Mathematical modeling and sensitivity analysis of

the cardiovascular arm arteries with anastomosis

R. Gul and S. Bernhard

Abstract
Cardiovascular disease is one of the major problems in todays medicine
and the number of patients increase worldwide. To find the correct treatment,
prior knowledge about function and dysfunction of the cardiovascular system
is required and method need to be developed that identify the disease in an
early stage.
Mathematical modeling is a powerful tool for prediction and investigation
of the cardiovascular system. It has been shown, that the Windkessel model,
drawing an analogy between electrical circuits and fluid flow, is a simple but
effective method to model the human cardiovascular system.
In this paper we have applied parametric local sensitivity analysis to a linear elastic model of the arm arteries including anastomosis to find and rank
sensitive parameters that may be helpful in clinical diagnosis. A computational model for anastomosis (anterior ulnar recurrent) is carried out to study
the effects of some clinically relevant haemodynamic parameters like blood
resistance and terminal resistance on pressure and flow at different locations
of the arm artery. In this context we also discuss the spatio-temporal dependency of local sensitivities. The percentage effect on the measurable state
variables pressure and flow, with respect to percentage change in cardiovascular input parameters, is quantified using norms. This novel methodology
allows us to verify the qualitative results obtained by sensitivity analysis.
The sensitivities with respect to flow resistance (R), arterial compliance
(C) and flow inertia (L), reveal that the flow resistance and diameter of the
vessels are most sensitive parameters. Those parameters play a key role in
diagnoses of severe stenosis and aneurysms. In contrast, wall thickness and
elastic modulus are found to be less sensitive. Results from time dependent
sensitivities show that R is sensitive at diastole and systole of the flow and
pressure wave respectively. While C and L are most sensitive at systole of
both flow and pressure waves respectively.
Sensitivities w.r.t. R and Z are obtained from inferior ulnar collateral anastomosis with anterior ulnar recurrent (IUC-AUR), which show that by increasing
R or equavalently by decreasing diameter of the anastomosis, the flow within
anastomosis decreases and the pressure difference yields back flow near endto-side anastomosis.
Keywords: computational cardiovascular model, cardiovascular parameters,
sensitivity analysis, anastomosis, Windkessel model.
Fachbereich

Mathematik, Freie Universit

at Berlin, Germany. gulrehman@ciit.net.pk
of Electrical Engineering and Information Technology, Pforzheim University of Applied
Sciences, Germany. stefan.bernhard@hs-pforzheim.de
Dept.

Introduction

With growing interest in the prediction and diagnosis of cardiovascular diseases,

different mathematical models have been developed and applied. Windkessel models
(electrical analogy to fluid flow) have shown to be an effective approach in modeling
the human cardiovascular system [18]. Westerhof, N. et al. [3] studied the design,
construction and evaluation of an electrical analog model. Quarteroni, A. et al. [1]
introduced a multiscale approach, where local and systemic models are coupled at
a mathematical and numerical level. He also introduced the Windkessel models for
different inlet and outlet conditions.
Within the Windkessel model the hemodynamic state variables (pressure (p)
and flow (q)), are interrelated to the model parameters like elastic modulus (E),
vessel length (l) and its diameter (d), wall thickness (h), the density of blood ()
and the network structure. Provided that the model parameters applied correspond
to a certain cardiovascular disease, the Windkessel model is a good way to study
variations of parameters, which are difficult to modify directly.
The basis for robust parameter estimation is on the one hand an optimal experimental measurement setup and on the other hand the development of models
that describe the hemodynamic state variables in a set of relevant parameters that
can be estimated with high accuracy. The design consists of several logical steps,
dealing with questions like:
Which vascular system parameters are most influential on the hemodynamic
state variables pressure and flow?
Which vascular system parameters are insignificant and may be fixed or eliminated?
Which region in flow and pressure waves are sensitive w.r.t. cardiovascular
parameters?
What are the parameters, variables and experimental measurement locations
within a clinical setting?
Sensitivity analysis is a powerful approach to find sensitive and therefore important
cardiovascular system parameters. The sensitive parameters can be further used to
design a measurement setup and to interpret measurements. in [9] for example, Sato,
T. et al. studied the effects of compliance, volume and resistance on cardiac output
using sensitivity analysis. Yih-Choung Yu et al. [10] used parameter sensitivity to
construct a simple cardiovascular model. Leguy, C.A.D et al. [15] applied global
sensitivity analysis on the arm arteries and showed that the elastic modulus is most
sensitive parameter, while arterial length is a less sensitive parameter.
In this paper we are interested in the effect of anastomosis on local sensitivities.
Anastomoses are the interconnection between vessels.
Beside providing a collateral circulation anastomoses also act as a second rout
of blood flow when main vessels are blocked by plaque, atherosclerosis or stenosis, which minimized the demage at tissue level. Another important concept while
studying anastomosis is valveless flow. William Harvey published a report explaining impedance defined flow , which explained a mechanism for valveless flow [11].
Later on, Weber [12] stated that heart is not able to pump blood alone but there
are other forces which help in circulation. There are several parameters in cardiovascular system which control the blood flow or simply create valveless flow like,
2

viscous and inertial effects and also elastic properties of two vessels [13]. Details of
the valveless flow mechanism are given in [14]. In this paper, we present a computational model of anastomosis around elbow joint (interior ulnar collateral anastomosis
with anterior ulnar recurrent, IUC-AUR). As luminal diameter or equavalently the
blood resistance play an impotant role in blood flow through anastomosis, so, here
we will study the effects of blood resistance and terminal impedance on pressure
and flow through anastomosis.
The methods developed in this paper, are seen as the first step towards cardiovascular system identification from cardiovascular measurements. Within this work we
derive a computational cardiovascular model for the arm arteries (with and without
anastomosis) by using the lumped parameter approach. In a first instance we apply
local sensitivity analysis (without anastomosis) to study the effects of cardiovascular
parameters on the hemodynamic state variables. We study time dependent sensitivities w.r.t. R, C and L to find sensitive regions in the pressure and flow waves.
Due to the network structure of the cardiovascular system it is helpful to determine
and discuss the location dependent sensitivities i.e which locations in arm arteries
are sensitive w.r.t. R, C and L. Finally to quantify and compare our results, we
apply the concept of norms.

Derivation of the model equations

Under the assumption that the arterial tree is decomposed into short arterial segments of length l with a constant circular cross-section and linear elastic wall behavior, the following one dimensional flow equations can be derived from the linearized
Navier-Stokes equation, the equation of continuity and the shell-equation for thin
walled, linear elastic tubes [1, 16]

p
q
= Rq + L ,
x
t

(1)

q
p
p
=
+C .
(2)
x
Z
t
Within these equations the state variables are the flow q and the relative pressure p.
The viscosity and inertial forces of the blood flow are described by the viscous flow
resistance R and blood inertia L per unit length respectively. The elastic properties
of the wall are modeled by a compliance C per unit length, while Z is terminal
resistance [16]. Integration of the two partial differential eqns. (1) and (2) along
flow axis leads to a system of equations (3) commonly used to describe electrical
circuits (see Figure ??). In this type of model each segment of the arterial system is
described by a set of two equations that are known as Windkessel equations. Here
(pin , qout) and (pin, pout) are the boundary conditions for non-terminal and terminal
nodes respectively. To model a mean venous pressure with a value of 15mmHg for
terminal nodes the equation system is setup by including an additional terminal
resistance Z. The matrix form of the Windkessel eqns. with boundary conditions
is
dX
= AX + B
(3)
dt
Where, X = {p1 q1, ..., pNs qNs } is a state vector, A is a state matrix and B contain
the boundary conditions. A complete description of state-space representaion is

Figure 1: Linear elastic model for fluid flow in nonterminal vessel segments (left)
and for terminal vessel segments (right).
given in section 2.3
For non-terminal segments (Figure1:left) X = (qin, pout)T
 R 1 
 pin 
L
L
L
A=
,
B(pin , qout) =
1
qout
0
C
C
For terminal segments (Figure1:right) X = (qin, qout)T

 R
 pin 
1
L
L
L
,
B(pin , pout) =
A=
1
1
out
pZC
ZC
C
The electrical parameters for ith segment in the arterial tree are related to the
physiological parameters of the fluid and vessel wall by:
Ri =

8l
,
r4

Li =

l
,
r2

Ci =

2r2l
.
Eh

(4)

where
E =
h =

Young modulus,
wall thickness,

l = length of vessel,
= blood viscosity,

r = radius of vessel
= blood density

The vascular network parameters required in A and B are given in Table 1:

2.1

Network structure and model equations

While modeling arm arteries (with and without anastomosis), we adopt domain
decomposition (DD) approach in which the domain of whole cardiovascular systen is
decomposed into number of subdomains of cardiovascular segments. Each segment
4

Nodes
units

E
kgm2 s2
105
4
4
4
4
4
4
8
8
8
8
8
16
8
8
8

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15

l
m
102
6.1
5.6
6.3
6.3
6.3
4.6
7.1
7.1
7.1
2.2
6.7
7.9
6.7
6.7
3.7

d
m
103
7.28
6.28
5.64
5.32
5
4.72
3.48
3.24
3
2.84
4.3
1.82
4.06
3.48
3.66

h
m
104
6.2
5.7
5.5
5.3
5.2
5
4.4
4.3
4.2
4.1
4.9
2.8
4.7
4.6
4.5

R
kgs1 m4
106
3.539
5.868
10.15
12.82
16.43
15.10
78.90
105
142.9
55.11
31.94
1173
40.19
50.22
33.60

C
m4 s2 kg1
1011
7.454
4.778
4.035
3.514
2.974
1.9
0.667
0.531
0.448
0.1207
1.067
0.0834
0.9366
0.80
0.3958

L
kgm4
106
1.539
1.898
2.648
2.976
3.369
2.76
7.838
9.042
10.55
3.647
4.844
31.88
5.434
6.075
3.693

Table 1: Numerical values of parameters for each node of the arm arteries(shown
in Fig. 2). The value of terminal resistance (Z) on three terminal nodes is 3.24
109 kgs1 m4 , = 0.004 kgs1 m1 and = 1050 kgm3 [4].
of the arm arteries in a network structure as given in Figure 2 is represented by an
electrical circuit as shown in Figure (1, top).
To built a cardiovascular network, relations between the arterial segments need to
be defined like at bifurcation flows diverge i.e. q1 = q2 +q3 and at anastomosis flows
merge i.e. q3 = q1 + q2. Pressure at bifurcation and at anastomosis remain same
(see Figure 1, bottom).
2.1.1

Model equations without anastomosis

In analogy to Kirchhoffs current and voltage law (Ohms law of hydrodynamics),

the arterial structure given in Figure 2, leads to the following system of coupled
ordinary differential equations for pressure and flow.
Flow equations:
q1

q4

q7

q10

q13

pin p1 R1 q1
p1 p2 R2q2
p2 p3 R3 q3
, q2 =
,
q3 =
L1
L2
L3
p3 p4 R4q4
p4 p5 R5q5
p5 p6 R6 q6
, q5 =
,
q6 =
L4
L5
L6
p6 p7 R7q7
p7 p8 R8q8
p8 p9 R9 q9
, q8 =
,
q9 =

L7
L8
L9
p9 p10 R10q10
p6 p11 R11q11
p11 p12 R12q12
, q11 =
, q12 =
L10
L11
L12
p11 p13 R13q13
p13 p14 R14q14
p14 p15 R15q15
, q14 =
, q15 =
L13
L14
L15
(5)

Pressure equations:

p1

p5

p9

2.1.2

q1 q2
q2 q3
q3 q4
q4 q5
, p2 =
,
p3 =
, p4 =
C1
C2
C3
C4
q5 q6
q6 q11 q7
q7 q8
q8 q9
, p6 =
, p7 =
, p8 =
C5
C6
C7
C8
q9 q10
q10 (p10 pout )/Z1
q11 q12 q13
, p10 =
, p11 =
C9
C10
C11

p12

p15

q12 (p12 pout )/Z2

q13 q14
, p13 =
,
C12
C13
q15 (p15 pout )/Z3
C15

p14 =

q14 q15
,
C14
(6)

Model equations with anastomosis

It was shown we have the AUR anastomosis in the model of arm artery (see Figure
2), the flow will split at nodes 3 and 6, and will merge at node 11. the inlet pressure
at node 11 is same as nodes 6 and 33. As we add three additional nodes so we have
now 5 new equations for pressure and flow. Additional equations for pressure and
flow of the ulnar recurrent artery are,
Anastomosis flow equations:
q3

q33

p2 p3 R3q3
p3 p31 R31 q31
, q31 =
L3
L31
p32 p6 R33 q33
L33

q32 =

p31 p32 R32 q32

L32
(7)

Anastomosis pressure equations:

p3 =

q3 q4 q31
,
C3

p31 =

q31 q32
q3 q33
q3 + q6 q11
, p32 = 2
, p33 = 3
(8)
C31
C32
C6 + C33

p33 and p6 are same according to the conservation of momentum.

2.2

Diverging and merging flows at junctions

Blood flow at junctions plays an important role in normal and pathological conditions of the cardiovascular system. In this section we briefly discuss pressure
and flow modeling at junctions for both diverging (at bifurcations) and merging
blood flows. In the arterial system merging flow conditions appear in the context
of anastomosis.
2.2.1

Diverging blood flow

In the above model structure diverging flows occur at the bifurcation of the vessels
at nodes 3, 6 and 11 (see Figure 2). Flows at junction mainly depend on angle and
radius of the daughter vessels. In this paper we limit our study to symmetric and
asymmetric bifurcations with respect to radius of the daughter vessels. For diverging flows the mass conservation equation at node 3 is, q3 = q31 + q4. For linearized
6

Figure 2: Simplified anatomy of the arm arteries (left) and model geometry of
brachial, anterior ulnar recurrent, ulnar and radial arteries, with number of nodes
Ns = 15 (without anastomosis), Nas = 18 (with anastomosis) and number of terminal nodes Nt = 3 (right).
system (eqns. 9, 10) the total pressure through bifurcation remain unchanged, i.e.
the output pressure at node 3 is the input pressure for both nodes 31 and 4.

2.2.2

Merging blood flow

To model the merging flows at junctions has a great importance to understand the
effect of anastomosis in bypass in the cardiovascular system. These type of flows
occur also in vascular grafting and arteriovenous fistula (AVF). According to the
conservation of mass, the flow at node 11 is, q33 + q6 = q11 and according to the law
of conservation of momentum, the total pressure remains continuous at nodes 11.

2.3

State-space representation

The state-space representation is a compact way to model a physical system as a set

of input, output and state variables related by first order differential equations [17].
In state-space form, we have a system of two equations: an equation for determining
state xt of the system (state equation), and another equation to describe the output
yt of the system (observation equation). The matrix form can be written as
x t
yt

= Axt1 + But ,
= Cxt + Dut .

(9)
(10)

Here xt is the state vector of the system, ut the input vector and yt the observation
vector. The dynamics of the system is described by the state dynamics matrix
A M (nn). The input matrix B M (ni) specifies the time dependency of the
in- and outflow boundary values and the observation matrix C M (m n) defines
the observation locations within the state-space system, i.e. the nodal location
in the network. Here m denotes the number of observations. Finally the input
to observation matrix D M (m i) adds the influence of the input vectors to
the observation vectors. Besides the computational advantage the state-space form
allows the integration of experimental measurements (observations) into the model
building process. This step is essential for the adjecent model parameter estimation
from experimental measurements, that are planned in a future study.
7

The state vector xt contains the flow and pressure functions at all network
locations, whereas the output vector yt contains the flow and pressure at selected
nodes i. For a m = 4 dimensional observation vector, the output vector is e.g.
y(t) = (q5(t), p5 (t), q6(t), p6 (t))T where y Rm at nodes 5 and 6. The state-space
system for the arm artery given in Figure 2, using eqn. (5) and eqn. (6) is defined
by

Aij =

Bij =

R
i+1

L
i+1

Ci

L i+1

12

L i+1

i = 1, 3, 5, ...29, j = i
i = 2, 4, 6, ...30, j = i 1
i = 1, 3, 5, ...29, j = i + 1
i = 3, 5, 7, ...29, j = i 1 and i 6= 21, 25

Ci

Zk C 2i

Li

Zj11 C i

Zj1 C i

Zj1 C 2i

also for i = 21, j = 12 and i = 25, j = 22

i = 2, 4, 6, ...30, j = i + 1 and i 6= 20, 24
also for i = 12, j = 21 and i = 22, j = 25
k = 1, 2, 3, i = 20, 24, 30, j = i
otherwise

i=j=1
i = 20, j = 2
i = 24, j = 3

(
1
, Cij =
0

i = 1, 2, 3, 4, j = i + 8
otherwise

i = 30, j = 4
otherwise

and Dij = 0.
For AUR anastomosis we expand our state-space model by using eqns. (7, 8). The
system is solved using the MATLAB built in solvers ODE 45 and lsim.

Methods of local sensitivity analysis

Cardiovascular parameters which have influence on output variables like, pressure

and flow are identified by sensitivity analysis. To understand the interdependence of
the state variables and the parameters of the cardiovascular model, it is imperative
to quantify the sensitivity of the state variables with respect to model parameters. In
this work we study the local sensitivities at different nodes in the vascular network.
Mathematically the sensitivity coefficient can be calculated as,

Sij =

yi
j

Where yi is i-th model output and j is the j-th parameter.

(11)

3.1

Sensitivity by finite difference method

In local sensitivity analysis, parameters are varied segmentwise by some portion

around a fixed value and the effects of individual perturbations on the observations
are studied [18]. Using differential calculus the sensitivity coefficients are
Si =

yi ( + ) yi ()
yi
= lim
,
0

(12)

where yi is the i-th model output, is the model input parameter and is the
change in model parameters. There are various methods to compute the sensitivity
coefficients in eqn. (12), within this work we use the method of finite difference:
Si =

yi
yi ( + ) yi ()
'

(13)

Equation (13) produces a set of two sensitivity time series Si (t) (one for pressure
and one for flow) per parameter and per network node (see figure 4).
Here we discuss three different scenarios of sensitivity analysis, (i) sensitivity analysis with respect to structural parameters ELdh, (ii) electrical parameters RCL
and (iii) location dependent sensitivities. In this context we also discuss the time
dependence of the sensitivities and compare the results of the two scenarios with
and without anastomosis
The sensitivity results are finally compared to the 2-norm of the distance vector of
the state variables of two time series.
3.1.1

Sensitivity with respect to E, l, d and h

To study the effects of Eldh on pressure and flow, we first solve our model with
both 10% variation of Eldh from its nominal values (see table 1). The sensitivities
are computed using eqn. (13). In this case is either Eld or h. The sensitivities for
pressure and flow w.r.t Eldh were calculated for node 7 (radial artery) and plotted
in figure 3.
3.1.2

Time dependent sensitivity

Sensitivity time series with respect to R, C and L, can be used to find sensitive
regions in pressure and flow waves at each location of the arm arteries. For this
we first solve eqn. (9) and eqn. (10) with Matlab built in solvers, then calculate
sensitivity time series for both pressure and flow at nodes 5 and 7 by using eqn.
(13). Results are presented in figure 4
3.1.3

Sensitivity with respect to R, C and L

To find the sensitivity of the electrical parameters R, C and L on cardiovascular

pressure and flow, we first solve eqn. (9) and eqn. (10) numerically using the
CVODES solver, which is a part of SUNDIALS software suit [19, 20] and then
calculate the sensitivities using eqn. (13). Results are summarized in figure 5
3.1.4

Network structure and sensitivity

To study the influence of the vascular network structure, we use a non-physiological

network structure with identical node parametrization, i.e. the parameters Ri, Ci
9

and Li are identical for each node. This allows us to analyze the influence of the
network structure onto sensitivity values at different node locations.
3.1.5

Mean sensitivity over time

For section 3.1.3 and section 3.1.4, we calculate mean of sensitivity time series for
all arm arteries nodes. Therefore we compute the nodal sensitivity time series Si for
all nodal parameters as described in previous section. For each node, n, we obtain
two sensitivity time series (one for the pressure and one for the flow) per parameter
variation. The total number of possible parameter variations is 3n2, so we end up
with a set of 6n2 time series. To reduce the complexity, we average out the time
dependency by computing the mean of the absolute value of every time series. The
6n2 real values are used for further analysis in section 4. They are displayed into 6
matrices that characterize the sensitivity of pressure and flow (sensitivity nodes) in
the network structure based on changes in the electrical parameters RCL (selected
nodes). Each cell in Figure 5 and 6, represents the mean absolute value of time series
(Si ). Due to the fact that we use the pressure as an input and output boundary
condition, the change in pressure with any parameter at all inlet and terminal nodes
will be zero.
3.1.6

Sensitivity within anastomosis

Blood flow in AUR anastomosis (collateral circulation) depends on size and mainly
on diameter of anastomosis, smaller the diameter lesser the flow in anastomosis and
vice versa. In this paper we limit our study to end-to-side AUR anastomosis and
show the sensitivity of anastomosis structure on pressure and flow by changing blood
resistance Ra and Rb and terminal resistance (Z) [1114]. Here we consider Ra =
R31 + R32 + R33 and Rb = R4 + R5 + R6, where Ra and Rb are the total resistance
through AUR anastomosis and through its parallel brachial artery respectively.
Further more we discuss 5 different scenarios (a) when Ra ' Rb, (b) Ra  Rb, (c)
Ra  Rb, (d) Z is small and (e) Z is large. The sensitivity is computed as earlier
stated.

3.2

Sensitivities by using norms

To obtain a measure to validate the sensitivities, we calculated the mean Euclidean

distances of the observations made in a model with different parameter sets 1 , 2 .
Here 1 is the nominal parameter set and 2 is 10% change in 1 .
k1 , 2 k:= mean
tT

k yi (2 , t) yi (1 , t) k2
kyi (1 , t)k2

i = 1, 2, 3...2 Ns ,

The results are compared to sensitivity measures in table 2 and table 3.

Results and discussion

In this section we discuss the results obtained from section 3.

10

E sensitivity on flow at node 7

x 10

4.5
4

pressure [Pa]

flow [ml/s]

2.5

Phase and
amplitude shift

1.5
1
0

0.1

0.2

x 10

0.3

0.4
time [s]

0.5

Phase and
amplitude shift

0.9

0.6

0.7

0.8
0

0.8

0.2

1
0

0.1

pressure [Pa]
0.2

0.3

0.4
time [s]

0.5

0.6

0.7

0.8
0

0.8

pressure [Pa]

3
2.5

0.1

1.4

A7

pressure [Pa]

1.3

3
2.5

Norm sensitivity=0.58

0.2

0.3

Phase and
amplitude shift

1.5
0.1

0.2

0.3

0.4
time [s]

0.5

0.5

10%
+10%
Nominal values

0.6

0.7

0.8

Norm sensitivity=0.13

A8

1.2
1.1
Phase and
amplitude shift

0.4
time [s]

h sensitivity on pressure at node 7

x 10

Norm sensitivity=0.36

3.5

1
0

d sensitivity on pressure at node 7

1.5
10%
+10%
Nominal values

0.8

0.8
0

0.8

h sensitivity on flow at node 7

x 10

0.7

0.7

1.1
1

0.6

0.6

0.9
0.5

0.5

1.2

1.5
0.4
time [s]

0.4
time [s]

1.3

0.3

0.3

10%
+10%
Nominal values

1.4

0.2

0.2

x 10

1.5

4.5

0.1

Norm sensitivity=1.34

0.1

Phase and
amplitude shift

0.9

3.5

1
0

0.8

1.1

1.6
10%
+10%
Nominal values

0.7

1.2

d sensitivity on flow at node 7

x 10

4.5

0.6

Norm sensitivity=0.32

1
Phase and
amplitude shift

1.5

0.5

1.3

0.4
time [s]

10%
+10%
Nominal values

1.4

2.5

0.3

l sensitivity on pressure at node 7

x 10

Norm sensitivity=0.95

A3

0.1

1.5
10%
+10%
Nominal values

3.5

flow [ml/s]

1.1

l sensitivity on flow at node 7

flow [ml/s]

1.2

flow [ml/s]

Norm sensitivity=0.13

1.3

10%
+10%
Nominal values

1.4

x 10

Norm sensitivity=0.36

3.5

E sensitivity on pressure at node 7

1.5
10%
+10%
Nominal values

0.9

0.6

0.7

0.8
0

0.8

0.1

0.2

0.3

0.4
time [s]

0.5

0.6

0.7

0.8

Figure 3: Pressure and flow for a 10% change in E, l, d and h in the radial artery
at node 7. The results reveal that d and l are sensitive parameters while E and h
are less sensitive, for the reason of linear appearence E and h have same effects on
cardiovascular pressure and flow.

4.1

Sensitivities with respect to E, l, d and h

The sensitivity for pressure and flow are obtained by a variation of the cardiovascular
parameters E, l, d and h of arm arteries by 10%. From figure 3 it is directly evident,
that diameter and length are most sensitive parameters, while the elastic modulus
and wall thickness are comparatively less sensitive parameters.

11

4.2

Time dependent sensitivity

Time dependent sensitivity is very important specially when we want to estimate

cardiovascular parameter from pressure and flow waves. It further guide us to find
optimal positions or regions (systole and diastole), from where we can get more
information for parameters. In this section we study +10% change in R, C and L
in all locations. Time series w.r.t R, C and L were calculated at nodes 5 and 7, by
using eqn. (13). The results are summarized as,
(a). R sensitivity time series:
According to the hydrodynamics form of Ohms law, the flow resistance R can be
calculated from the ratio of pressure gradient and flow, i.e. R = p
q . It is evident
from the figure 4 (B1 , B2 )) that blood resistance R has an inverse relationship to
blood flow q and has linear relationship with the pressure gradient. When arteries
constrict the resistance increases and when arteries dilate the resistance will decreases. Due to this fact resistance is most sensitive in the diastolic region of the
flow wave, while it is less sensitive in the the systolic region. On the other hand
resistance is sensitive at diastolic region of pressure waves, which shows clear agreement with Ohms law of hydrodynamics.
(b). C sensitivity time series:
Compliance C, is the change in arterial blood volume V , due to the change in
arterial blood pressure p, i.e. C = V
p . It is thus the slope of the pressure-volume
curve, so it depends on the pressure level at which the compliance or elastance is
calculated. The results show, that the compliance has a larger effect on diastolic
parts of both pressure and flow waves (see Fig. 4 (C1 , C2)).
(c). L sensitivity time series:
Blood inertia L relates pressure drop with flow rate i.e. L = p
q . Blood inertia
plays a role in acceleration (in systole) and deceleration (in diastole) the blood flow
in cardiovascular vessels. From Figure 4 (D1 , D2) it is clearly shown that L is most
sensitive in diastolic parts of pressure and flow waves as compared to systolic parts.

4.3

Sensitivities with respect to R, C and L

The sensitivities w.r.t. R, C and L were calculated by finite difference approach

eqn. (13), using the SUNDIALS software. The sensitivity pattern obtained by
variation of the viscous flow resistance R in any segment of brachial artery (see
Figure 5 (top)) indicates a strong local (within brachial artery) influence on flow
and has significant global influence on all following nodes of the brachial, ulnar and
radial arteries. In contrast, changing R in the parallel association of the ulnar and
radial arteries have negligible local and global effects, because in parallel arteries
1
1
1
the total flow resistance is given by the fraction Rtotal
= Rulnar
+ Rradial
, i.e. due
to the increment in total diameter the flow resistance reduces. Physically a change
of R in one branch redirects the flow into the other branch while the overall flow
is maintained. The sensitivity of flow resistance in parallel branches is thus smaller
than in series connections.
In contrast to the flow resistance, the sensitivity of arterial compliance C in
the brachial part has small influence (locally and globally) on pressure and flow
(see Figure 5 (middle)). This can be explained by the fact, that in series segments
the totalP
compliance is given by the sum of all segmental compliances in series
Ctotal = 6i=1 Ci. The total compliance is larger than the individual compliances,

12

15

x 10

R sensitivity on flow at nodes 5 and 7

0.5

R sensitivity on pressure at nodes 5 and 7

Node 5
Node 7

B1

x 10

0.5

1.5

Node 5
Node 7

B2

R sensitivity

R Sensitivity
2
0

100
5

1.5

x 10

200

300

400
500
Time [ms]

600

700

100
13

C sensitivity on flow at nodes 5 and 7

Node 5
Node 7

C1

2
0

800

300

400
500
Time [ms]

600

700

800

C sensitivity on pressure at nodes 5 and 7

x 10

Node 5
Node 7

C2

0.5

200

0.5
2
1
C sensitivity

1.5
2
0

100
14

x 10

1.5

200

300

400
500
Time [ms]

C sensitivity

600

700

6
0

800

Node 5
Node 7

D1

100

x 10

300

400
500
Time [ms]

600

700

800

Node 5
Node 7

200

L sensitivity on pressure at nodes 5 and 7

L sensitivity on flow at nodes 5 and 7

L sensitivity

0.5
0

0
2

0.5
L sensitivity

1
1.5
0

100

200

300

400
500
Time [ms]

600

700

4
6
0

800

100

200

300

400
500
Time [ms]

600

700

800

Figure 4: Flow and pressure waves (A1 , A2) and sensitivity time series with respect
to R, C and L (B1 , B2 , C1, C2, D1 , D2) at nodes 5 and 7. Results shows that R is
sensitive at diastolic and systolic part of flow and pressure wave respectively. While
C and L are most sensitive at systolic part as compared to the diastolic part of
pressure and flow waves at respective nodes.
thus a change of C in any node has a small effect on pressure and flow in the whole
arm artery. In contrast a variation of the arterial compliance C in the ulnar and
radial arteries have a large (local) effect on pressure and large (global) effect on flow
especially in the brachial artery.
From eqn. (4) it is obvious that viscous resistance and blood inertia are inversely
related to r4 and r2 respectively. Which means in large arteries blood inertia plays
an important role, while in small arteries viscous resistance is more important. A
variation of blood inertia in the first node of the brachial artery has large influence on
pressure and flow of all following nodes of the arm arteries (see Figure 5 (bottom)).
However, we observe only minor local (within brachial artery) effects on flow when
we change L in each segment of the brachial artery. Further, due to the fact that
1
1
1
the total inductance Ltotal
= Lulnar
+ Lradial
reduces at the furcation, the flow and
pressure in the ulnar and radial arteries are less sensitive with respect to L.
13

R mean absolute sensitivity for flow

2

R mean absolute sensitivity for pressure

14

x 10
6

Selected nodes

Selected nodes

Brachial artery
4

10
2

10

12

10

Radial artery

12

10

8
10

C sensitivities of flow
on brachial artery

12

14
2

6
8
10
Sensitivity nodes

12

2
14

7
6

10

12

2
1
2

6
8
10
Sensitivity nodes

12

14

L mean absolute sensitivity for pressure

12

x 10

Selected nodes

Brachial artery

1.5

8
1

10
12

13

x 10

14

Selected nodes

12

14

L mean absolute sensitivity for flow

6
8
10
Sensitivity nodes

C mean absolute sensitivity for pressure

x 10
14

Selected nodes

Selected nodes

14

Sensitivity nodes

C mean absolute sensitivity for flow

10

14

14
12

Ulnar artery

14

16
Brachial artery

12

12

x 10

Brachial artery

x 10
5
4

6
8

Radial artery

10

12

0.5

Ulnar artery

14

14
2

6
8
10
Sensitivity nodes

12

14

6
8
10
Sensitivity nodes

12

14

Figure 5: Effects of viscous flow resistance R (top), vessel compliance C (middle)

and blood inertia L (bottom) on pressure and flow in the arm arteries. From
the results changing flow resistance, R and blood inertia, L in brachial artery has
strong local and global effects on flow and pressure waves respectivily. While flow
in brachial artery is sensitive w.r.t. changing vessel compliance, C in all nodes of
the arm aretry.

4.4

Network structure and sensitivity

In order to know important locations in arm arteries (without anastomosis), we

consider identical node parameter values in whole arm artery. Results show that R
is sensitive for flow in brachial artery while other locations in radial and ulnar are
less sensitive. On the other hand changing R in brachial artery has influence on
pressure for all following nodes. C and L sensitivities also have some local impact
in brachial artery, while distal nodes are less sensitive (see Figure 6). Location
dependent sensotivity along with computational model of anastomosis, indicates a
feasible location for creating anastomosis (vascular grafting) and AVF, e.g. one can
create anastomosis from the sensitive parts of the arm arteries.

14

R mean absolute sensitivity for flow

R mean absolute sensitivity for pressure

13

x 10
6

2
Brachial artery

10

Selected nodes

Selected nodes

12

6
8
10
Sensitivity nodes

12

Radial artery

10
2
Ulnar artery

2
x 10

12

14

Selected nodes

1.5
1

12

4
6

8
2

10
12

0.5

14
2

6
8
10
Sensitivity nodes

12

L mean absolute sensitivity for flow

Selected nodes

8
6
10
4

12

6
8
10
Sensitivity nodes

12

12

14
4

x 10

15

4
6

10

8
10

12

14

6
8
10
Sensitivity nodes

10

L mean absolute sensitivity for pressure

12

x 10
12

14

14

13

x 10
4

10

6
8
10
Sensitivity nodes

2.5

C mean absolute sensitivity for pressure

4
6

14

14

C mean absolute sensitivity for flow

Selected nodes

12
1

14

Selected nodes

Brachial artery

x 10

14

14

6
8
10
Sensitivity nodes

12

14

Figure 6: Effects of viscous flow resistance R (top), vessel compliance C (middle)

and blood inertia L (bottom) on pressure and flow in the arm artery, with identical
nodes. From the results R is sensitive parameter for flow in brachial artery, while
changing R in brachial have both local and global effect on pressure. C and L also
have some local sensitivities in brachial part of arm artery.

4.5

Sensitivity within anastomosis

In this section we consider 5 different scenarios as,

(a). Ra ' Rb:
To study the blood flow through AUR anastomosis, we first consider identical values
of RCL of nodes 31 , 32 , 33 to nodes 4, 5 and 6 respectively. As it is mentioned earlier, the diameter of vessels plays an important role in pressure and flow distribution
of cardiovascular system, so we limit our study in changing the blood viscosity in
AUR anastomosis and its parallel brachial artery, i.e. Ra ' Rb, where the values
of Ra and Rb are given in section (3.1.6). the Results show that pressure and flow
distribution in AUR and its parallel brachial artery are the same (see Figure 7 A,
B).
(b). Ra  Rb:
Here we consider small diameter of AUR anastomosis as compare to its counterpart
b
brachial artery (ideal case). The values of Ra = R
50 . Due to decrease in diameter
or equivalently, decrease in total blood resistance through AUR anastomosis, the

15

flow decreases. Results show some back flow at node 33 relate to the pressure drop
resulting from the increased blood resistance through the AUR anastomosis (see
Figure 7 C, D).
(c). Ra  Rb:
A similar behavior of pressure and flow distribution can be seen by decreasing the
diameter of nodes (4 , 5 and 6) of brachial artery. Figure (7 E, F) shows back flow
at node 6 due to the pressure drop by setting Rb = 50Ra.
(d). Z is small:
Physical activities lead to a reduction in terminal resistance (Z), which increases
the blood flow and mean cardiovascular pressure. As a result the cardiac output
will increases. These are the temporary changes which appear only when we do
some physical exercise. Another artificial reason for low terminal resistance is the
implantation of arteriovenous fistula (AVF), which is abnormal connection between
a peripheral artery and vein. The decrement in terminal resistance increases blood
pressure and flow and finally increases cardiac output. In this study we consider
Z
small terminal resistance, Zs = 10
(see Figure 7 G, H).
(e). Z is large:
In this section we consider large terminal resistance (Z) at terminal node 15 of
the ulnar artery i.e. ZL = 10Z, which appears when we have no physical activity.
Figure (7 I, J) shows that if the person is at rest, we have large value of Z, then
flow will reduce (Ohms law of hydrodynamics). Results are obtained at different
locations of the arm artery, which reveal that by increasing Z the flow will reduce
and pressure will increase near the terminal.

4.6

Sensitivities by using norms

Finally we compare the results obtained by sensitivity analysis with those obtained
by using norms. We found that the diameter and length of vessel are most influential
parameters and that the norm computed for the wall thickness and elastic modulus
has identical values (see table 2 and 3).

16

0.8

1.5

Node 5
Node 6
Node 11
Node 32

1
Flow [ml/s]

Flow at selected nodes with R =R

x 10

B1

Node 33

0.6
0.4
Merging flows
0.2
100
6

200

300

400
500
Time [ms]

600

700

1.3
1.2

Node 3

2
3

1.1
1

0.8
0

800

Flow at selected nodes with R >>R

a

x 10

1.5

B3

x 10

1.3

200
300
400
500
600
700
Time [ms]
Pressure at selected nodes with Ra>>Rb

Node
Node
Node
Node

1.4

Node 33

10

100
4

Pressure [Pa]

Node 5
Node 6
Node 11
Node 32

15

Flow [ml/s]

Node 5
Node 6
Node 11
Node 3

0.9

0
0

20

Pressure at selected nodes with R =R

x 10

1.4
Pressure [Pa]

1.2

5
6
11
3

800

B4

Node 33

1.2
1.1
1

0.9

Backflow
100

20

x 10

300

400
500
Time [ms]

600

700

0.8
0

800

Flow at selected nodes with Ra<<Rb

Node
Node
Node
Node

15

Flow [ml/s]

200

5
6
11
32

1.5

B5

x 10

300

400
500
Time [ms]

600

700

800

B6

1.3

200

Pressure at selected nodes with Ra<<Rb

Node 5
Node 6
Node 11
Node 3

1.4

Node 33

10

100
4

Pressure [Pa]

5
0

Node 3

1.2
1.1
1

0.9
Backflow

5
0

100

200

300

400
500
Time [ms]

600

700

0.8
0

800

Flow with R >>R and small Z at node 15

x 10
a
b
5

Node
Node
Node
Node
Node

Flow [ml/s]

1.5

6
11
14
15
3

1.4

x 10

7
1.3

1
0.5
0

1.2

200

300

400
500
Time [ms]

600

700

800

Pressure with R >>R and small Z at Node 15

a

Node
Node
Node
Node
Node

Pressure [Pa]

2.5

100
4

6
11
14
15
3

1.1
1
0.9

Backflow
100
6

20

x 10

Flow [ml/s]

15

200

300

400
500
Time [ms]

600

700

0.8
0

800

Flow with R >>R and large Z at Node 15

a

Node
Node
Node
Node
Node

6
11
14
15
3

100
4

1.5

x 10

1.4
1.3

Pressure [Pa]

0.5
0

10

200

300

400
500
Time [ms]

600

700

800

Pressure with Ra>>Rb and large Z at Node 15

Node
Node
Node
Node
Node

6
11
14
15
3

B10

1.2
1.1
1

0
0.9

Backflow
5
0

100

200

300

400
500
Time [ms]

600

700

0.8
0

800

100

200

300

400
500
Time [ms]

600

700

800

Figure 7: Effects of viscous flow resistance R and terminal resistance Z on pressure

and flow at different locations of the arm arteries (with anastomosis). From the
results, it is seen that increasing flow resistance decreases the flow and produces
some back flow and flow will increase by decreasing terminal resistance.

17

E
l
d
h

N0
q0
0.45
0.54
1.96
0.45

N1
q1
0.44
0.53
1.76
0.44

N2
q2
0.42
0.47
1.75
0.42

N3
q3
0.36
0.43
1.55
0.36

N4
q4
0.34
0.39
1.43
0.34

N5
q5
0.27
0.32
1.16
0.27

N6
q6
0.24
0.31
1.06
0.24

N7
q7
0.36
0.95
1.34
0.36

N8
q8
0.39
0.86
1.19
0.39

N9
q9
0.19
0.48
0.80
0.19

N10
q10
0.18
0.46
0.78
0.18

N11
q11
0.18
0.25
0.87
0.18

N12
q12
0.11
0.13
0.45
0.11

N13
q13
0.24
0.34
0.10
0.24

N14
q14
0.14
0.19
0.63
0.14

N15
q15
0.12
0.15
0.54
0.12

Table 2: 10% change in E, l, d and h at node 7 of the arm artery (see Figure 2, without anastomosis) and their corresponding
percentage change in flow at each node.

18

E
l
d
h

N0
p0
0
0
0
0

N1
p1
0.02
0.03
0.13
0.02

N2
p2
0.03
0.05
0.15
0.03

N3
p3
0.05
0.06
0.21
0.05

N4
p4
0.06
0.08
0.27
0.06

N5
p5
0.08
0.09
0.33
0.08

N6
p6
0.08
0.10
0.36
0.08

N7
p7
0.13
0.32
0.58
0.13

N8
p8
0.13
0.32
0.58
0.13

N9
p9
0.14
0.37
0.63
0.14

N10
p10
0
0
0
0

N11
p11
0.09
0.11
0.38
0.09

N12
p12
0
0
0
0

N13
p13
0.09
0.11
0.40
0.09

N14
p14
0.10
0.12
0.44
0.10

N15
p15
0
0
0
0

Table 3: 10% change in E, l, d and h at node 7 of the arm artery (see Figure 2, without anastomosis ) and their corresponding
percentage change in pressure at each node. Change in pressure with respect to any parameter is zero, because we use pressure as a
boundary condition.

Conclusion

In this work we have applied different methods of sensitivity analysis to a lumped

parameter Windkessel model of the arm arteries. The results indicate a strong dependence of the pressure and flow state variables onto a variation in vessel diameter
and length. According to the elastic properties and the thickness of the arterial
wall, a much lower sensitivity was found. Alternatively we can say that blood viscosity and compliance of vessel are sensitive parameters while blood inertia is less
sensitive parameter.
Results from time dependent sensitivity show that R is sensitive at diastole and
systole of the flow and pressure waves, also C and L are sensitive at systolic part of
the flow and pressure waves, While in location dependent sensitivities, R is sensitive
for flow in the brachial artery and changing R in brachial artery have both local
and global influences on the pressure.
Further we have presented a computational mathematical model for anastomosis
which can predict pressure and flow through anastomosis. Results reveal that by
increasing flow resistance or equavalently decreasing diameter of the anastomosis
will decrease the flow and produces some back flow near end-to-side anastomosis,
while flow near terminal nodes will increase by decreasing terminal resistance and
vice versa.
Finally, we have used the concept of norms to quantify the results and to compare
the variation in state variables according to parameter changes. We found a good
agreement to the results obtained by sensitivity analysis.

Future work

The methods applied, give satisfactory results if the cardiovascular parameters are
independent, in the real scenarios however, they are often interdependent like e.g.
the observation of a high correlation between the extension of the elastic walls and
the tangential tension caused by transmural pressure. To study these type of effects
in a more general way, we plan to apply global sensitivity analysis to a closed loop
and open loop cardiovascular system models, which deal with variations in many
parameters at a time.

Acknowledgments

The author would like to thank the HEC/DAAD for their financial support also to
Rudolf Huttary for fruitful discussions and encouragement.

References
[1] Quarteroni, A., Ragni, S. and Veneziani, A., Coupling between lumped and
distributed models for blood flow problems. Computing and Visualization in
Science. Volume 4, pages 111-124 (2001)
[2] Milisic, V. and Quarteroni, A., Analysis of lumped parameter models for blood
flow simulations and their relation with 1D models. Mathematical Modeling
and Numerical Analysis (1999)
19

[3] Westerhof, N., Bosman, F., De Vries, C.J. and Noordergraaf, A., Analog studies
of the human systemic arterial tree. Journal of Biomechanics, Volume 2, pages
121-143 (1969)
[4] Noordergraaf, A., Verdouw, P.D. and Boom, H.B.K., The use of an analog
computer in a circulation model. Progress in Cardiovascular Diseases. Volume
5, number 5 (1963)
[5] Phillips, C., A simple lumped parameter model of the cardiovascular system.
PhD thesis, Colorado State University Fort Collins, Colorado (2011)
[6] Shim, E. B., Sah, J. Y. and Youn, C. H., Mathematical modeling of cardiovascular system dynamics using a lumped parameter method. Japanese Journal
of Physiology. Volume 54, pages 545-553 (2004)
[7] Yobing, S., Lumped parameter modeling of cardiovascular system dynamics under different healthy and diseased conditions. PhD thesis, University of Sheffield
(2013)
[8] Sherwin, S. J., Franke, V., Peiro, V., and Parker, K., One dimentional modeling
of a vascular network in space time variables. Kluwer Academic Publishers.
Printed in the Netherlands. (2003)
[9] Sato, T., Yamashiro, S.M., Vega, D. and Grodins, F.S., Parameter sensitivity analysis of a network model of systemic circulatory mechanics. Annals of
Biomedical Engineering. Volume 2, pages 289-306 (1974)
[10] Yu, Y.C., Boston, J.R., Simaan, M.A. and Antaki, J.F., Sensitivity analysis of
cardiovascular models for minimally invasive estimation of systemic vascular
parameters. Proceedings of American Control Conference, San Diago, California, (1999)
[11] Harvey, W., Exercitatio anatomica de motu cordis et sanguinis in animalibus
, Frankford, chapter 14 (1628)
[12] Weber, EH., De pulsu, resorptione, auditu et tactu. Annotationes Anatomicae
et Physiologicae, Lipsiae, (1834)
[13] Liebau, G., Die bedeutung der tragheitskrafte fur die dynamik des blutkreislaufs. Zs kreislaufforschung. Volume 4, pages 428-438 (1957)
[14] Maximilian, M., Johnnie, W. H., Graham, S. S., Thomas, K., Noordergraaf,
A.,Impedance defined flow: Generalization of William Harveys concept of the
circulation-370 year later. International Journal of Cardiovascular Medicine
and science, Volume 1, pages 205-2011 (1998)
[15] Laguy, C.A.D., Bosboom, E.M.H., Belloum, A.S.Z., Hoeks, A.P.G. and van de
Vosse, F.N., Global sensitivity analysis of a wave propagation model for arm
arteries. Medical Engineering and Physics. Volume 33, pages 1008-1016 (2011)
[16] Jager, Gerad S., Westerhof, N., Noordergraf, A., Oscillatory flow impedance
in electrical analog of arterial system: representation of sleeve effect and nonnewtonian properties of blood. Circulation research. Volume XVI, pages 121133 (1965)
20

[17] Bernhard, S., Al Zoukra, K. and Sch

utte, C., Statistical parameter estimation
and signal classification in cardiovascular diagnosis. Environmental Health and
Biomedicine. Volume 15, pages 458-469 (2011)
[18] Zi, Z., Sensitivity analysis approaches applied to systems biology models. IET
system biology. Volume 5, issue 6, pages 336-346 (2011)
[19] Hindmarsh, A.C., Brown, P.N., Grant, K.E. et al. SUNDIALS: suite of nonlinear and differential/algebraic equation solvers, ACM Transactions on Mathematical Software. Volume 31, Number 3, pages 363-396 (2005)
[20] Serban, R., Hindmarsh, A.C., CVODES: the sensitivity-enabled ODE solver in
SUNDIALS. Proceedings of IDETC/CIE (2005)

21