications, the scale s and tim e t are
restricted to discrete values only. T he natural discretization
of the scaling param eter is s m 5 r m , w here m [ Z and the
step r 5 ¤ 1. W ithin the scale r m the signal is sampled only at
tim es t n 5 n r m s , s > 0, which m eans that the sampling
rate is autom atically adjusted to the scale.
For certain mother wavelets, orthogonal basis can be
obtained by using r 5 2, resulting in a special application
of the wavelet transform, known as multi-resolution
analysis (M RA ). The original signal, sam pled at t s , is
split into a `blurred’ version on a coarser tim e scale 2t s
and a detail on scale t s . T his process is then repeated,
Downloaded by [California State University of Fresno] at 16:12 29 November 2012
namely f 1 2 f 2 ³ f 1 /4p f 0. The choice of f 0 thus determ ines
the current frequency resolution. W e have taken f 0 5 1.
The wavelet transform contains information on am pli-
tude, frequency and tim e. In ® gure 7 the absolute value of
the w avelet transform of the H RV signal in the time-
frequency plane is presented. However, it is di cult to
capture all this information at once, especially since several
almost periodic phenomena are present. Their am plitudes
and frequencies also vary in time. Therefore, various two-
dimensional projections are used. W e will use scalograms
averaged over tim e to compare wavelet transform s
obtained from diŒerent signals, in diŒerent subjects and

giving a sequence of more and more blurred versions under diŒerent conditions. An averaged scalogram of the
together with details removed at every scale. Several HRV signal is presented in the low er part of ® gure 6 b.
applications of M RA to cardiovascular signals were Compared to the spectrum calculated by Fourier trans-
reported [27,28], m ainly to capture the main features of form, the low frequency resolution is im proved, allowing us
the signal. to de® ne the frequency intervals for the characteristic
In this paper, w e follow instead the original idea of processes more precisely. By providing an improved
M orlet. B y choosing r 5 1.05, and a m other wavelet well estimation of the frequency content of the H RV signal, a
concentrated in both tim e and frequency, we can detect better understanding of the physiological mechanisms of
precisely the frequency content in a given time interval. the oscillatory processes involved becom es possible. To
M orlet proposed the use of a Gaussian function, m odu- gain new insight into those processes, w e now present
lated by a sine wave. In the time dom ain, it is w ritten as analyses of the peripheral blood ¯ ow, measured simulta-
2 1/4 neously with other cardiovascular signals.
w (u ) 5 p exp ( 2 i 2p f 0u ) 2 exp ( 2 2p f 02 /2) exp ( 2 u 2 /2) .
( 18)
3.2. Cardiovascular signals
The choice of f 0 is a com prom ise between localization in
Today several m easurem ent techniques enable non-inva-
tim e and in frequency. F or sm aller f 0, the shape of the
sive, continuous, observation of a number of other
wavelet favours localization of singular tim e events, whilst
cardiovascular functions in addition to the electrical
for larger f 0 more periods of the sine wave in the window
activity of the heart. A sensor based on piezoelectric
make the frequency localization better. For f 0 > 0. 8, the
properties of some crystals can be used for pressure
value of the second term in (18) is so sm all that it can be
recordings in larger vessels. The sam e principle is used to
ignored in practice and a simpli® ed expression for the
follow the m ovem ents of the thorax in the inspiration and
M orlet wavelet in the time domain is
expiration phases of lung function. The Doppler principle
w (u ) p 2 1/4 exp ( i 2p f u ) exp ( u 2 2) .
5 2 0 ( 19) 2 /
The corresponding wavelet family consists of Gaussians,
centred at tim e t with standard deviations s . In the
frequency domain we have G aussians w ith a central
frequency f 5 f 0 /s and a standard deviation of 1 /2p s .
Therefore, the w avelet transform at a given scale s can also
be interpreted as band-pass ® ltering giving an estimation of
the contribution of the frequencies in this band. T he
relation between the scale and the central frequency for the
M orlet wavelet is
f0
f 5 . ( 20)
s
The frequency resolution changes with frequency: at low
frequencies (large scales) the resolution is better than at
high frequencies (sm all scales). A ccordingly, the tim e
resolution is better for high than it is for low frequency
components. In order for peaks to be detected at f 1 and f 2
(f 1 > f 2), they must be separated by at least one half of the Figure 7. The wavelet transform of the HRV signal in the time-
standard deviation of the peak at the higher frequency, frequency plane.
 Physics of the human cardiovascular system
allow s for measurem ents of blood ¯ ow velocity. The tissue
under observation is bathed in either ultrasound or
coherent light. Ultrasound penetrates deeply, how ever, so
that it can not be directed selectively. T hat is w hy, when
ultrasound based instruments are used clinically to measure
blood ¯ ow velocity, they are restricted to large vessels.
For peripheral blood ¯ ow measurem ents optical sensors
with wide dynam ic range are used. T he laser light can be
directed into a very sm all area. Its depth of penetration is
smaller than that of the ultrasound and it can be controlled
by selecting the wavelength and emitted power. Thus, it can
detect blood ¯ ow in the capillary bed. A s already
Downloaded by [California State University of Fresno] at 16:12 29 November 2012
mentioned, this is w here matter and energy are exchanged
between the blood and tissue cells, so that w e m ay expect all
processes involved in blood ¯ ow regulation to be re¯ ected
in the corresponding signal.
3.2.1. Peripheral blood ¯ ow. O nly four years after the
® rst w orking laser was dem onstrated by M aiman in 1960
[29], Cum mins et al. [30] proposed a m ethod of m easuring
the velocity of particles in solution by interpretation of the
Doppler-frequency-shifted light. Some years later, Riva et
al. [31] applied this technique to the measurem ent of red
blood cell velocities in the glass tube ¯ ow model. H ow ever,
it was Stern [32] who ® rst used the laser D oppler technique
for blood perfusion measurement in the undisturbed
microcirculation in 1975 . Nilsson et al. [33,34] subsequently
provided detailed technical and experim ental evaluations of
the technique.
The near-infrared laser is frequently used to measure the
velocity and concentration of red blood cells w ithin a
hemisphere of volume ~ 2 mm 3. A ® bre-optic probe carries
a beam of laser light which is then w idely scattered and
partly absorbed by the tissue. Light scattered from moving
blood cells undergoes a Doppler shift in the wavelength
while the wavelength of light scattered from static objects
remains unchanged (® gure 8). T he magnitude and fre-
Figure 8. Sketch showing how a beam of laser light is scattered
and absorbed in the tissue. From [35], with permission.
41
quency distribution of the w avelength changes are related
to the num ber and velocity of blood cells. The back-
scattered light is collected by a ® bre and converted into an
electrical signal. T his signal is proportional to the ¯ ow but
unrelated to the direction of blood cell movem ent.
This technique of blood ¯ ow measurem ent is limited by
the fact that in the case of occlusion, i.e. stopping of the
¯ ow through the measured area, there is a residual value
called the `biological zero’ , as illustrated in ® gure 9. Even
during occlusion some blood remains in the area under
observation. It is the random B rownian movement of the
remaining red blood cells that results in this residual value.
A s a result, the ¯ ow cannot be expressed straightforwardly
in absolute units (e.g m l / s / mm 3), but only in arbitrary units
(AU). To obtain an absolute m easure, the value of the
biological zero has to be determined for every measure-
m ent.
W e show in ® gure 9 that during occlusion the oscillations
vanish com pletely. M oreover, the detection of oscillatory
changes in the ¯ ow is not in¯ uenced by the lack of absolute
units, as long as we provide a calibration with a reference
value. The problem of biological zero is therefore irrelevant
to our quest for information about the oscillatory nature of
the peripheral blood ¯ ow. From ® gure 9 w e can also see
that the am plitude of oscillations is com parable in
m agnitude to that of the steady ¯ ow on w hich they are
superim posed, im mediately dem onstrating the importance
of the oscillations in characterizing the dynam ics of blood
¯ ow.
Figure 9. The peripheral blood ¯ ow (top) and its wavelet
transforms (below), before, during and after occlusion of the
vessels proximal to the measurement site. An apparent residual
¯ ow, named the `biological zero’ remains during occlusion,
whereas the oscillations vanish. The amplitude of oscillations,
before and after occlusion, is comparable in magnitude to that of
the steady ¯ ow on which they are superimposed, demonstrating
the importance of the oscillations in characterizing the dynamics
of blood ¯ ow.
 42 A. Stefanovska and M. BracÏ icÏ
Even in the early analyses of laser D oppler blood ¯ ow
recordings [34,36], the oscillatory nature of the ¯ ow was
noted. HoŒman et al., using a frequency histogram,
reported oscillations, synchronized w ith the heart rate
and also oscillations in the H F and LF range [37]. Both
windowed Fourier and wavelet analyses of signals recorded
for 20 min revealed ® ve characteristic frequencies in the
interval from 0.0095 H z to 2 H z [38,39]. The characteristic
peaks typically appear around 1 H z, 0.3 Hz, 0.1 H z,
0.04 H z and 0.01 H z, almost at the same frequencies as
those discussed above: the ® rst one in the ECG spectrum
and the latter four in the spectrum of the HR V signal.
Downloaded by [California State University of Fresno] at 16:12 29 November 2012
The energy of each particular oscillation varies with the
vessel’ s diameter and network density, i.e. the local ¯ ow
resistance. There are two possibilities for m aking the signals
comparable in terms of energy. One is a quantitative
assessm ent of the resistance in the adjacent netw ork, w here
the ¯ ow is m easured. A t present, a reliable, non-invasive
technique is unfortunately not available. The other
possibility is to choose m easurement sites where, based
on anatom ical evidences, similar vessel resistance is
expected. W hen the blood ¯ ow is m easured on the sites
with similar resistance, the contribution of each oscillatory
process does not depend on the m easurem ent site, or tim e
of the measurem ents [39]. In ® gure 10 the energy within
Figure 10. (a) Time and (b) space invariance of the oscillatory
components of the peripheral blood ¯ ow, measured in the areas
where the vessels have similar density and resistance. For details,
see text.
intervals around each of the ® ve peaks (labelled I± V) is
presented for two sets of measurem ents: (a) sim ultaneous
measurem ents on tw o diŒerent sites and (b) tw o consecu-
tive measurem ents on the sam e site. W e have show n that no
statistically signi® cant diŒerence exists and therefore that
the energy of each interval can be taken as tim e and space
invariant. W e will address the question of interval bounds
and statistical presentation later in section 3.2.4.
3.2.2. Sim ultaneo us measurem ents. In searching for the
physiological origin of the oscillations observed in the
ECG , HR V and peripheral blood ¯ ow we m ade simulta-
neous recordings of several cardiovascular functions
[38,40,41]. Signals of cardiovascular origin were m easured
on healthy young m ale subjects. During the m easurem ent
the subjects w ere lying still on a bed and they were asked to
relax. The E CG and blood pressure were sampled at
400 Hz, while a sampling rate of 40 H z w as used for
respiration and blood ¯ ow signals. The recordings lasted
20 m in. Before evaluating the frequency content the trend
was rem oved from all signals and all but ECG were
resampled to 10 H z. The left part of ® gure 11 presents a
25 s segment of the respiration, EC G, HR V, blood pressure
and peripheral blood ¯ ow on the right arm and the right leg
after pre-processing. The blood pressure was recorded on
the index ® nger of the left hand, w hile the sensors for blood
¯ ow measurem ents were placed over the bony prom inences,
of the wrist and ankle joint.
The peaks of the average w avelet transforms appear at
sim ilar, in som e cases even at exactly the sam e, frequencies
in all measured signals (® gure 11, right). DiŒerences exist,
however, in the am plitudes of the oscillations. For the ECG
and blood pressure signals the am plitude of the heart beat
frequency dominates the spectrum . T he spectrum of a
respiratory signal has one dom inant peak, at ~ 0.2 Hz. In
the HRV signal this peak is of com parable am plitude to
those of slower oscillation betw een 0.009 5 Hz and 0.15 H z.
In the signal of peripheral blood ¯ ow the am plitudes of all
oscillations are of the same range. Hence, we can see that
the peripheral blood ¯ ow re¯ ects the activities of both the
local and the central m echanism s of cardiovascular
regulation.
3.2.3. Physiological nature of oscillations. Not all of the
observed oscillations are yet understood in physiological
term s. Those that can be selectively observedÐ the heart-
beat and respirationÐ are of course relatively well under-
stood. In other cases only indirect evidence is available, and
we will see that the position is less clear.
The basic frequency in the ECG signal, around 1 H z
corresponds of course to the heart rate. At rest, its value
ranges from 0.6 H z in sportsmen to 1. 6 Hz in subjects with
im paired cardiovascular systems. The heart’ s pum ping
activity is manifested in every single vessel and is also
 Physics of the human cardiovascular system
Downloaded by [California State University of Fresno] at 16:12 29 November 2012
Figure 11. Simultaneously measured cardiovascular signals (left) and their wavelet transforms (right). HRV is in Hz, other values are
in arbitrary units.
present in the microcirculation through the capillary bed. It
is highly dom inant at the outlet of the heart, as well as in
the larger vessels. Its contribbution gradually diminishes
with decreasing vessel diameter. T he elasticity of the
vessels, and their structural properties, also aŒect the
magnitude of this ¯ ow component.
The spectral peak at around 0.2± 0.3 Hz, corresponding
to 12± 18 events per m inute, w ell known in physiology as the
breathing frequency [2]. T he existence of respiratory
modulation of the heart rate has long been recognized
[18]. This frequency, observed earlier in the HR V and
blood pressure signals, was designated [12,15,42] as the H F
interval, and attributed to the parasym pathetic autono-
mous control. In the peripheral blood ¯ ow signal the
respiratory origin of this peak was discussed by H oŒm an et
al. [37]. By simultaneous recordings of both respiratory and
blood ¯ ow signals, direct evidence was also obtained
[38,43].
In early analyses of the blood pressure and HRV signals,
oscillations with periods of ~ 10 s were associated with
blood pressure regulatory mechanisms [12,42]. Since then
several pieces of indirect evidence as to their local origin
have been reported. These oscillations are a manifestation
of the m yogenic activity of the smooth muscle cells
displaced in the walls of resistive vessels. The smooth
muscle cells respond continually to changes in the
intravascular pressure [44]. T his response is m ediated by
oscillations in the ion concentrations, mainly Ca 1 1 , across
43
the mem brane of vascular sm ooth m uscles. F or isolated
vessels, the m yogenic origin of these oscillations w as
dem onstrated either by measuring dynam ically the dia-
m eter changes of vessels [45], or ion concentrations [46],
though not in hum ans. In ® gure 11 the peak around 0.1 H z
is clearly visible in both blood ¯ ow signals, the frequency of
the peak typically being higher on the hands than on the
legs.
The peak of ~ 0.04 Hz was observed in both the HRV
and blood pressure signals [41,42] and in the peripheral
blood ¯ ow signal [38]. It is attributed either to m etabolic
[42], or to neurogenic processes. A lthough these oscillations
cannot be selectively measured, indirect evidence for their
origin was reported by K astup et al. [47]. After disconnect-
ing nerves from the vessels, known as denervation, which
suppresses the neurogenic regulation of the vessel radius,
they observed that the ~ 0.04 H z oscillations disappeared.
T his peak can be detected in the averaged scalograms of all
sim ultaneously m easured signals presented in ® gure 11. It
usually appears sm eared due to the variation of its period
w ith tim e.
Long recordings and good frequency resolution also
enabled us to isolate a peak at ~ 0.01 Hz. W e have found it
in all cardiovascular signals, although its exact position
diŒers from one to another, suggesting that it is of local
origin. T here is indirect evidence [48] that this oscillation is
related to the endothelial function. M oreover, some
experim ents suggest that nitric oxide, a m etabolic substance
 44
that is released from endothelial cells, has an in¯ uence on
the state of contraction of the vessel musculature. Recent
studies demonstrate that nitric oxide plays an essential
im munological and citotoxic role in the hum an organism.
The reader seeking physiological insight into the involve-
ment of this substance in the dynam ics of blood ¯ ow are
referred to [49] and the references therein.
3.2.4. Frequency band s and quantitative measures. T he
positions of the spectral peaks change with tim e and also
diŒer from one subject to another. F or every peak,
however, a frequency interval exists within w hich it is
Downloaded by [California State University of Fresno] at 16:12 29 November 2012
found in all subjects. The local minim a of the average
wavelet transform, were used to divide the frequency band
between 0.009 5 Hz and 2 Hz into ® ve intervals.
The intervals de® ned in ® gure 6 b diŒer from the
recomm ended standards for heart rate variability [17].
The frequency resolution of the wavelet transform enabled
us to detect several peaks in the VL F interval ranging from
0.003 to 0.04 H z, of which tw o w ere above 0.009 5 H z.
Therefore, w e have split this part of V LF interval into (I)
from 0.009 5 Hz to 0.002 Hz w here, according to our
studies [48,49], the metabolic process is m anifested and
(II) from 0.02 H z to 0.06 Hz, where the neurogenic process
is manifested [47]. The interval (III) from 0.06 Hz to
0.15 H z, where the myogenic process is most probably
manifested, corresponds to the LF interval; however, its
lower bound w as set to 0.06 Hz since a peak around
0.04 H z was detected. Interval (IV) from 0.15 Hz to 0.4 H z
on which the respiratory activity, m easured sim ultaneously,
has its dominant peak is the same as the HF interval, while
the last interval, (V) from 0.4 H z to 2 Hz is the interval of
heart frequency.
These intervals were chosen based on over 500 scalogram s
of around 100 subjects, either healthy subjects, athletes, or
subjects with some cardiovascular impairment. Neverthe-
less, a revision of the boundary values using, for example,
higher order spectral analysis is needed. The identi® cation of
high harmonics and linear combinations of basic peaks
could provide better distinction between the peaks.
The intervals were chosen in such a way that the peak
corresponding to any given physiological prosess is always
in the same interval. T hus, we can roughly ascribe the
energy w ithin one interval to the activity of that process,
and w e can thus obtain a quantitative measure of process
oscillations [39], which can be used to compare diŒerent
signals.
According to equation (17), the physical quantity behind
the scalogram is the energy density. T he average energy in a
given frequency band e i ( f i 1, f i 2) is
t 1 /f i 1
1 1 ~
e i ( f i 1, f i 2) 5 2
| g (s , t) | 2 d s d t .
t 0 1 /f i 2 s
A. Stefanovska and M. BracÏ icÏ
The energy is averaged over time and p = ô is tak en.
However, the absolute value of the energy de® ned by
equation (21) may sometim es be misleading. If the total
energy of the signal increases, it is very probable that the
energy in each band will increase. In such cases, it is in our
interest to ® nd out if and how the distribution of the energy
am ong the processes has changed. Therefore, w e shall
introduce the normalized energy in a given frequency band
e i (f i 1, f i 2)
e i ( f i 1, f i 2)
e i (f i 1, f i 2) 5 , ( 22)
e tot al
where e t otal is the energy of the signal contained in the
frequency band of our interest, i.e. betw een 0. 0095 Hz and
2 Hz.
To avoid anom alies in the average scalogram s intro-
duced by particular subjects, we will base our analyses on
hom ogeneous groups of subjects. Statistical plots will be
used to present the median values, 10% , 25% , 75% and
90% of the range. T he value that drops out of these limits is
represented as a cross. T o compare the values betw een the
groups, the M ann± W hitney test for statistical signi® cance
will be used [50]. If the probabilities of the m edian being
equivalent is below p = 0.05 the diŒerences between the
groups are set as signi® cant.
3.2.5. Reversible changes. Because the cardiovascular
system operates continuously, we can reveal many of its
functional characteristics even from a normal resting state.
Additional insight can be obtained by observing how it
reacts to perturbations. In living organisms, the choice of
ways of introducing perturbations is limited. One possibi-
lity is to induce changes by delivering pharm acological
substances, for exam ple vasodilator substances [49]. An-
other possibility is to impose changes by increased physical
activity. In the following, we will present the exercise-
induced changes, evaluated by the wavelet transform of the
peripheral blood ¯ ow signals.
Signals w ere m easured for nine young healthy male
subjects. In each case, three signals w ere recorded before,
and tw o after, 40 m in of exercise on a treadmill at a 3
s
uphill gradient [51]. Each of the signals w as recorded for
20 m in. The m aximal oxygen uptake (VO 2 m ax ) had been
evaluated a day prior to the measurem ents for individual
subjects, and the exercise intensity w as standardized to a
level of 80% VO 2 m a x . D uring the m easurem ent of peripheral
blood ¯ ow, the subjects were in a supine position in a room
in which the temperature w as maintained constant.
Sixty second segm ents of typical blood-¯ ow signals
recorded before and after exercise, and the group m edian
values of norm alized energies, obtained from the wavelet
transforms calculated from the entire signals, are presented
( 21)
in ® gure 12. It is obvious that both the steady level, and the
am plitude of oscillations, are markedly higher after
 Physics of the human cardiovascular system
exercise. A lthough the characteristic frequencies and
am plitudes of oscillations have changed, ® ve characteristic
peaks were found in the averaged scalograms of all signals,
both before and after the excercise.
The energy w ithin each frequency interval was calcu-
lated from the wavelet transform of the m easured signals.
The exercise induced a substantial increase in total energy,
comparing to that in the blood ¯ ow while resting (® gure
Downloaded by [California State University of Fresno] at 16:12 29 November 2012
Figure 12. A 60 s segment of the blood ¯ ow signal before and
after exercise (upper plot) and the group median values of
normalized energies and frequencies of all peaks (lower plot).
Figure 13. The total energy of the blood ¯ ow signals, the frequencies of the peaks and the normalized energy of intervals around each
peak, for all ® ve measurements, before and after exercise.
45
13). The energy in the last measurement, which on average
w as m ade between the 35 and 55 m in after exercise, had
decreased almost to the initial value. C hanges in the
frequencies of the peaks, and normalized energies within
the corresponding intervals, are illustrated in the lower
plot of ® gure 12, while the statistical plots are given in
® gure 13. It is evident that exercise induced signi® cant
changes in heart rate, and in the respiration and m yogenic
frequencies. It is well know n that the heart rate and
respiration frequency increase during exercise and remain
higher for som e time afterwards. The myogenic process
not only has a higher characteristic frequency imm ediately
after exercise, but its normalized energy has also increased
signi® cantly.
Indeed, w e may conclude that increased m yogenic
activity, with the characteristic frequency between 0.06 H z
and 0.15 H z, is the m ain change induced by a single episode
of exercise. It is interesting that the normalized energies of
the neurogenic process, w ith its characteristic peak around
0.04 H z, and m etabolic process, w ith its characteristic peak
around 0.01 Hz, are reduced after exercise. All changes
decay shortly after exercise. However, imposed dem ands of
the cells lead to enhanced m etabolic process as w ell as
increased energy of the heart peak w hen exercise is regularly
repeated. Such changes w ere observed in a group of
sportsmen, as com pared to the controls [39].
3.2.6. Irreversible changes. A single episode of exercise
induces short-time changes and, after their decay, the
original dynam ics is re-established. Impairment of the
cardiovascular system, on the other hand, can result in
irreversible changes. In this section w e present the results of
a clinical study w hich included subjects with cardiovascular
diseases. In addition to a control group, consisting of 17
 46 A. Stefanovska and M. BracÏ icÏ
young male subjects, a group of 15 subjects at least 4 days
after m yocardial infarction, and a group of 13 subjects with
diabetes were included in the study. The set of signals
described in section 3.2.2. was measured for all subjects,
while at rest. W e report only the main ® ndings here. T he
complete set of results w ill be presented and discussed
elsewhere.
W hile infarction aŒects mainly the pum ping function of
the heart, diabetes changes the metabolic e ciency of the
cells and by this resistance of the vessels. Consequently, the
heart is permanently imposed to an increased w ork load.
These changes are clearly m anifested in the H RV signal.
Downloaded by [California State University of Fresno] at 16:12 29 November 2012
The total energy of the signal is signi® cantly low er in the
two groups of patients than it is in the control group (® gure
14). For comparison, the total energy of the HR V signal
measured for a subject in com a is presented in the sam e
Figure 14. Total energy of the HRV signal measured in healthy
subjects, patients after myocardial infarction, diabetic patients
and a patient in coma.
Figure 15. Signals recorded from the cardiovascular system of an intensive care patient (left) and corresponding amplitude spectra
(right). HRV is in Hz, other values are in arbitrary units.
® gure. T he energy of this signal is practically zero,
representing a state in which there are almost no variations
in the heart rate.
The decreased variability of the heart rate is m anifested
as a sharpening of the characteristic peaks in all other
cardiovascular signals, as presented in ® gure 15. In this case
the arteriolar and venous pressure w ere measured inva-
sively, via an inserted ¯ uid-® lled catheter and transducer
system, as a part of the routine procedure in intensive care
units. Since the peaks hardly varied at all in tim e, the
Fourier method gav e distinct spectral peaks. M oreover,
peaks appearing at linear combinations of the characteristic
frequencies are also observable. It is also obvious that the
ratio betw een characteristic frequencies tends to becom e
rational.
W e can interpret these results as illustrating that
cardiovascular impairm ent results in signi® cantly reduced
interactions am ong the processes involved in control of the
the blood ¯ ow. T his, am ong other changes, results in
decreased variability of the heart rate and therefore smaller
energy of the H RV signal. T his m ay mean that the other
processes are less e cient in signalling their needs to the
heart, and / or that the heart is unable to cope w ith their
demands.
The reduction in heart variability is often interpreted as
decreased com plexity and chaos in the activity of the heart
[27,52]. However, it is the study of changes in the spectral
components that gives us insight into the pathology of
diseases [53,54]. Evaluation of the contribution of each
spectral com ponent m ight provide a diagnostic and
predictive tool for a whole range of diseases related to the
cardiovascular system.
 Physics of the human cardiovascular system
4. Nonlinear analysis
The ultimate goal of a physical description of a system is a
mathematical formulation as a set of diŒerential equations.
Before we proceed to such a description, w e m ust learn as
much as possible about the system in question. So far, w e
have analysed the cardiovascular signals in the frequency
dom ain. The analyses imply that, on a time scale of ~ 1
minute, ® ve alm ost periodic processes contribute to the
dynam ics of the blood ¯ ow . Developments in the theory of
nonlinear dynam ics, and its numerical applications to
chaotic time series over the last three decades, have
provided several methods for estimating the invariants of
Downloaded by [California State University of Fresno] at 16:12 29 November 2012
the dynam ics from scalar signals, as we shall now describe.
4.1. System characterization
Based on frequency analysis, the following conclusions can
be draw n regarding the physical nature of the system: (i)
well-de® ned spectral peaks at characteristic frequencies are
present in the signals; (ii) peaks at the sam e frequencies
were found in all signals, regardless of the m easured
quantity or the measurement site; (iii) although the peaks
are not sharp, and vary in time for each one of them a ® nite
frequency interval exists within w hich its variations are
con® ned; (iv) the existence of the peaks is not in¯ uenced by
irreversible or reversible changes, implying their robust
nature and the corresponding structural stability of their
sources; (v) from their response to perturbations w e may
conclude that a m utual dependence exists am ong the
sources. T hese results lead to the inference that the source
of each observed peak is an oscillator, and that the
oscillators are m utually coupled. Robustness and structural
stability are characteristic of certain nonlinear oscillators.
The time evolution of an oscillator m ay be obtained as
one of the solutions of a general diŒerential equation
(t) 5 ( 0) 5
d
Ç ( ¹ , (t)) 0, [ .
where (t) is a state vector, Ç (t) 5 d /d t and the ¯ ow is
some general nonlinear function. The vector of control
parameters ¹ is kept constant during the observations. By
changing the control parameter, transitions betw een
diŒerent types of behaviour can be induced. Graphically,
the solutions can be presented as time series
x 1(t), x 2(t) ...x d (t), or alternatively as a trajectory in a
geom etric phase space . d . The phase space, ® lled with
trajectories, is called the phase portrait of the dynam ical
system. After transients are over, the m otion of (t) settles
typically near a subset of the phase space, called an
attractor. The phase space volume occupied by a real
physical system can either be preserved or contracted by
tim e evolution. In the ® rst case, the system is conservative,
in the other dissipative. For dissipative system s, the volum e
occupied by the attractor can be relatively small compared
to the initial volum e of the phase space.
47
The sim plest attractor of a dynam ical system is a stable
® xed point. A periodic motion, on the other hand, results in a
limit cycle. A superposition of periodic m otions, quasi-
periodic motion, has an attractor in the shape of a torus.
F igure 16 presents the phase portrait of the respiratory signal.
T he alm ost periodic nature of the signal, already discussed in
the time and frequency dom ain, results in a limit cycle in
phase space. Since the other system s in¯ uence the respiratory
function only slightly, we may observe the limit cycle in two
dimensions, although it is broadened. In blood ¯ ow , the
contributions of all processes are comparable and a higher
dimensional phase space is needed to portray the attractor.
F or this alm ost quasi-periodic ¯ ow with ® ve characteristic
frequencies the attractor is expected to be a 5-torus.
However, w hen nonlinearities are involved, even a ® nite
dimensional system need not be quasi-periodic. The m otion
w ithin an attractor m ay be unstable in some directions.
T hese instabilities are manifested as an exponential
separation of trajectories. Such systems exhibit a sensitive
dependence on initial conditionsÐ which is the hallm ark of
chaotic behaviour.
The characterization of nonlinear dynam ical system s is
based on geometrical and statistical properties of the
attractor, such as its entropy, various dim ensions (informa-
tion, H ausdorŒ, correlation ...) and Lyapunov exponents
[55,56]. Its statistical properties become relevant as soon as
the dynam ics is su ciently com plicated that geom etrical
information about the shape of the attractor is no longer
available. From here on, it is the statistical theory which
can distinguish diŒerent degrees of com plexity. The basic
tool w hich enables us to measure statistical properties of
the system is the ergodic theory. Ergodic theory states that
a time average equals a space average. Or, in other words,
that a trajectory of the system explores the entire phase
space that is energetically available to it [57].
The numerical algorithms for calculation of the correla-
( 23)
,
tion dimension [58] and Lyapunov exponents [55] have
frequently been applied to characterize nonlinearities in
biological signals (see [59 ± 61] and references therein).
Figure 16. The respiratory signal embedded in two-dimensional
phase space.
 48
Since the very long observation tim e introduces non-
stationarities, the number of points necessary for a
successful estim ation of correlation dimension for high-
dimensional system s [62] can only be obtained by over-
sam pling. Oversam pling, however, emphasizes the noise in
the signal. Therefore the correlation dim ension cannot be
reliably estimated for the case of biological signals resulting
from high-dimensional system s [63]. It was proposed,
however, that the correlation dimension can still be used
for qualitative characterization [64]. By comparing correla-
tion integrals obtained from original tim e series and their
surrogates, i.e. random ized sequences with similar spectral
Downloaded by [California State University of Fresno] at 16:12 29 November 2012
and statistical properties as the original signal, one may
distinguish between determ inistic and stochastic or noise-
dominated signals. W e have show n that all m easured
signals are largely deterministic [40].
The number of points is crucial for all algorithm s used
for the characterization of nonlinear system from m easured
signals, including the estim ation of Lyapunov exponents.
As we shall see below, an algorithm is available for the
estimation of L yapunov exponents, based on an approx-
im ation of the local ¯ ow in phase space. In this case, the
quality of approximation is more important than the
number of points itself.
4.2. The Lyapuno v exponents
The L yapunov or characteristic exponents measure the rate
of convergence or divergence of nearby trajectories in the
phase space. Thus, they determ ine both the stability of the
trajectories and the system’ s sensitivity to initial conditions.
This m akes them one of the most meaningful characteriza-
tions of a nonlinear dynamical system. U sing L yapunov
exponents, one can distinguish between ® xed points and
periodic, quasi-periodic or chaotic m otions.
The Lyapunov exponents characterize the response to
small perturbations of the trajectories of the system (24).
The tim e evolution of a small perturbation is governed by
linearized equation in the tangent space
d Ç (t) 5 D ( ¹ , (t) ) d (t) d ( 0) 5
where D is the Jacobi matrix of the ¯ ow . A t the end of
the last century, A. M . Lyapunov introduced a measure of
average contraction of the perturbation to a given
trajectory as
1 5 d (t) | d 0 5
¸( d 0) 5 lim log
t® ¥ t 5 d 05
today know n as the Lyapunov exponents. U sing d linearly
independent initial conditions d 01 .. . d 0d , one obtains a
fundam ental system of solutions ¸1 ³ ¸2 .. . ³ ¸d . For
ergodic systems the set of ¸i does not depend on the initial
condition d 0i and so the ¸i are global properties of the
attractor [65]. As such, they bear no inform ation about the
A. Stefanovska and M. BracÏ icÏ
local behaviour on the attractor. Therefore, local L yapu-
nov exponents were introduced [66]. In this case, the limit
t® ¥ is not taken in equation (25) . T hese values may vary
signi® cantly over the attractor, but their means converge
tow ards the global exponents.
If any ¸i is positive, small perturbations will grow
exponentially. If all ¸i are negative, any perturbation w ill
decrease and the attractor is a stable ® xed point. In the case
of a zero exponent, the size of the perturbation does not
change in time. A stable periodic state, for exam ple, has one
zero exponent corresponding to a perturbation tangent to
the limit cycle, and all the other exponents are negative. A
quasi-periodic system with k incom mensurate frequencies
has k zero exponents, and all others are negative. Every
attractor of a sm ooth dynam ic system, given by equation
(23), has at least one zero Lyapunov exponent correspond-
ing to a perturbation tangent to the trajectory. Such a
perturbation simply m oves one along the sam e orbit. The
exponents of a Ham iltonian system com e in conjugate pairs,
consisting of a positive and a negative exponent of the same
magnitude, and two of them vanish [55]. One zero exponent
is associated with the conservation of energy and the other
with the fact that the evolution equations are diŒerential.
For system s whose equations of m otion are known
explicitly there is a straightforward method of com puting
all Lyapunov exponents [67]. This method cannot, how -
ever, be applied directly to experimental data. There are
two approaches to estim ating the exponents from m easured
signals. In the ® rst, introduced by W olf et al. [68] and
Rosenstein et al. [69], two nearby points in the phase space
are followed and only the largest exponent is evaluated.
The second approach, introduced by Eckm ann and Ruelle
[55], as well as by Sano and Sawada [70], is based on
estimating the Jacobians of the m ap.
For scalar time series s (t) the attractor can be
reconstructed by the method of delay coordinates, intro-
duced by Packard et al. [71]
( t) 5 [s (t), s ( t 1 s ), . .. , s ( t 1 (d 2 1) s ) ], ( 26)
where t 5 t s , 2t s , . .. n t s 2 (d 2 1) s , d is the dim ension of the
d 0, ( 24)
embedding space and the time lag s is an integer multiply of
the sampling time t s .
In the points along a chosen trajectory on the embedded
attractor, a set of neighbouring points is evolved for a
chosen time (evolution time). Based on these two sets, the
local ¯ ow is approximated by a set of basis functions. From
, ( 25) subsequent Jacobi matrices, d Lyapunov exponents are
calculated (for details see [56]). The great advantag e of this
method compared to the trajectory tracing method is that
one can deal with arbitrary vectors in tangent space while
the observed data points are used only to approximate local
¯ ow. Thus, we can calculate all exponents (including
negative ones) as long as the approximation of the ¯ ow is
adequate.
 Physics of the human cardiovascular system
The number of calculated exponents is equal to the
dimension of the reconstructed phase space d. If d is too
small, the attractor is not able to unfold and the calculated
exponents are erroneous. O n the other hand, if d is too
large, the reconstructed attractor is contained in a
subm anifold of dim ension m < d, and d Ð m spurious
exponents that are unrelated to the dynam ics of the system
are obtained. U nder time reversal, true exponents change
sign, w hile the spurious ones do not [56]. In this way, they
can be identi® ed.
In low em bedding dimensions, the global Lyapunov
exponents of the blood-¯ ow signal vary from one dim en-
Downloaded by [California State University of Fresno] at 16:12 29 November 2012
sion to the other, but as the dim ension reaches 10, two
patterns are observed: either four paired and one zero or
® ve paired exponents are calculated. In the latter case, one
pair equals zero w ithin the calculation error [72,73]. Am ong
the exponents of the reversed signal, four or ® ve pairs were
observed again. The other exponents were found to be
negative and of the same m agnitude as some of the
exponents obtained from the original signal, also with
negative sign. Therefore, they can be identi® ed as spurious.
To reveal local properties of the attractor, local
Lyapunov exponents were calculated. F igure 17 presents
the distribution of the values of the ® rst nine exponents,
calculated in an 11-dimensional em bedding space. Paired
values are obtained againÐ four paired and one zero
exponent.
The appearance of an exponent equal to zero within the
calculation error shows that the blood ¯ ow dynam ics is
governed by a determ inistic system . Paired values are an
indication of the almost Ham iltonian nature of the system
that regulates the ¯ ow of blood on a time scale of minutes.
M oreover, the number of pairs, nam ely ® ve, support the
hypothesis that ® ve oscillatory subsystems are involved in
the regulation.
5. System description
W e have obtained some convincing evidence that the
system is largely deterministic: the existence of character-
istic frequency peaks in all m easured signals; their non-
Figure 17. The distribution of local exponents of the blood ¯ ow signal of a healthy person, calculated from 40 diŒerent starting points.
49
vanishing autocorrelation function [40]; a zero L yapunov
exponent of peripheral blood ¯ ow signal; and diŒerences
between the correlation integrals of original signals and
their surrogates.
It is therefore possible to describe the dynam ics within
the cardiovascular system as a solution of a set of
diŒerential equations. The oscillatory nature of the
processes involved in blood ¯ ow regulation, and their ® nite
number, provided the starting point for the formulation of
these equations. It must be borne in mind that the processes
are mutually interdependent. Their joint action is directed
tow ards a single goal: to provide matter and energy
continuously to each cell. To achieve this goal, they m ust
act collectively so that, at any given m om ent, the reaction
of each depends on the state of all the others. In the
m athematical formulation, these mutual dependences
appear as couplings between the oscillators.
5.1. Coupling s
T he eŒect of coupling between tw o oscillators on their
behaviour depends on its strength. W eak couplings result in
a variation of the characteristic frequencies of the
oscillators, w hile strong couplings may lead to qualitative
changes in the system behaviour, i.e. phase transitions.
If couplings did not exist in the cardiovascular system,
w e w ould have sharp peaks and the characteristic
frequencies, which in healthy subjects at rest are only
tending towards a rational ratio, may further become
com mensurate. The resonance phenom enon, for example,
is one of the possible scenarios of how the system collapses.
T he other possible scenario is that in this case the
characteristic frequencies m ight becom e completely incom -
m ensurate. T he couplings enable the exchange of inform a-
tion am ong the processes and are therefore fundam ental to
the appropriate functioning of the cardiovascular system.
Understanding the physical and physiological nature of
these couplings is obviously essential to understanding how
the whole system works. The frequency and am plitude of
each observed oscillation re¯ ect both the activity of the
oscillator itself and the eŒect of all couplings. Although the
 50 A. Stefanovska and M. BracÏ icÏ

eŒect of couplings cannot be m easured separately, we may First, we shall introduce the ¯ ow (x i ) and its velocity (y i ),
however extract it from the data using a variety of contributed by each oscillator separately and then consider
techniques for the analysis of phase, frequency and their collective activity. T he sam e type of oscillator will be
am plitude couplings. used for all ® ve processes. In view of the experimental
One of earliest couplings to be identi® ed, and the m ost evidence showing that the oscillators are both robust and
frequently investigated, is that between the heart and nonlinear in nature, a form
respiratory activity, know n as the respiratory arrhythm ia. xÇ i 5 2 x i qi 2 y i 2p f i ,
Although it is usually studied from the H RV signal, it is
12
also visible in the peripheral blood ¯ ow signal. W e present yÇ i 5 2 y i qi 1 x i 2p f i , qi 5 a i ((x 2i 1 y 2i ) / 2 ai ) , ( 27)
in ® gure 18 the corresponding w avelet transform . Fre-
quency and am plitude variations of the ¯ ow com ponent, was chosen [38]. The index i denotes the ith oscillator, i= 1
driven by the heart are also show n. They are presented as the heart, i= 2 the respiratory, i= 3 the m yogenic, i= 4 the
Downloaded by [California State University of Fresno] at 16:12 29 November 2012

projections of the peak value of the wavelet transform onto
the tim e-frequency and time-am plitude planes. It is notable
that the heart frequency is modulated by the frequency of
respiration, while the am plitude of the ¯ ow, synchronized
with the heart beat, is modulated by slow er oscillatory
processes, dom inated by the m yogenic frequency of around
0.1 H z.
W e can see from the scalogram s that the peaks at lower
frequencies vary in time as well. It may be assum ed that
couplings, similar to the one presented in ® gure 18, also
exist am ong the other oscillators. Based on the analysis of
the system in diŒerent states, norm al and perturbed, w e
may presume that all couplings are weak.
5.2. Coupled oscillators
In the following, we present diŒerential equations which
govern the regulation of the peripheral blood ¯ ow in a
single point of the vessels network. W hile the N avier± Stokes
equations, presented in section 2 give the E uler description
of the ¯ ow , w e use the L agrangian description to follow the
neurogenic and i= 5 the metabolic oscillator. a i is the
am plitude, and f i the characteristic frequency of the
oscillator. T he constant a i determ ines the stability of the
limit cycle. It is interesting to compare (27) with the van der
Pol oscillator [74], w hich has often been used to describe
heart activity. The latter is a relaxation oscillator, and m ost
of the energy is exchanged in a particular part of a cycle. In
the cardiovascular system , this is true only at the output of
the heart, whereas in the periphery energy is uniformly
exchanged during the w hole cycle. T he shape of the ¯ ow in
one heart beat cycle, presented in ® gure 4, also diŒers from
the solution of the van der Pol equations.
Equation (27) describes an autonomous oscillator. The
couplings lead to additional terms in the equation. The
present understanding of couplings allow s us to specify
only the sign of coupling term s. T herefore, linear couplings
will be used. For the heart oscillator, we obtain
xÇ 1 5 x 1q1 2 y 12p f 1 1 g 2x 2 2 g 3x 3 2 g 4x 41
g 5x 5 2 g 6( 1 2 2) ,

contribution of individual processes to the ¯ ow. yÇ 1 5 y 1q1 1 x 12p f 1 1 g 2y 2 2 g 3y 3 2 g 4y 4 1 g 5y 5 . ( 28)

1 and 2 represent the in¯ ow and out¯ ow of blood at the

point observed. The coe cient g 2 represents the in¯ uence of
respiration to the heart rate. D uring inspiration the heart
beats faster, w hile in the expiration phase its rate decreases.
Both the myogenic and neurogenic activities contribute to the
stiŒness of the vessels and by this increase the resistance to the
¯ ow. T herefore, their in¯ uences are introduced by 2 g 3 and
2 g 4. T he m etabolic activity reduces stiŒness and decreases
resistance to the ¯ ow. T his coupling is introduced via g 5.
As it is evident from the H RV signal, all processes have
an im pact on the pum ping action of the heart and through
this on the ¯ ow and velocity of the blood. The respiratory
activity, on the other hand, is in¯ uenced least by the rest of
the system .
The equations for the respiratory ¯ ow and velocity
Figure 18. The wavelet transform of the peripheral blood ¯ ow
signal within the frequency interval corresponding to heart components can then be written as
activity. Variations of frequency and amplitude are plotted as xÇ 2 5 x 2 q2 2 y 2x 2 1 h 4x 4 1 h 5x 5 1 h 6( 1 2 2) ,
projections of the peak value onto the time± frequency and time±
amplitude planes. yÇ 2 5 y 2q2 1 x 2x 2 1 h 4y 4 1 h 5y 5 . ( 29)
 Physics of the human cardiovascular system
Only the neurogenic ( h 4) and m etabolic ( h 5) activities
modulate the ¯ ow com ponent, driven by the pressure
diŒerence generated by the respiratory activity. The
neurogenic in¯ uence is considered because breathing is
under a continuous autonom ous control. M oreover, the
bilateral transection of the vag us nerve has been shown to
result in slower breathing and deeper inspiration [2]. The
lungs also adapt their activity to the metabolic needs of the
cells. T he higher the m etabolic activity the deeper and
faster the respiration is.
In the proposed model, the other three systemsÐ
myogenic, neurogenic and m etabolicÐ w ere also described
Downloaded by [California State University of Fresno] at 16:12 29 November 2012
in an analogous w ay. The experimentally observed
characteristic frequencies were used to characterize each
of them, and their interrelations were approxim ated by
linear couplings, based on physiological evidence. H ow -
ever, additional studies are necessary to elucidate further
the origin and nature of these slower oscillators and their
mutual couplings.
N ote that, in form ulating our m athematical description,
we have consciously chosen an oscillator as the basic unit
for construction of our model. A lthough this choice is
based on the experimental and clinical evidence discussed
above, and although m any biological systems have been
show n to be governed by oscillators [75 ± 77], it is not the
only possible choice. In addition to van der Pol, Clynes [78]
also described the heart activity with an oscillator. H e used
a harmonic oscillator w ith a variable frequency and two
separate equations to describe the variability during
inspiration and expiration. T he m ajority of the proposed
models, how ever, interpret the observed oscillations in the
HRV and blood pressure signals in terms of nonlinearities
and tim e delays [79 ± 85]. They concentrate mainly on
short-term blood pressure control mechanisms, including
respiratory oscillations and oscillations w ith a period of
~ 10 s.
The reason w hy we favour coupled oscillators as a model
of the processes governing blood ¯ ow through the
cardiovascular system , rather than a set of delay-diŒer-
ential equations, is tw ofold. F irst, the delay is introduced
somew hat arti® cially to create an oscillatory behaviour that
is already inherent for our basic units. In reality, there is
physiological evidence that the underlying processes can
also function autonomously, supporting the idea of an
oscillator-based model. Secondly, whereas delays are
certainly important in reactions to isolated transients, the
cardiovascular system is constantly reacting to sm all
perturbations. Its existence is determ ined by its ability to
cope on a continuing basis w ith changes resulting from the
mutual interaction of the processes involved. In the coupled
oscillator model, the time needed for mutual interactions is
manifested as phase shifts betw een the oscillators.
The full version of the proposed model consists of ten
coupled ® rst order diŒerential equations. A n analytic
51
solution of such a system w ould be, to say the least, hard
to obtain. T he alternative possibility is a numerical
calculation, w hich is how ever sensitive to the choices of
integration method and of parameter values. A recently
discussed method based on analogue circuits [86] seem s
potentially a promising way of studying the solutions of
high order diŒerential equations of this kind.
For the present, how ever, instead of solving the
equations directly, we have obtained a solution starting
from the known physical and physiological properties of
the cardiovascular system [87]. The eŒects of all experi-
m entally observed processes on one pumping cycle of the
heart were considered and a m athematical description of
oscillations in the blood ¯ ow was acquired. Physical
characteristics of vessels were described by the windkessel ²
model [88]. This consists of a capacitor representing the
com pliant aorta, and a resistor representing the stiŒer
peripheral vessels, connected in parallel. Earlier variants of
this model attempted to describe either the wave m otion of
the blood [89] or the regulatory processes [8,90], but dealt
separately with individual sections of the cardiovascular
system. Given the couplings that are know n to exists
between the processes, however, a proper understanding of
the ¯ ow dynam ics clearly requires consideration of the
physics of the system as a w hole, treated as an entity.
5.3. Relations am ong the oscillators
T he relationship between the blood ¯ ow , the pressure P
and the vessel’ s resistance R is de® ned by O hm ’ s law
P
5 .
R
E ach of these variables oscillates around its steady-level
5 0 1 D 0, P 5 P0 1 D P0 , R 5 R0 1 D R0 .
T he ¯ uctuations are determ ined by the state variables of the
oscillators x i . From the existing physiological evidence
discussed above, we may assum e that the heart’ s activity
(x 1) directly increases the blood ¯ ow value
D 5 e 1x 1,
and the respiratory activity (x 2) increases the pressure
diŒerence
Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð
² Th e w in dkessel m odel is based on m echanical properties of vessels and
describes the pulse-w ave propagation of the blood. V essels that consist of
a rela tively large proportion of elastic fibres, such as the aorta, the
pulm onary arteries and the adjacent parts of the great arteries, are called
w indkessel vessels. W ith each pulse any given segm ent of such a vessel
distends to store a volum e of blood. As it subsequently contracts back to
its original dim ensions, it pushes blood on to the next segm ent. Th e nam e
w indkessel (Germ an for air cham ber) has been given to these vessels and
their function because of the resem blance to the air-filled cham bers that
sim ilarly affect the velocity and pressure of fluids driven by pistons
through system s of pipes.
 52 A. Stefanovska and M. BracÏ icÏ

the left atrium [92]. There, the value of P (l, t) becomes 0.

D P 5 e 2x 2 .
By analogy, w e may assume the value of ¯ ow at this
Furthermore, the myogenic (x 3) and the neurogenic (x 4) boundary condition to be (l, t) 5 0.
activity constrict the vessels and resist the ¯ ow , while the The wave motion of the blood along the cardiovascular
metabolic activity (x 5) dilates the vessels and decreases its system can be w ritten as
resistance
­ ­ P
D R 5 e 3x 3 1 e 4x 4 2
5 2 j 1 2 j 2x 5 , j i >0
e 5x 5 . ­ t ­ z

From the point of view of the capillary bed, where the cells
­ P ­
of the human body have direct access to the blood, the 5 2 ¹1 1 ¹ 2x 3 1 ¹ 3x 4 , ¹i > 0, ( 33)
­ t ­ z
peripheral ¯ ow consists of an in¯ ow 1, which is driven by
the heart through the arteries, and an out¯ ow 2 through where j i , ¹ i are control param eters.
Downloaded by [California State University of Fresno] at 16:12 29 November 2012

the veins, which is pumped by the pressure diŒerence The values of these control parameters in equations (31)±
generated by the lungs (33) may diŒer with respect to the part of the system under
0 consideration: arterial, capillary or venous. T he blood
Ç 1 5 n (D 1 D R2 1) ,
R0 pressure values are higher at the origin of the aorta and
lower at the entrance of the vena cava into the heart. The
Ç 1
2 5 v ( D P 2 2) . ( 30) arterial ¯ ow near the heart is dom inated by the heart pum p
R0
oscillations, w hile on the way to the capillaries the
The constants n and v determ ine the ¯ ow dynam ics. myogenic activity takes charge over the regulation of the
¯ ow. At the capillary bed the am plitude of oscillations of
the metabolic activity dom inates.
5.4. The ¯ ow along a vessel The model w as built to summarize the experimental
Above, w e have considered the oscillations of the blood ® ndings and known physiological facts. It still needs to be
¯ ow at a single point in the cardiovascular system. T he evaluated, analytically and numerically, and further experi-
contribution of each process diŒers at diŒerent points of ments are also needed. W ith appropriate choice of
the system . H ow ever, all ® ve characteristic frequencies parameter values, we expect it to serve in obtaining the
were found in diŒerent physiological signals and on initial conditions for locally oriented models, such as
diŒerent sites along the vessel’ s network. Therefore, w e described by N avier± Stokes equations. M oreover, equa-
may assum e that blood ¯ ow s in the form of travelling tions for wave motion of the blood may reduce locally to
waves. The heart’ s contribution to the ¯ ow at each point Navier± Stokes equations.
results not only from the in¯ uences of local regulatory
mechanism s, but also from the values of pressure and ¯ ow
along the entire system. T herefore, equations (28) and (29) 6. O utlook
were rearranged [91] On the way tow ards reaching an understanding of the
cardiovascular system, physics, m athematics, physiology
xÇ 1 5 x 1q1 2 y 1x 1 1 g 2x 2 2 g 3x 3 2 g 4x 4 1 g 5x 51 and clinical m edicine meet under the wings of nonlinear
l l dynam ics. Therefore, the current comprehension of the
g 6 P ( z , t) d z 1 g 7 ( z , t) d z , physical and physiological properties of the processes
0 0
involved in the control of blood circulation has been
yÇ 1 5 y 1q1 1 x 1x 1 1 g 2y 2 2 g 3y 3 2 g 4y 4 1 g 5y 5 , ( 31)
combined with experimental ® ndings in order to arrive at a
l characterization of the system dynam ics.
xÇ 2 5 x 2q2 2 y 2x 2 1 h 4x 4 1 h 5x 5 1 h 6 P ( z , t) d z 1 Processes on many diŒerent time scales and of w idely
0
l diŒerent structural complexity are re¯ ected in a single cycle
h 7 ( z , t) d z , of blood through the system . On one hand, this fact m akes
0
the understanding of the system di cult. However, we have
yÇ 2 5 y 2q2 1 x 2x 2 1 h 4y 4 1 h 5y 5 . ( 32)
show n that the system exhibits clear signatures of
Here, ( z , t) is the ¯ ow at any point of the circulatory determ inistic dynam ics. T he sam e dynam ic properties
system. It is generated by the heart, therefore characterize all signals generated by the system , regardless
( 0, t) 5 e 1x 1( t). The pressure P is generated by the of the measurement site, and are also preserved in tim e.
lungs and P ( 0, t) 5 e 2x 2(t) . W e have assum ed the ¯ ow in Therefore, the system can be treated as a single entity. This,
the z direction, and l is a length in this direction. T he on the other hand, provides a simpli® cation that oŒers the
blood returns to the right atrium of the heart at a pressure possibility of evaluating the m icroscopic and m acroscopic
of 0 Pa , and at almost 0 Pa from the pulmonary vein to mechanisms together.
 Physics of the human cardiovascular system
Several methods of linear and nonlinear analysis have
been applied to extract the dynam ics characterizing the
measured signals. Although an enormous development of
nonlinear dynamics has occurred during the last three
decades, m uch w ork remains to be done on the theory and
num erical m ethods for dealing with ® nite high-dimensional
systems. W e believe that studying the physics of the
cardiovascular system may provide a motivation for further
progress in the ® eld. Its understanding is not only a
theoretical challenge, but m ay also provide useful diag-
nostic and predictive tools for this life-giving system .
Downloaded by [California State University of Fresno] at 16:12 29 November 2012
Acknow ledgem ents
W e w ish to thank Peter V. E . M cClintock for continuous
encouragem ent and support during the writing of the
manuscript and his most useful comm ents. A .S. is grateful
to Herm ann H aken and colleagues at the Institute of
Theoretical Physics I, University of Stuttgart for num erous
discussions. She is indebted to the A lexander von Hum -
boldt Foundation.
The work was supported by the Slovenian M inistry of
Science and Technology and partially by the Slovenian±
Germ an Science and T echnology co-operation. W e ac-
knowledge E . W eber-Jensen for providing m easurem ents
on the subject in coma, D . SÏ tajer and V. U rbancÆicÆ
-R ovan
for patient selection and clinical support during recordings
on patients after m yocardial infarction and diabetic
patients, and M . CÆ encÆ
ur for technical support. Some of
the measurements w ere made possible by M oor Instru-
ments, who kindly lend us the laser Doppler ¯ ow meter.
References
[1] K eele, C. A., N eil, E., and Joels, N . (editors), 1982, Samson Wright’ s
Applied Physiology (Oxford: Oxford U niversity Press).
[2] Schm idt, R. F., and Th ews, G . (editors), 1989, H uman Physiology
(Berlin: Springer-V erlag).
[3] Fung, Y. C ., and Zw eifach, B. W ., 1971, Ann. R ev. Fluid Mech., 3,
189± 210.
[4] Zweifach, B. W., 1966, ASM E, Biomechanics, 1, 121± 131.
[5] K arnowsky, M . J., 1968, J. Gen. Physiol., 52, 64± 95.
[6] M cGrew, R. E., 1985, Encyclopedia of Medical History (London:
M acm illan).
[7] Lippold, O. C. J., and Winton, R. F. (editors), 1979, H uman
Physiology (Edinburgh: Churchill Livingstone).
[8] Strackee, J., and Westerhof, N ., (editors), 1993, The Physics of Heart
and Circulation (Bristol: Institute of Physics Publishing).
[9] Sutera, S. P., and Skalak, R., 1993, Ann. R ev. Fluid M ech., 25, 1± 19.
[10] van Steenhoven, A. A., 1993, Velocity pro® les in large arteries. The
Physics of Heart and Circulation, edited by J. Strackee and
N . W esterhof (Bristol: Institute of Physics Publishing), pp. 295± 320.
[11] Fung, Y. C., 1996, Biomechanics: C irculation (New York: Springer).
[12] H yndm an, B. W ., K itney, R. I., and Sayers, B. McA., 1971, Nature,
233, 339± 341.
[13] Sayers, B. McA., 1973, Ergonom ics, 16, 16± 17.
[14] Chess, G . F., Tam , R. M . K ., and Calaresu, F. R., 1975, A mer. J.
Physiol. , 228, 775-780.
53
[15] Akselrod, S., G ordon, D ., Ubel, F. A., Shannon, D. C., Berger, A. C.,
and C hoen, R. J., 1981, Science, 213, 220± 222.
[16] D i Renzo, M., Parati, G ., Pedotti, A., and Zanchetti, A. (editors),
1997, Frontiers of Blood Pressure and Heart R ate Analysis, (Am-
sterdam: IOS Press).
[17] Task Force of the ESC and the N ASPE, 1996, C irculatio n, 93, 1043±
1065.
[18] H ales, S., 1773, Statical Essay s II, H aemastaticks (London: Innings
M anby).
[19] Ludw ig, C., 1847, A rch. Anat. Physiol., 13, 242± 302.
[20] H ildebrandt, G ., 1987, Th e autonomous tim e structure and its reactive
m odi® cations in the hum an organism . Temporal Disorder in H uman
Oscillatory Systems, edited by L. Resing, U. van der H eiden and M . C.
M ackey (Berlin: Springer), pp. 160± 174.
[21] Fourier, J. B. J., 1888, Theorie analytique de la chaleur. Oeuvres de
Fourier, Tom e premier, edited by G . D arboux (Paris: G authiers-
Villars).
[22] Press, W . H ., Teukolsky, S. A., Vetterling, W . T., and Flannsry, B. P.,
1992, Numerical Recipes in C (Cam bridge: C am bridge University
Press).
[23] M orlet, J., 1983, Sam pling theory and w ave propagation, Issues in
Acoustic Signal /Image Processing and Recognition, N ATO ASI Series,
Vol. I, edited by C. H. C hen (Berlin: Springer-V erlag).
[24] G rossm ann, A., and M orlet, J., 1984, SIA M J. M ath. A nal., 15, 723±
736.
[25] D aubechies, I., 1992, Ten Lectures on Wavelets (Philadelphia:
SIAM ).
[26] K aiser, G ., 1994, A Friendly Guide to W avelets (Boston: Birkh Èa user).
[27] Thurner, S., Feurstein, M . C ., and Teich, M . C., 1998, Phys. Rev.
Lett., 80, 1544± 1547.
[28] Stefanovska, A., Leger, P., BracÆ icÆ
, M ., BracÆ
icÆ
, T., Bendayan, P., and
Boccalon, H ., 1998, Technol. Health C are (in the press).
[29] M aim an, T. H ., 1960, Nature, 187, 493± 494.
[30] C umm in s, H . Z., K nable, N ., and Yeh, Y., 1964, Phys. Rev. Lett., 12,
150± 153.
[31] Riva , C., Ro ss, B., and Benedek, G. B., 1972, Invest. Opthalm ol., 11,
936± 944.
[32] Stern, M . D ., 1975, Nature, 254, 56± 58.
[33] N ilsson, G. E., Tenland, T., and O È berg, P. A., 1980, IEEE Trans.
Biomed . Eng., 27, 597± 604.
[34] Salerud, E. G., Tenland, T., N ilsson, G . E., and O berg, P. A ., 1983,
Int. J. M icrocirc.: Clin. Exp., 2, 91± 102.
[35] Perim ed, A. B., J Èarf Èa lla , Sweden, http:// www.perim ed.se.
[36] W ilkin, J. K ., 1986, Am. J. Physiol., 250, H 765± H768.
[37] H oŒman, U., Yanar, A., Franzeck, U . K., Edvards J. M ., and
Bollinger, A., 1990, M icrovascular R es., 40, 293± 301.
[38] Stefanovska, A., 1992, Dissertation, U niversity of Ljubljana, Ljublja-
na.
[39] BracÆ icÆ
, M., and Stefanovska, A., 1998, Bull. Math. Biol. 60, 919 ± 935.
[40] Stefanovska, A., and K rosÆ elj, P., 1997, Open Sys. & Information Dyn.,
4, 457± 478.
[41] BracÆ icÆ
, M ., and Stefanovska, A., 1998, Open Sys. & Information Dyn.
(in press)
[42] K itney, R. I., Fulton, T., McD onald, A. H ., and Linkens, D . A., 1985,
J. Biomed. Eng., 7, 217± 224.
[43] M uÈ ck-W eym ann, M . E., Albrecht, H.-P., H ager, D., Hiller, D .,
H ornstein, O. P., and Bauer, R. D., 1996, Microvasuclar Res., 52, 69-
78.
[44] Johnson, P. C ., 1991, NIPS , 6, 41± 42.
[45] Bertuglia, S., Colantuoni, A., and Intaglietta, M., 1994, M icrocirc.
Res., 48, 68± 84.
[46] G ri th, T. M ., and Edwards, T. H ., 1994, Am. J. Physiol., 266,
H 1801± H 1811.
[47] K astrup, J., B uÈ hlow , J., and Lassen, N . A., 1989, Int. J. Microcirc.:
Clin. Exp., 8, 205± 15.
 54 A. Stefanovska and M. BracÏ icÏ
[48] Stefanovska, A., BracÆ icÆ
, M ., and Kvernm o, H. D., 1998, IEEE Trans.
Biomed . Eng. (subm itted).
[49] K vernm o, H . D., Stefanovska, A., K irkeb en, K. A., and K vernebo,
K ., 1998, M icrovascular R es. (subm itted).
[50] Iversen, G . R., and G ergen, M., 1997, StatisticsÐ The C onceptual
Approach (New York: Springer).
[51] K vernm o, H . D., Stefanovska, A., BracÆ
icÆ
, M ., K irkeb en, K. A., and
K vernebo, K., 1998, M icrovascular R es. (in the press).
[52] Ivanov, P. C h., Ro senblum , M . G ., Peng, C.-K., M ietus, J., H avlin,
S., Stanley, H. E., and Goldberger, A. L., 1996, Nature, 383, 323±
327.
[53] Lom bardi, F., Sandrone, G., Torzillo, D., and M alliani, A., 1995,
Spectral analysis of blood pressure and heart rate in patients with
m yocardial infarction. C omputer Analysis of Cardiovascular
Downloaded by [California State University of Fresno] at 16:12 29 November 2012
Signals, edited by M . D i Renzo, G. M ancia, G . Parati,
A. Pedotti and A. Zanchetti (Am sterdam : IOS Press), pp. 267±
274.
[54] W eise, F., and H eydenreich, F., 1995, Pow er spectrum analysis of
heart rate in diabetic patients: A marker of autonomic dysfunction.
Computer Analysis of Cardiovascular Signals, edited by M. D i Renzo,
G . M ancia, G . Parati, A. Pedotti and A. Zanchetti (Am sterdam: IOS
Press), pp. 283± 292.
[55] Eckmann, J.-P., and Ruelle, D ., 1985, Rev. M od. Phys., 57, 617± 656.
[56] Abarbanel, H . D . I., Brown, R., Sidorowich, J. J., and Tsim iring,
L. Sh., 1993, R ev. M od. Phys., 65, 1331± 1392.
[57] Lichtenberg, A. J., and Lieberm an, M . A., 1983, Regular and C haotic
Dynamic s (New York: Springer-V erlag).
[58] G rassberger, P., and Procaccia, I., 1983, Phys. Rev. Lett., 50, 346±
349.
[59] G lass, L., and K aplan, D ., 1993, M edic. Progr. Technol., 19, 115± 128.
[60] G ri th, T. M., 1996, Cardiovascular Res., 31, 342± 358.
[61] Peng, C. K ., H alvin, S., H ausdorŒ, J. M ., Mietus, J. E., Stanley, H. E.,
and Goldberger, A. L., 1997, Fractal m echanisms and heart rate
control: Long-range correlations and their breakdown w ith disease.
Frontiers of Blood Pressure and H eart Rate Analysis, edited by M . Di
Renzo, G . Parati, A. Pedotti and A. Zanchetti (Amsterdam : IOS
Press), pp. 3± 14.
[62] Eckmann, J.-P., and Ruelle, D ., 1992, Physica D , 56, 185± 187.
[63] Stefanovska, A., Strle, S., and K roselj, P., 1997, Phys. Lett. A, 235,
24± 30.
[64] Theiler, J., Eubank, S., Longtin, A., G aldrikian, B., and Farmer, J. D .,
1992, Physica D , 58, 77± 94.
[65] Oseledec, V . I., 1968, Trans. Moscow Math. Soc., 19, 197± 231.
[66] Abarbanel, H . D . I., Brow n, R., and Kennel, M . B., 1991, J. Nonlinea r
Sci., 1, 175± 199.
[67] Bennetin , G ., G algani, L., G iorgilli, A., and Strelcyn, J. M ., 1980,
M eccanica, 15, 9± 30.
[68] W olf, A., Swift, J. B., Swin ney, H., and V astano, J. A., 1985, Physica
D , 16, 285± 317.
[69] Rosenstein, M . T., Collins, J. J., and De Lu ca, C. J., 1993, Physica D,
65, 117± 134.
[70] Sano, M ., and Saw ada, Y., 1985, Phys. R ev. Lett., 55, 1082± 1085.
[71] Packard, N ., Crutch® eld, J., Farm er, D ., and Shaw, R., 1980, Phys.
Rev. Lett., 45, 712± 716.
[72] BracÆicÆ
, M ., and Stefanovska, A., 1998, Bull. Math. Biol., 60, 417± 433.
[73] BracÆicÆ
, M ., and Stefanovska, A., 1998, Lyapunov exponents of
simulated and measured quasi-periodic ¯ ows. Signal Analysis and
Prediction, edited by A. Prochazka, J. U hlir, J. P. Rayner and
N . G. K ingsbury (Boston: Birkh Èauser), pp. 479± 488.
[74] van der Pol, B., and van der M ark, J., 1928, Phil. M ag. [Suppl.], 6,
763± 775.
[75] H aken, H ., 1987, A dvanced Synergetics (Berlin: Springer).
[76] M urray, J. D., 1989, Mathematical Biology (Berlin: Springer).
[77] W in free, A. T., 1990, The G eometry of Biologica l Tim e (Berlin:
Springer).
[78] Clynes, M ., 1960, J. Appl. Physiol., 15, 863± 874.
[79] K itney, R. I., Selman, A., M cDonald, A., and Linkens, D ., 1982,
Automedica, 4, 141± 153.
[80] deBoer, R. W ., K arem aker, J. M ., and Strackee, J., 1987, A m. J.
Physiol. , 253, H 680± H 689.
[81] Baselli, G., Cerutti, S., C ivardi, S., M alliani, A., and Pagani, M., 1988,
IEEE Trans. Biopmed . Eng., 35, 1033± 1046.
[82] Saul, J. P., Berger, R. D., H ui C hen, M ., and Choen, R. J., 1989, Am.
J. Physiol., 256, H 153± H 161.
[83] tenV oorde, B. J., Faes, Th . J. C., Janssen, T. W . J., ScheŒer G. J., and
Rom pelman, O., 1995, Respiratory modulation of blood pressure and
heart rate studied with com puter m odel of barore¯ ex control.
Computer Analysis of C ardiovascular Signals, edited by M . D i Renzo,
G . Mancia, G . Parati, A. Pedotti and A. Zanchetti (Am sterdam: IOS
Press), pp. 119± 134.
[84] Seidel, H ., and Herzel, H ., 1995, M odelling heart rate variability due
to respiratio n and barore¯ ex. Dynamical Phenomena in Livin g
Systems, edited by E. M osekilde and O.G . M ouritsen (Berlin:
Springer), pp. 202± 228.
[85] Cavalcanti, A., and Belardinelli, E., 1996, IEE E Trans. Biomed . Eng.,
43, 982± 989.
[86] Luchinsky, D . G ., and M cClintock, P. V. E., 1997, Nature, 389, 463± 466.
[87] N ovkovski, N ., and Stefanovska, A., Oscillations in the blood
circulation, preprint.
[88] Frank, O., 1899, Z . Biol., 37, 483± 526.
[89] Li, J. K .-J., 1996, Comparative Cardiovascular Dynamics of M ammals
(Boca Raton: CRC Press).
[90] K itney, R. I., and Seydenjad, S. R., 1997, Investigation of HRV by
m odel analysis. Frontiers of Blood Pressure and H eart R ate Analysis,
edited by M . Di Renzo, G. Parati, A. Pedotti and A. Zanchetti
(Am sterdam: IOS Press), pp. 67± 88.
[91] Stefanovska, A., Strle, S., and BracÆ
icÆ
, M ., H aken, H., 1998, Open Sys.
& Information Dyn. (in the press).
[92] Folkow, B., and N eil, E., 1971, Circulation (New York: Oxford
U niversity Press).
Aneta Stefanovska is head of the G roup of
Nonlinear Dynam ics and Synergetics, in the
Faculty of Electrical Engineering, University of
Ljubljana, Slovenia , where she received her PhD
degree. Part of the research for her PhD she
carried out at the Institute of Theoretical Physics
and Synergetics at the University of Stuttgart,
Germ any, to which she returns regularly, cur-
rently as a Fellow of the Alexander von
Hum bold t Foundation . She initiated her PhD
studie s by investigating the eŒects of electric
currents on bloo d ¯ ow in soft tissues, but then
became fascinated by the nonlinea r and complex
dynamic s of the blood circulation. The resultan t
experiments and analyses led to her proposa l in
1992 of a m odel of coupled autonomo us oscilla-
tors regulating blood ¯ ow in the cardiova scular
system . Her research is devoted to extending and
deepenin g understan ding of the beauty and
function of this life giving system .
icÆreceived her D iploma Degree in
M aja BracÆ
electrical engineering in 1993 from the University
of Ljubljana, Slovenia , for work on the Lyapu-
nov exponents characterizing bloo d ¯ ow. As a
young researche r at the Faculty of Electrical
 Physics of the human cardiovascular system 55

Engineering at the University of Ljubljana, she
continue d her work in developing and applying
the techniques of linear and nonlinear dynamic s
to studie s of real system s. Am ong them, the
huma n cardiova scular system prove d to be one of
the most challenging. She perform ed extensive
analyses of its behaviour by analysin g signals
Downloaded by [California State University of Fresno] at 16:12 29 November 2012
related to its function which all illustrate that in
spite of being complex and nonlinear, it is not
complicated beyond understan ding. Her main
interest lies in the analysis of the coupling s within
the system , and she is currently preparin g her
PhD thesis on this topic.