D EPARTMENT OF M ATHEMATICS
D OCTORAL PROGRAM IN
M ATHEMATICAL M ODELS AND M ETHODS IN E NGINEERING
Supervisor:
Prof. Paola Causin
Co-advisor:
Prof. Riccardo Sacco
Tutor:
Prof. Roberto Lucchetti
Chair of the Doctoral Program:
Prof. Irene Sabadini
In this thesis we propose mathematical models for blood and solute transport in large
microcirculatory networks which try to overcome this gap.
The thesis is divided in two Parts. In Part I, we present the assumptions and the mathe-
matical techniques to reduce the 3D balance equations to a 1D framework. We present
the numerical solution techniques and the simulations results, using coherent physi-
ological data. In the model, the vessel network is represented as graph of distensi-
ble/rigid vessels embedded in a metabolically active tissue. Blood is described as a
mixture of two components, plasma and red blood cells. Flow of blood in each ves-
sel is described by a generalized Poiseuille’s law featuring a conductivity parameter
function of the area and shape of the tube cross section. Structural models of the ves-
I
sel wall, both phenomenological and mechanically-founded (thick-walled ring model
for arterioles or thin-walled ring model including a buckling model for venules), are
used to compute the geometry of the deformed tube section under external and internal
stimuli (transmural pressure, solute concentration, vasodilator/constrictor molecules).
In addition, diffusion-transport-reaction equations are employed to study the transport
and delivery of solutes (in particular, oxygen) by blood along the network, taking into
account the leakage of solute through the vessel walls into the surrounding tissue.
In Part II, we focus on a specific microcirculatory district, the eye retina microcir-
culation. The aim of this part is to employ the previous described mathematical models
to assess the role of altered physiological parameters in the progression of several eye
pathologies, and specifically glaucoma.
The resulting geometrical model is adapted to the 3D retina anatomy. The model is
validated against experimental data on humans and is used to carry out sensitivity anal-
yses, both via numerical and semi-analytical techniques. On the same line, a further
investigation is carried out to address drug delivery in the posterior segments of the
eye, a delicate and difficult target even for modern pharmacology.
II
Contents
Introduction 1
III
Contents
IV
Contents
Bibliography 169
V
Introduction
The thesis is divided into two main parts. The first part is devoted to the mathe-
matical modelling of large microcirculatory systems described as graph networks of
distensible tubes. Numerical solution techniques and simulation results are presented
at the end of each chapter. The simulations carried out using the proposed models show
interesting local features of the network response, which are pretty different from what
1
Contents
should be expected on the basis of experiments on isolated vessels (see, e.g, [79, 118])
and of theoretical studies with lumped models (see, e.g, [10,80,189,217]). The follow-
ing topics are addressed:
• in the first chapter we give a brief overview on the microvascular structure and
function
• in the second chapter we describe the geometrical modelization of the circulatory
network. In particular, we consider networks obtained in silico from fractal di-
chotomic relations, where the vessel diameter and length at each bifurcation are
determined by the following fractal relationships [199]:
Dfn = Ddm1 + Ddm2 , L = 7.4D1.15 .
We also consider a completely unstructured network generated by diffusion lim-
ited processes (DLA) [214];
• in the third chapter, first we discuss a mathematical model for blood transport in
a network. Blood is represented as a mixture of two Newtonian incompressible
fluids, plasma and red blood cells. The pressure, velocity and hematocrit fields of
blood in the arteriolar/venular branches are described by 1D equations, obtained
upon averaging the 3D Navier Stokes equations
∂vs 1 ∂(rvr )
+ = 0,
∂s r ∂r
∂(vs HD ) 1 ∂(rvr HD )
+ = 0,
∂s r ∂r
∂ 2 vs 1 ∂vs ∂pb
µb ( 2 + )− = 0.
∂r r ∂r ∂s
Second, we discuss a mathematical model for transport and delivery of solute,
and in particular oxygen, by blood flow in the network. The oxygen content in the
arteriolar/venular vessels is described by the following steady state 1D equations
∂J s
A + 2παc Lp pc fc (R) =
∂s
∂J w
−A + 2παc Lp βt pt,w (rfc (r)) ,
∂s r=R
2
Contents
discussed in the second chapter. However, in this case, the blood flow results to
be dependent on a conductivity parameter, function of the area and shape of the
tube cross section and the biophysical properties of blood itself, defined as
b4 Z
R
σ= v ∗ dA∗ .
µb A ∗ s
Both empirical (experimentally-derived) and theoretical models are used to de-
scribe the relationship between the cross section area and the pressure loads.
When dealing with thin–walled venules, the possibility of section buckling is also
considered. A mathematical model to recover the unloaded configuration is also
discussed;
• in the fifth chapter a simple model of autoregulation mechanisms is proposed on
a single vessel, in which the vessel diameter is regulated by the concentration of
oxygen and nitric oxide in the vessel wall. Nitric oxide is assumed to influence the
level of calcium ions in the smooth muscle cells which, in turn, give rise to muscle
contraction and thus diameter variation. A radial solution for solute transport
along the vessel wall, described as a three–layered structure (endothelium, smooth
muscle cells layer and tissue) is obtained using Bessel–type functions. The radial
displacement of the wall is governed by the law:
∂ 2η ∂η h0 E(F ) h0 k2
ρw h0 R0 2
+ λh0 + 2
[η + κ1 η 2 ] = P0 − F;
∂t ∂t R0 (1 − ν ) R0
In the second part of the thesis, we consider a specific microcirculatory district, the
eye retina circulation and we apply the mathematical and numerical models, developed
in the first part, to its study. The following topics are addressed:
• in the sixth chapter we give a brief overview of the anatomical feature of the eye
retina;
• in the seventh chapter we couple the models discussed in the third chapter for
the blood flow and oxygen transport along the network with a model for the oxy-
gen transport in the retinal tissue. The model accounts for the three-dimensional
anatomical structure of the eye retina shown in Fig.1
The model is validated against available experimental results to identify baseline
conditions. Then, a sensitivity analysis is performed to assess the relevance of
the variation of different parameters of clinical interest on a selected number of
significant output variables;
• in the eighth chapter we focus our attention on the retinal tissue. We solve the
following diffusion-reaction equation for solute transport and consumption with a
Michaelis Menten term
∂ 2 pt
−Dt αt 2 = Q(pt , pc ),
∂z
by a semi-analytical technique which makes use of the so–called homotopy per-
turbation method [188]. We perform a sensitivity analysis to theoretically weight
out the relative contributions of various factors to the oxygen profile in the retinal
tissue, in particular oxygen source terms and consumption rates;
3
Contents
• eventually, in the ninth chapter, we study transport of solutes with different char-
acteristics (molecular weight, lipophilicity, . . . ) along the complete posterior part
of the eye, including sclera, choroid, retina and vitreous. This study is devoted to
develop a pharmacokinetical mathematical model to assess drug levels subsequent
to a trans–scleral drug implant. The domain is partitioned into different regions,
each of them with characteristic physiological structure and function.
The membranes limiting the different regions, which have fundamental impor-
tance since they represent physical obstacles to drug diffusion, are described by
appropriate partition laws including permeability parameters and active transport
effects. The filtration velocity is described according to the steady-state Darcy
equation while the drug transport by a diffusion-reaction equation. In particular,
the oxygen transport equation in the retina, described by a mixture of tissue and
blood vessel phases, is:
∂ 2 CRt
∂CRt ∂CRt
(1 − φ) = (1 − φ) DR − (vRt + vact ) − kRt CRt
∂t ∂x2 ∂x
+β(CRb − CRt ),
dCRb
φ = SRb (t, x) − β(CRb − CRt ) − φkRb CRb .
dt
A theoretical analysis is carried out to establish lower and upper bounds for the
drug in the retina as a function of the initial conditions, external sources and
boundary conditions. The problem is then solved by the use of numerical tech-
niques, establishing the drug profile as a function of time from drug implant. A
sensitivity analsysis is also carried out for different molecular properties.
4
Part I
5
CHAPTER 1
Microvascular Structure and Function:
Biophysical and Mathematical Background
7
Chapter 1. Microvascular Structure and Function: Biophysical and Mathematical
Background
itself. These reflect in very different network patterns depending on the location in the
body (see Fig.1.3). However, despite these different topological arrangements, certain
common features are found in all vascular beds. The arteriolar and venous networks are
similar in branching architecture and are organized in a parallel manner. Venous ves-
sels are more numerous and have relatively wider lumens with respect to the arterioles
(approximately 35-50%). Blood vessels have a hierarchical organization, where each
Figure 1.2: Pressure drop along the systemic circulation. Image modified from Pearson Educational,
Inc.
8
1.1. Organization of microvascular districts
vessel bifurcates/converges into/onto two (or in some cases, more then two) daughter
vessels whose diameters and lengths decrease going away from the main artery/vein.
The capillary networks, unlike arteriolar and venous networks, don’t have a tree-like
topology but consist of numerous and denser meshes some 20-30µm apart.
Figure 1.3: Microvascular networks in different tissues obtained by scanning electron microscopy. Scale
bars indicate a length of 40µm. Observe the wide variety of microvascular patterns. Figure from
[205]
The vessel walls of arterioles and venules are composed of three distinct layers or tunics
(see Fig.1.4), whose width decreases with decreasing vessel diameter. We recognize:
• tunica intima: the innermost tunic surrounding the central space of the vessel in
which blood flows (called lumen), containing the endothelium. The endothelial
cells of the arteries are linked by tight junctions that establish a blood-retinal bar-
rier to prevent the movement of large molecules or plasma proteins in or out of the
retinal vessels. In venules, instead, the endothelial cells exhibit loosely or poorly
organized tight junctions [190]. These latter make venules much porous then ar-
teries. Both in arterioles and in venules, small molecules, such as oxygen, can
easily pass through the endothelial cell membranes through the ionic channels;
• tunica media: located in the middle layer, consists mostly of one or more layers of
smooth muscle cells whose principal function is, in different measure in each type
of blood vessels, the active regulation of blood vessel diameter (see Sect.1.4);
9
Chapter 1. Microvascular Structure and Function: Biophysical and Mathematical
Background
• tunica adventitia (or tunica externa): located in the outer most layer, is composed
of collagen fibrils that protect the vessel and anchor it to surrounding structures.
Although venules and arterioles have the same vessel wall structure, the venular wall
thickness is thinner than their arteriolar counterpart and muscular components are less
prominent and absent in venules with diameter smaller than 50µm [160]. In systemic
arterioles, instead, 70-80% of the vessel wall is constituted by smooth muscle cells.
In spite of arterioles and venules, capillary walls are only one cell thick to make pos-
sible an easy exchange of nutrient and waste products between blood and tissue. As a
matter of fact, capillaries provide a large surface area for exchange. In the peripheral
circulation the surface area is estimated to total about 70m2 in humans [160].
Figure 1.4: Schematic structure of the blood vessel wall. Image modified from Person Educational, Inc.
10
1.3. Oxygen transport in blood
on the contrary, carries a large fraction of the RBCs from the parent vessel. Therefore
the hematocrit in a vessel results to be a function of hematocrit in the upstream vessels,
in addiction to its diameter.
Let us denote by the symbol [·] the concentration of each species and let [Hbtot ] =
[HbO2 ] + [Hb] be the total (constant) haemoglobin concentration. Let cHb = 1.344
ml
of O2 per gram of heme be the O2 carrying capability of the heme site [156]; then, the
concentration of bound O2 in RBCs is given by [O2,b ] = cHb [HbO2 ].
A quantity of interest is the saturation function which represents the momentary frac-
tion of bound haemoglobin sites ([HbO2 ]) over the total available haemoglobin sites
11
Chapter 1. Microvascular Structure and Function: Biophysical and Mathematical
Background
([Hbtot ]). If the dissolved and bound oxygen are assumed in chemical equilibrium
within the RBCs (i.e. the concentrations of dissolved and bound oxygen remain con-
stant in time), a closed quantitative description of the oxygen dissociation dynamics
(saturation function) can be expressed analytically [159]. We use in this work the Hill
equation (see, e.g., [156] and references therein)
pnHill
SO2 = . (1.1)
pnHill + pn50Hill
The functional form of SO2 , see Fig.1.6, depends on the partial pressure of free oxy-
gen p, usually measured in mmHg, relative to the half-saturation constant p50 and the
exponent nHill . The value nHill = 2.7 fits well the data for normal human blood in
the saturation range of 20-98%, for which p50 is about 26 mmHg. Several other fac-
tors influence oxygen saturation, including carbon dioxide levels, pH and temperature.
Their effect can be modelled by letting p50 and nHill be a mathematical function of such
parameters [48].
Figure 1.6: Saturation function curve obtained by Hill’s equation 1.1 with nHill = 2.7 and p50 =
26mmHg.
The oxygen partial pressure p in (1.1) can be computed from the free O2 concentra-
tion [O2 ], using Henry’s law
p = α−1 [O2 ], (1.2)
where α is the Bunsen solubility coefficient.
Finally, observe that combining relation (1.1) with the definition of the bound oxy-
gen concentration in RBCs, we have (with a slight abuse of notation)
[O2 ]b = CHb SO2 , (1.3)
where CHb = cHb [Hbtot ].
12
1.4. Regulation of blood flow in microcirculation
the blood flow and driving pressure is generally mathematically described in term of
vessel resistance (see Chap.3 for more details). This latter is influenced by several
interacting factors, among which the geometrical and mechanical characteristics of the
vessel and the biophysical properties of blood [205] (e.g. blood viscosity, see Chap.3)
are the main determinants. While the effect of the blood viscosity is described in the
previous sections, in this section we focus on the mechanical ("physical") characteristic
of the vessel.
Passive regulation
It is well known that the elastic fibers included in the tunica media provide vessels with
an elastic component. The passive constriction or swelling of the radius of the vessel
is due to the transmural pressure (difference between luminal and interstitial pressure).
Moreover, especially on the venous side, the occurrence of a negative value of transmu-
ral pressure can lead the vessels to lose their circular cross section and to buckle until
collapse. These vessels assume then the properties of a collapsible tubes [24, 99]. This
passive mechanism is usually mathematically depicted by an area-pressure relationship
called "tube law" [66, 117]. Fig 4.6 shows an example of tube law for a thin-walled
latex rubber tube along with typical cross section shapes.
Figure 1.7: Sketch of the tube law relating the transmural pressure difference and the dimensionless
cross section area ratio. A0 represent the cross section area of the tube in the unloaded condition.
Figure from [153].
The main features of arterial/venous tube laws are: i) the S-shaped profile deter-
mined by the passage from a distensible to a collapsing regime, when the transmural
pressure match the buckling (transmural) pressure (when A/A0 ' 1); ii) positive con-
vexity of the curve in the range of positive transmural pressure; iii) the saturation at a
maximal cross section area for high transmural pressure; iv) the "snap action” (i.e. a
change in shape over a pressure range so small as to be consider negligible) in proximity
of the zero/negative transmural pressure ( region in Fig.4.6).
In venules, the buckling pressure is about zero, while in the arterioles the critical
13
Chapter 1. Microvascular Structure and Function: Biophysical and Mathematical
Background
buckling pressure is much lower. Some experiments evidence a mild reduction of arte-
rial cross section area until the transmural pressure reaches −10 to −20mmHg [66] or
−50 to −40mmHg [53] thanks to the thick muscularis wall.
In literature there are different mathematical relationship describing the passive be-
haviour of the vessel wall. Two widespread currents can be distinguished: some authors
approximate the tube law by a relationship obtained from the mechanical description
of the thin/thick vessel wall, (i.e. by Laplace’s law) [65, 66, 80, 110]; others use an em-
pirical (polynomial) relationship [29,66,110,117,143] obtained by fitting experimental
data.
Active regulation
In addiction to the passive behaviour, many tissues and organs within the body are
able to intrinsically regulate their blood sources in order to meet their metabolic and
functional needs. This regulation mainly takes place in arterioles with a range of di-
ameter variations that can be as much as ± 100% of the baseline value. Regulation
mechanisms can be originated within blood vessels (due to variation of shear stress or
release of vasoconstriction /vasodilation molecules) or from the surrounding tissue (due
to changes in metabolic demand and interstitial pressure).
Namely, we consider (see also Fig.1.8):
• myogenic autoregulation, due to variations in transmural pressure (Bayliss effect
or compliance phenomenon): an increase in transmural pressure causes an in-
crease of wall tension and intensification of contractile activity of SMCs. A de-
crease of transmural pressure leads to the opposite effect;
• shear stress dependent autoregulation,related to the release of a vasodilating/con-
tractor factors by endothelial layer in response to variations in shear stress on
lumen-vessel wall interface;
• metabolic autoregulation, related to the maintenance of a suitable supply of both
oxygen and other nutrients in tissue. This mechanism alters blood flow in response
to variations of metabolic activity in tissues and/or to keep a constant chemical
content in bloodstream.
Figure 1.8: Schematic representation of autoregulation mechanisms. From left to right: myogenic, shear
stress dependent and metabolic autoregulation.
14
1.4. Regulation of blood flow in microcirculation
the endothelial layer, among which the most relevant species is Nitric Oxide (NO), a
powerful vasodilator (see Chap.5 for details).
Figure 1.9: Vessel diameter as a function of pressure in the intraluminal pressure. The external pressure
is assumed to be constant (=15mmHg). Solid curves are model predictions of [33] for the active
(considering different blood flow values) and passive regulation response of a retinal vessel. Symbols
are the corresponding experimental results.
15
CHAPTER 2
Geometrical Models of Microcirculatory Networks
The network geometry (vessel shapes, diameters, length, topology,...) and, of course,
the boundary conditions associated with network inputs and outputs deeply influence
the blood flow distribution along the network itself.
The geometry of a real vessel network can be extracted from digitized images of ex-
perimental measures, for example computed tomography angiography scans, scanning
electron microscopy. Another approach is to construct a realistic vessel network ex-
novo on a computer, for example using fractal trees or stochastic models. These models
are able to obtain results (i.e. velocity and pressure fields) ) consistent with the experi-
mental ones. In Fig.2.1 we show an example of the microcirculatory network geometry
obtained by digitized imaging (left) and fractal algorithm (right). In the following, we
give a brief description of these two approaches.
17
Chapter 2. Geometrical Models of Microcirculatory Networks
Figure 2.1: Left: digitized image from rat mesenteric circulation; right: fractal tree generated on a
computer.
fact that some of the above cited techniques are not able to catch images of small vessel
or capillaries, which are under the power resolution of such techniques. Another prob-
lem can derive from the difficulty to extrapolate from the image the complete geometry
(cross section area and shape, wall thickness) of blood vessels. Since the reproduc-
tion of a real geometry is very difficult, the alternative is to resort to mathematical and
computational models which can help reducing this gap in knowledge generating an
artificial-realistic network based on experimental measurements. The simplest mathe-
matical models are compartments. In these models, vessels are grouped in classes (i.e.,
small/large arteriole/venule, capillary) with the assumption that vessels within a certain
class are similar and exhibit the same properties. Different classes are coupled by con-
servation laws. This approach maintains a low number of unknowns and is capable of
reproducing the behaviour of the system in an average manner. However, the spatial
distribution of field variables is lost as well as the complexity of the internal interactions
between vessels in the microcirculatory network. These latter aspects are considered
instead in a different set of papers, which represent networks as vessel graphs. Some
of these graphs are constructed by stochastic growth models ( [35, 205]) (i.e. diffusion
limited approach), others others are determined by imposing the radius and length of
vessels between successive orders by fractal relationships ( [35, 50, 149, 199]) or by
Horton-Strahler approach (based on the Strahler ordering) [69, 186].
However, once the network skeleton is obtained and topological connectivity struc-
tures are available, from the modelling viewpoint there is no difference in the treatment
of a geometry, whatever it is its original source.
In the following section we describe two different methods to obtain a realistic net-
work of the retinal microcirculation which are used to perform simulations of the math-
ematical models presented in the following chapters. Namely, in Sect.2.2 we describe
a simple method to obtain a fractal geometry based on a modified Murray’s Law where
its 3D structure is recovered by using a rotation maps considering solid angles cho-
sen in a range which respects anatomical features. In Sect.2.3 we use the same fractal
18
2.2. Generation of fractal trees for microcirculation simulation
19
Chapter 2. Geometrical Models of Microcirculatory Networks
Figure 2.2: Example of a network (arterial side only) considered in the simulations, represented in the
pseudo–3D space. The network is asymmetric (cd1 /f = 0.7), as is evident from the colour mapping
the value of the radii and the different branching structure. To improve the readability of the figure,
the smallest vessels are not shown.
We can analyse the influence of the asymmetry of the network by considering four
different networks with progressively increasing symmetry (that is, with increasing in-
dex cd1 /f ), till reaching a symmetric dichotomic network. In Tab.2.1, we report the
values of different features of these networks (total number of vessels, min and max
route distance of the leaves of the tree, total cross section) for various values of cd1 /f .
There are considerable differences in the characteristic parameters between the four
networks. An increase of the asymmetry of the network leads to an increase of num-
Table 2.1: Characteristic values of parameters of networks (arterial side only) generated by different
degrees of asymmetry in branching (increasing symmetry moving to the right, 0.784=symmetry).
The last column indicates the trend of each parameter for increasing symmetry. Data correspond
to a minimal diameter of 5µm, inlet pressure 40mmHg, outlet pressure 20mmHg (values chosen as
in [199]). Parameters undergo variations determined by increasing homogeneity of the network,
relation between radius and vessel length, constraint of not trespassing the diameter of 5µm. These
elements combined together result into an equivalent network resistance which is more than doubled
passing from cd1 /f = 0.5 to cd1 /f = 0.784. The resistance is computed by the Poiseuille’s Law as
the ratio between the total pressure drop and the inlet blood flow (see Chap.3).
20
2.3. Diffusion limited aggregation algorithm
bers of network vessels. The route distance of blood flow through the consecutive
smaller daughter branches within the network of the great asymmetrical bifurcations
(cd1 /f = 0.5) is shorter than those within the other networks. The opposite situation
occurs, instead, for the maximum route distance,defined as the path length through con-
secutive largest daughter branches within the network. The large differences between
the path length and total cross section of the vessel suggest that a greater asymmetry
leads to a smaller resistance of the network.
1. fix a maximum number of particles M , representing the total mass of the cluster;
create a 2D square lattice with (nDLA × nDLA ) sites;
3. launch another particle starting from the boundary of the lattice and allow it to
move at random (diffuse) on the lattice sites;
4. if the particle comes to a lattice site which is in contact with a site already occupied
by any of the particles of the cluster, let it stick to the existing particle; if the
particle does not stick, let it wander again till it sits in a position;
The fractal dimension of the cluster df relates the mass attached with the maximum
radius rDLA of the cluster (defined on the lattice as the maximal number of sites sepa-
rating a point of the cluster from the center), so that M ∼ (rDLA )df . As shown in [173],
a sticking coefficient, which defines the probability for attachment of a particle to the
cluster, can be used to artificially alter the value of df , since the sticking coefficient
tending to zero yields a denser cluster with a higher fractal dimension.
Fig. 2.3(A) shows a typical DLA cluster generated by the above algorithm, imple-
mented in a Matlab code, with M = 3000 and nDLA = 600, considering sticking
coefficient equal to 1.
21
Chapter 2. Geometrical Models of Microcirculatory Networks
Figure 2.3: A) Realization of fractal tree generated by the DLA algorithm; B) Successive application
of morphological operators (pixels marked in grey are eliminated) to a detail of the tree: I to II:
skeletonization, II to III: opening of closed loops, III to IV: elimination of trifurcations. C) arteriolar
tree segments represented in scale, colours mapped according to the value of the radius.
A local system of cylindrical coordinates (r, θ, s), with s ∈ [0, L/Ne ], θ ∈ [0, 2π] and
r ∈ [0, R], is introduced in each segment E. Elements belonging to the same vessel
share homogeneous mechanical properties. Notice that each element in a vessel has its
own cross section area and the number of elements for each vessel may vary but it has
its own cross section area.
Consecutive vessels (or elements) are simply coupled to obtain the arteriolar and ve-
nous networks, which are been geometrically described by two trees. We denoted these
latter by TA and TV , respectively.
SNc Referring to Fig. 2.4, each network T is split into Nc
vessels V, such that T = i=1 Vi .
Blood flows in the union of the luminal spaces Ωi,j f (corresponding to vessel seg-
22
2.4. Mathematical notation for vascular network modelling
Figure 2.4: Schematic representation of a portion of a microcirculatory network. Each vessel is de-
scribed as a duct with straight longitudinal axis and is further partitioned into a series of consecutive
short elements. The central part of the figure depicts one of such elements, of constant length Le , with
highlighted the domain Ωf occupied by the blood flowing inside the luminal space and the wall struc-
ture. The right part of the figure represents the cross section A of the luminal space with thickness h
of the vessel wall. The local system of coordinates (r, θ, z) on the element is reported as well.
ment E i,j , jth-element belonging to vessel V i ), so that the entire blood flow domain is
SNc SNei i,j
represented by ΩF = i=1 j=1 Ωf .
Junctions between vessel elements and different vessels are simply modelled as
nodal points. Let Nint be the number of junction nodes. For each node nk , k =
1, . . . , Nint , we denote by Ik the set of elements which converge in that node. Moreover,
we denote by Ik− the subset of Ik for which nk is the first endpoint of the element, i.e.,
an element belongs to Ik− if its local axis coordinate is such that z = 0 in nk . Similarly,
we denote by Ik+ the subset of Ik for which nk is the second endpoint of the element,
i.e., an element belongs to Ik+ if its local axis coordinate is such that z = L in nk .
23
CHAPTER 3
Models of Blood Flow and Oxygen Transport in a
Rigid Network
In this chapter we will introduce one dimensional models to describe the coupling be-
tween blood flow and O2 transport in a large vessel network. These models can be used
as an alternative to the more complex three dimensional models or in conjunction with
them as discussed in Chap.7, to reduce the computational cost. The price to pay to use
1D mathematical models is that only averaged quantities are obtained.
The model presented in this chapter consider a three-dimensional Navier-Stokes equa-
tions for the blood that is considered as a mixture of two different fluids, the RBCs
and plasma. The transport equation is then reduced to the one-dimension by a stan-
dard averaging procedure on the vessel cross section. Upon doing this procedure the
blood flow results to be described by a generalized Poiseuille’s law. Observe that the
description of blood as a mixture allows us to include the Fahraeus-Lindquist blood and
plasma skimming effects (see Chap.1). The oxygen transport is represented by a 3D
diffusion-transport-reaction equation. As in [19, 129, 141, 203], we consider an impor-
tant aspect, rarely presents in literature, regarding the oxygen transport in blood: the
oxygen is present in blood in both free and bound (to RBCs) form. While the averag-
ing procedure for the blood flow equation is well assessed in literature, very few work
address the equivalent cross-section averaging for solute concentration [47, 189, 218],
carrying out, however, many further simplifications, for example considering compart-
mental models or flat concentration profiles. In this work, instead, we derive an aver-
aged one-dimensional model for the total (free and bound) oxygen concentration along
the network, considering a parabolic oxygen profile.
The simulation results of this model are presented in Chap.8 in the context of the retinal
circulation.
For the mathematical notations employed in the following sections, we refer to Sect.2.4.
25
Chapter 3. Models of Blood Flow and Oxygen Transport in a Rigid Network
multiscale mathematical modeling and numerical simulation" by Causin, P., Guidoboni, G., Malgaroli, F., Sacco, R., Harris, A. ,
appeared in Biomechanics and modeling in mechanobiology, 15(3), 525-542(2016).
26
3.1. Mathematical Model of blood flow in a rigid vessel
where the first two equations represent the balance of mass for the overall mixture and
for the RBC phase and the last one is the balance of linear momentum, respectively.
In system (3.1), µb is an effective mixture coefficient representing the whole–blood
viscosity.
- Continuity equation
Z
R 1 ∂(rvr )
∂s A
vs dA + dA, = 0
A rZ ∂r (3.2)
R 1 ∂(rvr HD )
∂s A vs HD dA + dA = 0,
Ar ∂r
Observe that, using the streamline boundary condition, the second terms of the two
above integral equations are equal to zero. Introducing the averaged axial velocity:
Z
vs = vs dA/|A| (3.3)
A
and, then, the blood flow rate Qb = |A|v s , the averaged version of the continuity equa-
tions reduce to:
∂s (HD Qb ) = 0,
(3.4)
∂s (Qb ) = 0;
Observe that, under the assumption of constant cross section area along the vessel
axis, the combination of first and second equation of (3.2) leads to ∂HD /∂s = 0 for
s ∈ (0, L). This latter, combined with the initial assumption of HD being independent
from r, leads to the conclusion that HD is constant along each tube.
- Momentum equation:
Before averaging the momentum equation, observe that, after a derivation of the third
equation of (3.1) along the s-axis combined with the second equation of (3.1), we ob-
tain that ∂s2 pb = 0 for s ∈ (0, L) which leads ∂s pb to be a constant along the longitu-
dinal axis. This allows to use the quantity ∂s pb as a scaling parameter. Introducing the
non–dimensional variables r∗ = r/R b and v ∗ = (−µb /R
s
b2 /∂s pb )vs , R
b being a character-
istic length of the cross section, we write the dimensionless form of the third equation
of (3.1)3 as [213]:
find vs∗ such that
4r∗ ,θ vs∗ = −1 in A∗ , (3.5)
27
Chapter 3. Models of Blood Flow and Oxygen Transport in a Rigid Network
Almost all parametric empirical descriptions of apparent blood viscosity depend on ves-
sel diameter and hematocrit. The most common law in literature is the one of [167,168],
where the viscosity law is obtained comparing the experimental data in rat mesentery
with theoretical prediction. The authors of [167, 168] proceed in the following manner.
First, they use a scanning procedure to determine experimentally the network topology
(connection matrix) and architecture (diameter and length of each vessel segment). The
boundary conditions in vessel segments entering or leaving the network are experimen-
tally determined, as well. The hemodynamic problem is then solved by an iterative
procedure in order to determine the viscosity law. A system of liner equations, ob-
tained by imposing conservation of blood flow at each network bifurcation, is used to
asses the pressure at each node, the vessel hematocrit (computed using 3.13), the blood
flow. The blood blood viscosity is then post-computed through the assumption of the
Poiseuille law. The resulting viscosity law is:
" 2 # 2
(1 − HD )C − 1
µ0.45 D D
µb = µp 1 + −1 · , (3.9)
µp (1 − 0.45)C − 1 D − 1.1 D − 1.1
28
3.1. Mathematical Model of blood flow in a rigid vessel
where D is the human diameter rescaled from that of rat [69] and
µ0.45 = 220 exp(−1.3D) + 3.2 − 2.44 exp(−0.06D0.645 ) (3.10)
is the blood viscosity with HD =45% and
C = (0.8 + exp(−0.075D)) · (−1 + (1 + 10−11 D12 )−1 ) + (1 + 10−11 D12 )−1 (3.11)
is a constant describing the dependence of the viscosity on the hematocrit.
In Fig.3.1, we represent the viscosity law described by (3.9) as a function of vessel
diameter and for different values of hematocrit. Observe how the apparent viscosity of
blood declines when the diameter is less then 103 µm and this continues up to a criti-
cal diameter value (around 40-50 µm). Below this value the apparent blood viscosity
steeply rises due to the above discussed Fahraeus-Lindqvist effect. For any given diam-
eter, the viscosity curve is shifted up with progressively increasing blood hematocrit.
Figure 3.1: In-vivo apparent blood viscosity as a function of vessel diameters and vessel hematocrit.
Curves obtained from Eq.3.9 (see [168] ).
More complex models of blood viscosity have been considered in further works of
Pries et al.where a mathematical model studying the influence of IEL (a relatively thick
layer lining luminal vessel wall) on blood viscosity is presented [166]. However, due to
the still ongoing controversial nature of this work, we stick to their basic, yet effective,
description.
29
Chapter 3. Models of Blood Flow and Oxygen Transport in a Rigid Network
The quantity pb,j in (3.12)1 represents the blood pressure in the vessel j evaluate in the
node i, whereas the blood flow Qb,j in each vessel j of the network at the converging
node i is synthesized by the last equation of system (3.8), if losses at the entry and exit
sections at points of convergence of segments are ignored.
The presented model requires, in addition, a law describing the RBCs repartition
law along the network. To do this, let us denote by F Qb,d = Qb,d /Qb,f and F Qrbc,d =
Qb,d HD,d /(Qb,f HD,f ) the flow rate fraction of blood and RBCs, respectively, into the
daughter vessel d. According to the model of [167], the fractional flow of RBCs
(F Qrbc ) into one of daughter branch is computed as a function of its fraction of blood
flow (F Qb ) as follows [205] :
F Qrbc,d = 0 if F Qb,d ≤ X0 ,
!−1
F Qb,d −X0
−B1 −B2 logit
F Qb,d −2X0 (3.13)
F Qrbc,d = 1+e if X0 ≤ F Qb,d ≤ 1 − X0 ,
F Qrbc,d = 1 if F Qb,d ≥ 1 − X0 ,
where logit(x) = log(x/(1−x)), and the parameters B1 , B2 , and X0 defining the phase
separation characteristics of the bifurcation. In this work we consider for B1 , B2 and
X0 the expression from [166], namely:
(Dd1 /Dd2 )2 − 1 1 − HD,f
B1 = −13.29 ,
(Dd1 /Dd2 )2 + 1 Df
1 − HD,f 1 − HD,f
B2 = 1 + 6.98 , X0 = 0.964 .
Df Df
where D is the vessel diameter (in µm), d1 and d2 denote the daughter vessels and f the
father vessel. In a diverging bifurcation (on the arterial side of network), after determin-
ing F Qrbc,d of a daughter vessel by Eq. (3.13), we adopt the repartition relation [163]
HD,d = (F Qrbc,d /F Qb,d )HD,f ,
to obtain the value of haematocrit in the daughter vessels. In a converging bifurca-
tion (on the venous side of the network), after having computed the value of HD in
the smaller daughter vessel, the value of HD in the larger vessel is determined by the
assumption of conservation of RBCs flux. Namely:
HD,f = (HD,d1 F Qb,d1 + HD,d2 F Qb,d2 )/F Qb,f ,
30
3.1. Mathematical Model of blood flow in a rigid vessel
In Fig. 3.2, we show the shape function vh,i = vh,i (z) ∈ Vh,(g1 ,g2 ) relative to node i. This
shape function is linear on each element belonging to Ii+ ∪ Ii− and such that vh,i = δi,r ,
r = 1, 2, . . . , Ntot , where we have set Ntot = (Nint + Nbdr ). These characteristics reflect
the web–like geometry of the domain. We define, ∀Qh , Zh , wh ∈ Wh,z1 , vh ∈ Vh,(g1 ,g2 ) ,
the bilinear forms:
Z Z
8µ dvh
A(Qh , wh ) = 4
Qh wh dz, B(vh , Qh ) = Qh dz, (3.15)
T πR T dz
Z
dvh
C(wh , Zh , Qh ) = Zh Qh dz. (3.16)
T dz
The numerical solution procedure, including a fixed point procedure, is carried out as
described in Algorithm 1.
The pressure field pb,start is chosen as the one computed by the model with HD,start ,
which, in turn, is set to 0.45 everywhere.
Referring again to Fig. 3.2 (and omitting for brevity the k superscripts of the inter-
nal iteration procedure), we explicitly write the relations obtained from system (3.17)
in Algorithm 1 for a bifurcation with joining node ni , with converging vessel ele-
ments E l , E k , E m (here, again for brevity, we have used a shortened notation for vessel
elements):
where Ri represents the vessel resistance in the element i. Notice that here we have
implicitly made use of the fact that functions in Vh,(g1 ,g2 ) are piecewise linear continuous
over T , so that they ensure the automatic satisfaction of the pressure coupling condition.
The value of the pressure for all the vessels converging in the node ni is thus uniquely
identified by pi . Substituting Eqs. (3.18),(3.19),(3.20) (generalized Ohm’s laws) in
31
Chapter 3. Models of Blood Flow and Oxygen Transport in a Rigid Network
Algorithm 1 : fixed point iteration to compute the fluid field on the network
given pb,start , HD,start ;
fix toll, kmax ;
(0)
set p(0) = pstart , Q(0) = Qstart , HD = HD,start , k=0;
while and(err ≥ toll, k ≤ kmax ) do
solve for
k+1
given Qkh ∈ Wh,0 , find HD ∈ Wh,0.45 such that:
∀ wh ∈ Wh,0
k+1
C(wh , HD ; Qkh ) = 0
k+1
compute µb = µb (HD ) by Eq.(3.9)
solve for
(k+1) (k+1)
find (Qh , ph ) ∈ (Wh,0 × Vh,(pin ,pout ) ) such that,
∀wh ∈ Wh,0 , ∀vh ∈ Vh,(0,0)
(k+1) (k+1)
A(Qh , wh ) + B(ph , wh ) = 0,
(3.17)
(k+1)
B(vh , Qh )=0
(k+1) (k) (k)
err= max{kHD − HD k/kHD k, kp(k+1)
b − p(k) (k)
b k/kpb k };
k=k+1;
end while
(k) (k)
set HD = HD , pb = pb .
Eq. (3.21), we end up with a reduced relation in the sole nodal pressure unknowns.
Carrying out this procedure for all the nodes, we obtain the linear algebraic system
MP = F, (3.22)
where P ∈ RNtot ×1 is the vector of nodal pressure dofs, F ∈ RNtot ×1 is the right–hand
side and M ∈ RNtot ×Ntot is the stiffness matrix. Boundary conditions are then enforced
in system (3.22) to ensure uniqueness of the solution. To recover fluxes, we use on each
element the corresponding Ohm’s law.
32
3.2. Oxygen transport and dynamics
Figure 3.2: Bifurcation of the network including the joining node ni and the converging vessel ele-
ments E l , E k , E m . On each element, the arrow indicates the positive direction of the local z axis.
This choice implies that Ii+ = {l}, Ii− = {k, m}. The linear “web–like” shape function vh,i ∈ Vh
is also represented.
Diameter ranges
Arterioles 1 2 3 4 5
pb [mmHg] 33.21 ± 2.38 38.8 ± 0.53 39.93 ± 0.38 40.67 ± 0.25 41.09 ± 0.25
HD 0.44 ± 0.05 0.45 ± 0.01 0.45 ± 0.006 0.45 ± 0.005 0.45 ± 0.001
µb [cP] 8.33 ± 1.99 2.83 ± 0.28 2.31 ± 0.03 2.36 ± 0.03 2.48 ± 0.04
Venules 1 2 3 4 5
pb [mmHg] 23.28 ± 1.75 19.95 ± 0.31 19.29 ± 0.23 18.83 ± 0.15 18.56 ± 0.16
HD 0.45 ± 0.05 0.45 ± 0.01 0.45 ± 0.006 0.45 ± 0.005 0.45 ± 0.001
µb [cP] 7.57 ± 2.32 2.58 ± 0.19 2.33 ± 0.04 2.44 ± 0.04 2.57 ± 0.04
Table 3.1: Mean values and standard deviations of blood pressure Pb , hematocrit HD and blood vis-
cosity µb along the network, binned in 5 diameter ranges. Arterioles: 1: 0 ≤ D < 20.5, 2 :
20.5 ≤ D < 41, 3 : 41 ≤ D < 61.5, 4 : 61.5 ≤ D < 82, 5 : 82 ≤ D < 123; venules: 1:
0 ≤ D < 24.5, 2 : 24.5 ≤ D < 49, 3 : 49 ≤ D < 73.5, 4 : 73.5 ≤ D < 98, 5 : 98 ≤ D < 147.
The vessel diameter D is expressed in µm.
value of blood viscosity in the first bin, both in arteriolar and venous tree, is due to the
Fahraeus-Lindquist effect.
multiscale mathematical modeling and numerical simulation" by Causin, P., Guidoboni, G., Malgaroli, F., Sacco, R., Harris, A. ,
appeared in Biomechanics and modeling in mechanobiology, 15(3), 525-542(2016).
33
Chapter 3. Models of Blood Flow and Oxygen Transport in a Rigid Network
Under the hypothesis of invariance of the O2 distribution with respect to the az-
imuthal coordinate, the multiphase model for O2 transport in a vessel is constituted by
the following equations:
a) O2 dissolved in RBCs:
∂(HD [O2 ]) ∂ ∂
+ HD [O2 ]vs − (Dc HD [O2 ])
∂t ∂s ∂s
1 ∂ ∂ (3.23a)
+ rHD ([O2 ]vr − (Dc [O2 ])
r ∂r ∂r
0 O2
= −kHD [O2 ] · [Hb] + k HD [HbO2 ] + J1,2 ,
b) O2 bound in RBCs:
∂(HD cHb [HbO2 ])
∂t
∂ ∂
+ HD cHb [HbO2 ]vs − (HD DHb cHb [HbO2 ])
∂s ∂s (3.23b)
1 ∂
+ rHD cHb [HbO2 ]vr
r ∂r
= kcHb HD [Hb] · [O2 ] − k 0 cHb HD [HbO2 ],
c) O2 dissolved in plasma:
∂((1 − HD )[O2 ])
∂t
∂ ∂
+ (1 − HD )([O2 ]vs − Dc [O2 ])
∂s ∂s (3.23c)
1 ∂ ∂
+ r(1 − HD )([O2 ]vr − (Dc [O2 ])
r ∂r ∂r
O2
= J2,1 ,
where vr and vs are the radial an longitudinal blood velocity, Dc and DHb are the dif-
fusivity coefficients (assumed to be equal in each direction) in plasma and RBCs of
O2 O2
dissolved O2 and hemoglobin, respectively, and J1,2 = −J2,1 is the O2 flux between
the RBC and plasma phases. Due to the small size of the vessels, in the above equa-
tions we have neglected the haemoglobin radial diffusion, according to the observations
in [94] related to the limited possibility of RBCs rotations and excursions from straight
streamlines in vessel with diameter less than 100µm.
Further on, the detailed O2 tension distributions numerically calculated by [145] show
that, in the range of microvessel diameters we are interested in, processes within and
in the near vicinity of RBCs have a negligible effect on the resistance to O2 transport.
This finding suggests that we can study a simplified version of Eqs. (3.23a),(3.23b) and
(3.23c) by considering that: i) as suggested by the O2 tension distributions calculated
in [145], it is reasonable to assume that dissolved and bound O2 are in chemical equi-
librium within the RBCs and, consequently, concentration gradients within the RBCs
34
3.2. Oxygen transport and dynamics
are negligible; ii) locally, plasma and haemoglobin dissolved O2 are in equilibrium,
which implies that the O2 tension is continuous across the RBC membrane. Let i = 1
denote the RBCs component and i = 2 the plasma component. From the mathematical
viewpoint, assumptions i) and ii) allow to:
1. use a closed quantitative description of the oxygen dissociation dynamics, by
means of the O2 saturation function SO2 . In this work, we consider the satura-
tion function described by the Hill equation [156]
pnc1Hill
SO2 = , (3.24)
pnc1Hill + pn50Hill
where pc1 is the dissolved oxygen tension in RBCs (see Sect. 1.3 for details about
functional form (3.24) )
2. neglect diffusive fluxes within RBCs;
3. assume pc1 = pc2 = pc .
Let αc = α1 HD + α2 (1 − HD ) be the mixture solubility coefficient; then, adding
together Eqs. (3.23a),(3.23b) and (3.23c) and using the above assumptions, we obtain
the following single-equation model for the total (bound+free) O2 concentration in the
blood mixture:
∂
(αc pc + HD CHb SO2 (pc ))
∂t
∂
+ vs (αc pc + HD CHb SO2 (pc ))
∂s
∂ DHb (3.25)
−Dc (αc pc + HD CHb SO2 (pc ))
∂s Dc
1 ∂ ∂
+ r vr (αc pc + HD CHb SO2 (pc )) − Dc αc pc
r ∂r ∂r
= 0,
35
Chapter 3. Models of Blood Flow and Oxygen Transport in a Rigid Network
vessel wall and tw is the wall thickness, estimated as a function of the vessel diameter
according to [111]. In the following section we present a more detailed model that
takes into account the passage of O2 across different layers of which the vessel wall are
composed.
36
3.2. Oxygen transport and dynamics
the lateral boundary condition (3.26). Notice that, the function form fc in (3.28) is
similar to that used in the blood flow. Moreover, observe that functions fc (r), and also
RR
fv (r), enjoys the property 0 rfc , v(r) dr = R2 /2, irrespectively of the value of δ.
Observe that, thanks to this property,
where fO2 (r) = fHD (r) = 1. Before starting with the averaging procedure, we intro-
duce the linearization of the saturation function as SO2 (pc ) ≈ S̃O2 pc , where the quantity
S̃O2 is a suitable constant value (see [159] and [218] for thorough discussion on this
topic). In this work, relying on fixed point map used for the numerical resolution, we
consider
pk,n Hill −1
c
S̃O2 = k,nHill pc , (3.30)
pc + pn50Hill
where pkc is the oxygen partial pressure computed at the previous internal iteration.
Introducing thus the variable p∗ in Eq.(3.25), written in the form (3.27), we integrate
the transport equation Eq.(3.25), over the cross section A and then we come back to the
variable pc by the equivalence (3.29), obtaining the following contribution:
∂R ∂R ∂R
vr|r=R = vs|r=R + = ,
∂s ∂t ∂t
and observing that
R Z R
R2 R2
Z
fc (r)rdr = , fc (r)S̃O2 rdr = S̃O2 ,
0 2 0 2
Z R
R2
S̃O2 rdr = S̃O2 ,
0 2
we obtain:
37
Chapter 3. Models of Blood Flow and Oxygen Transport in a Rigid Network
∂
A1 = A ( αc pc +HD CHb S̃O2 pc ) +
∂t
− 2π (vr (αc pc + HD CHb SO2 (pc ))r)
r=R
observing that:
R R
R2 R2
Z Z
fv (r)rdr = fv (r)S̃O2 rdr = S̃O2
0 2 0 2
and defining the moments
R
2(δ − 3)
Z
2
ω1 = 2 fv (r)fc (r)rdr = ,
R 0 3(δ − 2)
Z R
2
ω2 = 2 fv (r)fc (r)S̃O2 rdr = ω1 S̃O2 ,
R 0
we obtain
∂
A2 = Aαc pc v s pc ω1 + (1 − ω1 )βpt,w +
∂s
AHD CHb S̃O2 v s pc ω1 + (1 − ω1 )βpt,w
reduces to:
A4 = 2π rvr αc pc + HD CHb SO2 (pc ) .
r=R
38
3.2. Oxygen transport and dynamics
39
Chapter 3. Models of Blood Flow and Oxygen Transport in a Rigid Network
related to the contribution of the O2 partial pressure in the surrounding tissue. Notice
that, due to the assumption of constant interstitial pressure, J w is constant along each
vessel, and therefore its axial derivative is equal to zero.
Remark 1. Observe that if in Eq.(3.31) we consider fc (r) = 1 (flat profile) and thus
ω1 = 1, the permeability coefficient Lp necessarily must be zero in such a way the
relation (3.26) is respected and then Eq. (3.32) reduces to:
∂J s (3.35)
= 0.
∂s
where Js is the flux of the unbound O2 in the axial direction (including diffusive and
convective components).
40
CHAPTER 4
Models of Blood Flow in a Distensible Network
As described in the previous section, where it is proposed a model describing blood and
oxygen transport along a large rigid network, the distribution of flow and pressure drop
is a highly interlaced function of vessel resistances. In turn, several interacting factors
have an impact on vessel resistance: the geometrical and mechanical characteristics of
the vessel and the biophysical properties of blood [205] are the main determinants along
with, in a coupled fashion, the distribution of flow and the pressure loads themselves.
The microcirculatory networks are, situated in a pressurized environment that elicits a
mechanic responses of the distensible vessel walls and then the vessel cross section area
and geometry undergo deformations under pressure loads. This in turn cause, possibly
strong, changes in the vessel resistance. As a matter of fact, microcirculatory networks,
characterized by low values of the luminal pressure, comparable to the surrounding
interstitial pressure, can experience severe reductions of the luminal cross section, till
collapse (especially on the venous side), both in physiological conditions and -more
dramatically- in presence of pathologies.
In this chapter, we propose a mathematical and computational models to study the
behaviour of distensible microcirculatory networks to analyse the interaction of the
passive elasticity of the vessel wall on the blood flow. A simplified fluid-structure in-
teraction approach is considered. Namely, blood flow in each vessel is accounted for by
a generalized Poiseuille’s law where the vessel resistance is function, among the others,
of the area and shape of the tube cross section. To dispose of these latter parameters,
we need of the so-called tube law, which related the transmural pressure to the cor-
responding deformed cross section area. Both theoretical (including the phenomenon
of venous buckling, see Sect.4.1) and empirical (experimentally-derived) (see Sect.4.6)
tube laws are used to describe the relationship between the cross section area and the
pressure loads. The numerical results and comments are reported at the end of each
41
Chapter 4. Models of Blood Flow in a Distensible Network
section. For the mathematical notations employed in the following sections, refer to
Sect.2.4.
can be re–formulated as a generalized Ohm’s law connecting flux and pressure gradient
dpb
Qb = −σ . (4.3)
ds
1 This section is part of the article "Blood flow repartition in distensible microvascular networks: Implication of interstitial and
outflow pressure conditions" by Causin, Malgaroli, F. appeared on Journal of Coupled Systems and Multiscale Dynamics, 4(1),
14-24(2016).
42
4.2. The structure equations
Notice that the conductivity σ is a function of the geometry of the vessel cross section,
this latter being itself an unknown of the problem. The coupling with equations for the
structure (vessel wall) through the pressure loads closes the problem.
Remark 2. In this work, we consider a dynamic viscosity depending, among the others,
on the vessel cross section diameter (and, more in general on the hydraulic diameter),
as presented in [163] and described by Eq.(3.9). If the vessel cross section change,
this yields a nonlinear coupling with the geometry, which is dealt with an iterative
technique, resulting at each iteration in a constant value for µb computed from values
of the past iteration. For this reason, as for the present section, we consider µb (with a
slight abuse of notation) to be a constant, given, value.
Gathering the averaged continuity equation (3.4)2 and the Ohm’s law (4.3), we ob-
tain the (equivalent) system:
find Q and p such that
dQ dp
= 0, Q = −σ in Ωf . (4.4)
ds ds
Remark 3. To obtain a coherent fluid–structure coupling in the present model, the
conductivity parameter in system (4.4) must be constant in each vessel element. For
this reason, albeit the blood pressure in each vessel element is a linear function of the s
coordinate, we consider the structure to be loaded on the lumen interface with a unique
constant pressure p function of p (for example, its average along the element length)
and thus σ = σ(p). This approximation is acceptable if the number of elements in each
vessels are chosen in a such a way that the pressure gradients are not excessively high.
plication to the eye retina circulation" by Causin, P., Malgaroli, F. appeared on Journal of Mechanics in Medicine and Biology,
15(02), 1540027(2015).
43
Chapter 4. Models of Blood Flow in a Distensible Network
pressure. This procedure makes a limited use of physical and mechanical parameters,
with no resolution of various vessel characteristics. In fact, in principle, a different em-
pirical relationship should be extrapolated from experimental data of different vessels.
Each vessel element is modelled as a thin walled elastic ring assumed to be circular
in undeformed and deformed conditions. The same cylindrical coordinate system of
the fluid model is considered. As described in Sect.3.1.1, in the case of circular cross
section, Eq.(4.1) reduces to:
πD4
σ= , (4.5)
128µb
D being the deformed radius. Thus Eq.(4.3) becomes the well known Hagen-Poiseuille
relation:
128µb ∂p
Q= . (4.6)
D4 π ∂s
In literature, the diameter–pressure relationship is commonly found in the form DD0 =
f (pt ), where the vessel diameter (or area) is normalized to the vessel (or area) diameter
(D0 ) at reference vascular pressure pt,0 (or in undeformed status). In this context, we set
pt,0 = 0. From a mathematical view point, the function f : R+ → R must be: i) smooth
enough to guarantee that when the diameter–pressure relation is used in Eq. (4.6), this
latter has a unique solution; ii) such that f (pt,0 ) = 1 and f (z) > 0, ∀z ∈ R+ , in
order to have physiological values of diameters. Examples of f will be discussed in
the following section of simulation results and can be found in [66, 117, 142]. Observe
again that, albeit the blood pressure in each vessel is a linear function, we assume that
the deformed radius is determined by a unique constant pressure value along the vessel
(see Remark.3).
Thanks to the separability of D0 and f (pt ) (the key–step) and to the above regularity
proprieties of f , the Eq.(4.6) becomes:
dpt 128µb Qb
(f (p))4 =− (4.7)
ds πD04
where pt,in and pt,out are the inlet and outlet pressures of the vessel, respectively, and
R0 = 128µL/(πD04 ) is the vessel resistance at reference pressure. Following [117], we
let B : R+ → R be a primitive of the fourth power of f (p), representing an abstract
notion of pressure accounting for vessel distension and such that B(pt,0 ) = 0. Then,
Eq.(4.8) can be rewritten in analogy to Ohm’s law as
B(pt,in ) − B(pt,out ) = R0 Qb . (4.9)
If losses at the entry and exit sections at points of convergence of segments are ignored,
a repeated application of Eq. (4.9) synthesizes the flow in each vessel of the network.
44
4.2. The structure equations
Once having set up the model for a single segment, we can move towards the study of
a network T of vessels. From a mathematical viewpoint, this means to find suitable
joining conditions at the branchings. At each bifurcation of the network, conservation
of flow and uniqueness of pressure hold.
If the same distensibility function f (p) is adopted in a subset S ⊆ T of the network,
continuity of pressures implies as well continuity of the antiderivative B = B(p) in S,
due to the fact that B(p) is invertible since ∂B(pb )/∂pb = (f (pb ))4 > 0, and the inte-
gration constant for B has been fixed. When S = T , further relations exist between
B(p) and the blood pressure in the undeformed (rigid) condition which simplify the
resolution of the network (see [117] for a detailed discussion).
Solution procedure
To computationally deal with the problem on the complete network, we adopt a do-
main decomposition like approach. Namely, SMwe assume that there exist M disjoint
connected subtrees S1 , . . . , SM , such that j=1 Sj = T . Each subtree is character-
ized by its own uniform distensibility relation. Let Nj be the set of nodes of subtree
j, consisting in Nj,I internal nodes, Nj,IF interface nodes shared with the other sub-
trees and Nj,B boundary nodes (see Fig.4.1). In each subtree flux conservation and
continuity pseudo–pressures B is enforced at nodes Nj,I , upon using the constitutive
law (4.9), and boundary conditions at nodes N1,B . Closure of the system is obtained by
enforcing interface conditions between the subdomains, and, namely, continuity of the
physical pressures and conservation of blood flow. Notice that since the subdomains
have different distensibility laws, continuity of the abstract pressures is not required.
Such a procedure yields for each subdomain a linear algebraic system in the abstract
pressures Bj at nodes Nj . Once the system is solved, the values of physical pressure
are computed by inverting the function B(·). Well-posedness of the problem on each
subtree implies that there exists at least an interface or boundary node on which pres-
sure, and thus its corresponding B, is given. Here, for ease of exposition and adherence
with the actual case we study, we consider two subtrees, S1 and S2 , corresponding to
all the arterioles and the first half (in length) of capillaries and to all the venules and the
second half of capillaries, respectively (see Fig. 4.1). At the inlet of S1 , the flux QIN is
given, while at the outlet of S2 , the pressure pOUT is assigned. In order to solve the split
problem, we introduce an iterative procedure which entails the solution of a sequence
of boundary value problems on each subdomain, plus relaxation conditions at the inter-
face. The two subdomain share the interface nodes IF. Let us denote by the subscript
(j, IF) the restriction of field variables of the j–th domain to nodes Nj,IF . With a slight
abuse of notation, we denote by Qj,IF the fluxes in the segments of domain j converg-
ing in the nodes Nj,IF . The numerical solution procedure is carried out as described in
Algorithm2 (the algorithm can be adapted to more complex situations).
Simulation results
45
Chapter 4. Models of Blood Flow in a Distensible Network
work is generated by the fractal algorithm presented in Sect.2.2 with cd1 /f = 0.784
(symmetric branching pattern) and consists of n = 14 generations for both arterioles
and venules, with diameters ranging from 108 µm for larger arterioles, down to 8 µm
for capillaries, up to 147 µm for larger venules. Each terminal arteriole splits into four
capillaries which converge into a terminal venule. The blood viscosity is assumed to
vary vessel-by-vessel as a function of radius by the mathematical expression proposed
by Haynes in [88],
µ∞
µ(r) = , (4.10)
(1 + δ/r)2
where µ∞ is the asymptotic blood viscosity, set to 3.2 · 102 Poise and δ = 4.29. Notice
that here, viscosity is computed according to the reference diameter of each vessel and
then held fixed.
46
4.2. The structure equations
Figure 4.1: Microvascular network representing eye retina microcirculation. Black dots indicate nodes
of the trees. Sketch adapted from [199].
47
Chapter 4. Models of Blood Flow in a Distensible Network
Figure 4.3: Relative diameter variation as a function of vessel hierarchy for Test 1 in arteries (left panel)
and veins (right panel). Inlet flux values vary linearly from 0 to 1.3 · 10−4 cm3 /s moving from bottom
to top of each panel in the direction of the red arrows.
Figure 4.4: Relative diameter variation as a function of vessel hierarchy for Test 2 in arteries (left
panel) and veins (right panel). Inlet flux values vary linearly from 0 to 1.3 · 10−4 cm3 /s. The red
arrows indicates the displacement of the numerical solutions under increasing inlet blood flow.
48
4.2. The structure equations
In Fig.4.5 we show the blood flow distribution and radius variation in symmetric vs
unsymmetric networks (both constituted of ten vessel generations). The unsymmetric
network is obtained by manually reducing the arteriolar reference radius of 20% in the
rightmost district (from 3rd to the 9th generation). The same tube laws as in Test case
2 are assumed, namely the active–passive law fA for arterioles and the passive law fP
for veins.
The high blood flow variation in the radius-modified portion is due to the high pressure
(caused by the reduction of the undeformed radius of vessels in this region) and to
resulting high vessel resistance. Since the mean blood velocity in each vessel is defined
as the ratio between the blood flow and the vessel area, comparing two graphs, we see
that the mean blood velocity is reduced in the affected district with respect to that of the
symmetric case. Leftmost branches and veins in the rightmost branches show a slight
increased radius, whereas the rightmost arteriolar branches show a pronounced radius
decrease. This behaviour is, again, due to the shape of the tube laws (in particular to fA )
and to the different pressure levels corresponding to different distensibility responses
in the network.
Figure 4.5: Relative blood flow and diameter variation along the asymmetric network compared to the
symmetric case. The asymmetric network is obtained by reducing of 20% the arteriolar vessel radii
in the rightmost district. The same tube laws as in Test 2 are assumed.
The above results show that considering a network, albeit simplified, including a
wide range of vessel diameters, allows to connect network geometry and behaviour.
This can be, in perspective, an important factor for understanding the long term conse-
quences on vasculature of hemodynamics, possibly leading to pathological conditions,
via remodelling [170, 171].
49
Chapter 4. Models of Blood Flow in a Distensible Network
In the following we present a theoretical and mechanical method to compute the ves-
sel conductivity from a detailed physically-motivated structural model. We anticipate
in Fig.4.6 the tube law resulting from the present model. Observe in particular the
different behaviour of arterioles (thick-walled vessels) and venules (thin–walled ves-
sels). Observe also how, for these latter, there exists a physiologically plausible value
of transmural pressure under which the tube is not any more circular but assumes a
buckled configuration. Notice that the cross section does not need to be completely
closed for the vessel to be "functionally lost to the network". As a matter of fact, it suf-
fices the section to be small enough to prevent red blood cells passage to compromise
its physiological function [64].
Figure 4.6: Tube laws (transmural pressure vs. area relations) for arterioles and venules as obtained
from the model of Sect.4.2.2. As customary when representing this curve, the cross section area is
normalized over the cross section area at zero transmural pressure. Characteristic cross sections are
sketched for various values of the transmural pressure. Observe the “snap action” of venules (i.e., a
change in shape over a pressure range so small as to be considered negligible) in proximity of zero
transmural pressure. Venules with A/A0 < 10−2 are practically collapsed. The curves are obtained
using the same data considered for Fig.4.11.
In the structural model, each vessel segment is modelled as an elastic ring made of
elastic (Young modulus E) and incompressible material (ν=0.5), assumed to be circular
in undeformed conditions (radius Ru , thickness hu ). The same cylindrical coordinate
system of the fluid model is considered, if not otherwise specified. Small deformations
are considered, similarly to several works in this field, see, e.g., [62, 138]. In Fig.4.7,
we report the notation required for the mathematical discussion and the definition of
the relevant configurations we will consider.
On applying the internal and external pressure loads, radial and circumferential stresses
arise in the ring. We assume axisymmetry and plane stress conditions. Let η = η(r)
be the radial displacement of a point of the vessel wall. Then, we establish the strain–
50
4.2. The structure equations
Figure 4.7: Characteristic configurations in the structural model of Sect.4.2.2. Configuration I is the
experimentally measured “in-vivo” geometry, supposed to be circular. The arrows indicate the steps
followed in the computations to obtain a certain configuration from this configuration. According to
the different modelling chosen as a function of the wall thickness-to- radius ratios (see 4.2.2) the left
side of the figure corresponds to thick–walled rings (the internal radius is indicated), the right side
to thin–walled rings (the mean fiber radius is indicated). Configurations II and IV are the unloaded
configurations, corresponding to the stress-free geometry for the thick–walled ring and to the zero
transmural pressure geometry in the thin–walled ring, respectively. Notice that in the case of a thick–
walled ring the undeformed geometry II differs from the zero transmural pressure geometry III. We
refer to Sect. 4.2.3 for a detailed discussion on the computation of the unloaded configuration. In the
last row, we represent generic deformed geometries of the thick and thin–walled ring cross sections,
respectively. In the case of the thin–walled ring, we also consider the possibility of section buckling,
so that the generic deformed cross section is circular if in pre–buckling conditions (left) or with
a general shape if in buckled conditions (right). The same terminology adopted for the radii also
applies to the vessel wall thicknesses in the various conditions.
dη η
εN = , εT = , (4.11)
dr r
εN being the radial strain and εT the circumferential strain, respectively, and the pseudo–
elastic constitutive equations
E E
σN = (εN + νεT ), σT = (εT + νεN ), (4.12)
1 − ν2 1 − ν2
σN being the principal radial stress and σT the principal hoop stress, respectively, and
E = E(pt ) a functional representation of the Young modulus to be discussed at the end
of this section. We close the problem considering the equilibrium equation
dσN 1
+ (σN − σT ) = 0, (4.13)
dr r
51
Chapter 4. Models of Blood Flow in a Distensible Network
Figure 4.8: Wall thickness relative to inner vessel radius (parameter γ in the model) as a function of
intravascular pressure (image adapted from [165], data obtained from a meta–analysis of literature
studies). Arterioles and venules are designed to withstand different ranges of luminal pressure. The
arteriolar wall is a thick muscularis layer, while the venular wall is a thin structure. Further, similar,
data can be found in [121, 176].
A useful simplification of the expressions in Eq. (4.14) can be obtained for thin–
walled structures, since in this case hu , h2u Rm,u , with Rm,u = (Ru +Ru,e )/2 (virtual
position corresponding to the mean fiber radius). We then obtain the approximations
Rm,u
σN ' 0, σT ' pt , (4.15)
hu
where the second relation represents the well–known Laplace’s law. Thin–wall models
are considered admissible till γ ' 1:10 [207], γ being the ratio between the thickness of
ring with respect to the radius. As shown in Fig.4.8, the venule wall can be considered
a thin structure, since γ is in the range 1:20 to 1:50 [165, 176]. Much different is
the situation for arterioles, for which γ ' 1 : 3 [121], and thus the use of the full
expressions in Eq. (4.14) is required.
In order to derive the expression of the cross section deformed radius, we combine
Eqs. (4.12) with (4.11) and Eq. (4.14) (for thick rings) or Eqs. (4.15) (for thin rings),
respectively, obtaining
(1 − ν) (1 + ν) B2
R = Ru 1 + E B1 − E R2 thick–walled ring,
u
(4.16)
(1 − ν 2 )
R = Ru 1 +
pt thin–walled ring,
γE
52
4.2. The structure equations
Figure 4.9: Infinitesimal wedge–shaped radial section used for the derivation of the balance equa-
tions for ring model with circular cross section. Left: radial deformation. Right: tangential (hoop)
stress σT , normal (radial) stress σN with their increments acting on the wedge faces, along with the
pressure loads.
where for thick vessels R denotes the internal radius (blood–vessel interface) while,
with a slight abuse of notation, for thin vessels R denotes the mean fiber radius. The
deformed ring thickness can be post-computed from incompressibility, yielding h =
p
R2 + h2u + 2hu Ru − R for thick vessels and h = hu Ru /R for thin vessels.
Functional representation of the Young modulus. Whilst for large blood vessels, es-
pecially the carotid, several data have been collected based on experimental measures
possibly supported by the use of mathematical models (see, e.g., [66], [98]), there is a
substantial paucity of data and models for the Young modulus in microvessels. Given
these premises, one can think to simply consider E =const. Then, relations (4.16) be-
come linear in the transmural pressure, but the corresponding relation trasmural pres-
sure vs. cross section area exhibits two non–physiological features: (i) concave form
and (ii) absence of saturation at a maximal cross section area for high transmural pres-
sures. These features can be related to the different reaction to loads of the components
of the vessel wall (collagen, elastin). In this work, we use, thus, a more complex ex-
pression, considering a linear functional dependence of Young modulus with respect to
transmural pressure. This relation is obtained by fitting data from the measurements
obtained in [219] by wire myography in small deactiveted arteries and veins of the rat
mesenteric circulation. The computed steepness of the linear relation is such that the
Young modulus passes from a basal value Eb at zero transmural pressure to roughly its
double when the transmural pressure is increased to 50 mmHg. In the present example
of application of the model to the retinal circulation, we assume Eb = 0.022 MPa for
arterioles and Eb = 0.066 MPa for venules ( basal values chosen as in [80] for the same
microcirculatory district). Analogous trends can be obtained also considering different
sets of measurements, for example the ones in [98] for human coronary arteries.
When considering the possibility of reaching buckled configurations, the model must
also keep into account the bending actions which actually lead to the loss of axialsym-
metry. It is convenient in this context to fix a system of Cartesian axes on the bottom
point of the section (see Fig. 4.10, left). We let s be the arc–length parameter describ-
ing the wall mid-line in counter-clockwise direction from the origin of the axes and
53
Chapter 4. Models of Blood Flow in a Distensible Network
Figure 4.10: Left: coordinate system for the thin–walled ring model in buckled configuration. Right:
tangential (hoop) stress σT , normal (radial) stress σN and bending moment M with their increments
acting on the wedge faces, along with the pressure loads.
we denote by ϕ = ϕ(s) the angle between the positive direction of the x axis and the
tangent to the cross section. The Cartesian coordinates x = x(s), y = y(s) of a point P
identified by arc-length s are given by
Z s Z s
x= cos ϕ ds, y = sin ϕ ds. (4.17)
0 0
Fig. 4.10(right) shows an element wedge of arch length ds along with the normal
stress σN , the tangential stress σT and the bending moment M arising from the pressure
loads. According to the hypothesis of thin–walled structure, the internal actions have
dϕ
a constant average value in the radial direction. Let K(s) = be the local curva-
ds
ture of the section and K b = 1/Ru the curvature of the circular undeformed geometry
taken as reference configuration. From the approximate theory of curved beams (see,
e.g., [201]), the bending moment M = M (s) has the constitutive form
M = EI K − K b , (4.18)
where EI is the flexural rigidity, E being the constant Young modulus (assumed here
to be equal to the basal value one because of lack of data in this context) and I = h3u /12
the area moment of inertia of the cross section per unit length.
The balance of bending moments and forces on the infinitesimal wedge–shaped ra-
dial section of ring per unit axial length is given by the following equations
dM dσN pt dσT
= σN h, = KσT − , = −KσN . (4.19)
ds ds h ds
Combining Eqs. (4.19) with the Eq. (4.18) and using the definition of the curvature,
yields the first order differential equations:
K
ϕ
σN h
d K
= EI . (4.20)
ds
σN KσT −
p t
h
σT −KσN
54
4.2. The structure equations
Linear stability analysis of system (4.20) (see e.g., [197]) shows that a buckled non–
axialsymmetric solution exists for every pressure pt < pt,b , where pt,b = −3EI/Rb3 is
the critical transmural pressure corresponding to the lowest energy mode (azimuthal
wavenumber equal to 2). When pt = pt,b , the cross section (of radius Rb in incipient
buckling) loses its circular shape due to physical instability and buckles into an ellip-
tical shape. For pt < pt,b , progressively, the nearest opposite sides of the section get
close, until they touch if the contact pressure pt,c is reached. The contact point becomes
a straight line segment in contact if the pressure lowers to the contact line pressure pt,cl .
As the pressure is further decreased, the length of the contact line increases and the
associated section area tends to zero forming a dumbbell–like shape (see the character-
istic shapes reported in Fig. 4.6). The buckled configurations have a two-fold symmetry
(since they are related to the wavenumber 2), which allows for solving system (4.20)
just in a fourth of the domain. The approach to solve system (4.20) depends on the
value of the transmural pressure, and namely:
i) for pt,cl < pt < pt,b , we compute numerically the solution under the hypothesis of
isoperimetrical transformations (see also [39] for a similar assumption), by means
of the Matlab function bvp4c and using the following boundary conditions [61]:
– if pt,c ≤ pt < pt,b , we set ϕ = σN = 0 at s = 0 and ϕ = π/2, σN = 0 at
s = π/2;
– if pt,cl < pt < pt,c , we set ϕ = σN = 0 at s = 0 and ϕ = π/2, σT = 0 at
s = π/2;
An appropriate choice of the initial guess shape is of fundamental importance to
kick in the buckling instability in the computation [116]. We consider a guess
shape with a curvature which is a small perturbation (withparameter ε 1) of
1 s
the curvature of a circle of radius Ru , namely Kε = 1 + ε cos . This
Ru Ru
mathematically reproduces the existence in the vessel of imperfections which ac-
tually trigger the instability [66]. The value of ε must be tuned accordingly to the
imposed transmural pressure in order to obtain a physically coherent solution;
ii) for pt < pt,cl , the solution of (4.20) can be found from that for pt = pt,cl by the
similarity transformation [61]
ϕ(s) = ϕcl (scl ), K(s) = (pt /pt,cl )1/3 Kcl (scl ),
(4.21)
σN (s) = (pt /pt,cl )2/3 σN,cl (scl ), σT (s) = (pt /pt,cl )2/3 σT,cl (scl ),
with the coordinate transformation s = (pt,cl /pt )1/3 scl , where 0 < scl < s1 , s1
being the arc-length of the point of contact in the configuration corresponding to
pt = pt,cl .
Once the solution of system (4.20) has been computed, the non–circular buckled
geometry of the section is reconstructed in Cartesian coordinates from Eqs. (4.17).
The knowledge of the geometry of the deformed luminal cross section as a function of
the pressure loads allows for computing the element conductivity from relation (4.1).
In detail, we proceed as follows:
55
Chapter 4. Models of Blood Flow in a Distensible Network
pre–buckling post–buckling
4
πRu4
thick–walled (1 − ν) (1 + ν) B2 (p)
σ(p) = 1+ B1 (p) − /
ring 8µ E E Ru2
4 numerical
πRu4 (1 − ν 2 )
thin–walled
σ(p) = 1+ (p − pe ) solution
ring 8µ γE see Sect. 4.2.2
Table 4.1: Summary of the different expressions and techniques to obtain the conductivity for thick and
thin–walled ring elements. Only positive transmural pressure are considered for the thick–walled
rings. The quantities B1 and B2 are the linear functions of the pressure loads defined in Sect. 4.2.2.
Notice that here we have made explicit the dependence on the pressure indicator p.
- if the deformed section remains circular, relations (4.16) explicitly give the radius
of the blood-wall interface. The solution of problem (3.5) can be found analyti-
cally, and yields the usual parabolic Poiseuille velocity profile [213], from which
the conductivity can be straightforwardly computed by integrating (4.1);
- if the section is buckled, problem (3.5) is numerically solved with finite elements
on a triangulation of the deformed section. Vessel conductivity is then obtained
by 2D numerical quadrature of the integral (4.1). Observe that the buckled config-
uration is numerically computed only for a finite number of transmural pressure
values. However, a continuous conductivity curve can be reconstructed by inter-
polation.
Tab. 4.1 summarizes the different expressions/techniques which give the conductiv-
ity parameter for thick and thin–walled ring elements, respectively. Fig. 4.11 depicts
an instance of the computed vessel conductivity as a function of the transmural pres-
sure considering a representative arteriole with γ = 0.32 and venule with γ = 0.05,
both with Ru = 40 µm. We choose the Young modulus as discussed in Sect.4.2.2 and
we have set pe = 15 mmHg. The red curve with circular markers represents the con-
ductivity parameter of the arteriole, the continuous blue curve the conductivity of the
venule. The dashed blue curve in the region of negative transmural pressures repre-
sents, for comparison, the conductivity of the venule obtained from the second relation
in Tab.4.1 (thin-walled ring) considering a circular cross section with the same area of
the non-circular deformed geometry. Notice how this latter curve significantly differs
from the one obtained with the non–circular geometry, especially in the critical area
around the onset of the buckling. This motivates us to the explicit computation of the
buckled geometry.
We now consider the study of the fluid field in a complete compliant microcircula-
tory network (T ), organized into incoming arterioles, an intermediate capillary bed
and draining venules. As assumed in the case of rigid vessel, at the branching nodes,
we enforce pressure continuity for converging vessels and conservation of mass (see
Sect.3.1.2). At the inlet and outlet nodes (physiologically, more than one inlet/outlet
can be present in the network), we can apply inlet and outlet pressure values (that is,
we impose an overall pressure drop, as in the simulations presented in this work), or an
56
4.2. The structure equations
Figure 4.11: Vessel conductivity curve (log scale) plotted against the transmural pressure obtained for
a representative arteriole (red continuous curve) and venule (blue continuous curve) with radius
Rd = 40 µm. The external pressure is set to pe = 15 mmHg, the Young modulus is chosen as
discussed in Sect.4.2.2. The blue dotted curve represents the conductivity for a circular cross section
with the same area of the buckled configuration at the same value of transmural pressure. Significant
discrepancies arise as pt decreases in the negative half–plane.
Model summary
The following nonlinear boundary value system of PDEs is to be solved in the compli-
ant domain ΩF :
given the connectivity of T , the external pressure, the unloaded configuration and the
mechanical properties of the vessels, find the piecewise constant function Q satisfying
the conservation conditions and the continuous–piecewise linear function p, such that
in each element it holds
dQ dp
= 0, Q = −σ(p) , (4.22)
dz dz
where the conductivity σ(p) is determined as summarized in Tab. 4.1.
Solution procedure
It is convenient to think that the discrete counterpart of (4.4) corresponds to the adoption
of a primal mixed finite element method. As in Sect.3.1.3, the graph T represents the
“triangulation” of the domain, with elements E (see Sect.2.4). Referring to the notation
introduced in Sect.3.1.3, we let R(p) = 1/σ(p) be the non–negative tube resistance
per unit length (observe that σ(p) > 0 in the physiological range) and ∀Qh , wh ∈
57
Chapter 4. Models of Blood Flow in a Distensible Network
p(k+1)
h = ωp p(k+1)
h + (1 − ωp )p(k)
h ;
(k+1) (k+1)
Qh = ωQ Qh + (1 − ωQ )Q(k) h ;
err= max{kQ(k+1) h − Q (k)
h k/kQ(k) (k+1)
h k, kph − p(k) (k)
h k/kph k };
k=k+1;
end while
(k) (k)
set Qh = Qh , ph = ph .
58
4.2. The structure equations
be the generic expression of the structural operator, corresponding to the thick or thin-
walled rings models (direct problem). If the undeformed configuration were known, the
operator S would compute the measured geometry under given pressure loads. In this
context, we have to solve the inverse problem, where the unknowns are the unloaded
configuration and the stress field under which the measured deformed configuration is
in equilibrium. As, in general, S cannot be analytically inverted, we resort to the fixed–
point procedure described in Algorithm 4.
59
Chapter 4. Models of Blood Flow in a Distensible Network
Figure 4.12: Left: configurations I to IV used as unloaded geometries in the numerical experiment
described in the text. The red dotted-line represents the mean radius of the configuration. Right:
distribution of the hoop stress obtained loading configurations I to IV with p = 20 mmHg and
pe = 10 mmHg. Stresses color code the corresponding deformed configurations.
Configuration
Row I II III IV
1 R [µm] 40 38.9 38.5 38.0
2 ∆r R [%] / 2.8 3.5 5.3
3 h [µm] 12.8 13.1 13.2 13.5
4 ∆r h [%] / -2.2 -3.0 -5.6
5 ∆r σT [%] / 9.1 11.9 19.5
Table 4.2: Geometrical data and stress values for configurations I to IV and percentage variations with
respect to values of configuration I. Data refer to the example presented in Sect. 4.2.3 Row 1 and
3: mean radius and wall thickness of the initial configurations as in Fig. 4.12(left); rows 2 and
4: percentage variation of the mean radius and wall thickness in the deformed configuration as in
Fig. 4.12(right); row 5: percentage variation of the hoop stress σT at the mean radius of the deformed
configuration as in Fig. 4.12(right) The percentage variation is defined as: ∆r G := (G − Gd )/Gd .
the resulting deformation and hoop stress fields. A significant discrepancy in the stress
fields is evident. Configuration I yields stresses which differ of about 1 mmHg with
respect to the ones from configuration II (measured vs. unloaded geometry). More sig-
nificant differences arise if configuration IV is used instead of II (thin vs. thick structure
model). The discrepancies are an increasing function of the magnitude of the transmu-
ral pressure, the individual internal and external pressures, and of the Young modulus
(data not reported). Tab. 4.2 summarizes the geometry of each configuration and the
percentage difference in results with respect to the ones obtained from configuration I.
We conclude this section by noting that in the present work we do not consider the
existence of pre–stresses (residual–stresses). It is well know that, if cut radially, vessels
spring open releasing the residual stress and approaching the zero-stress state which is
a sector [66]. This aspect is rather delicate and deserves further future analysis.
Simulation results
The methodology described in the previous sections can address the solution of general
unstructured networks. However, here the simulations are run considering coupled
arteriolar and venular networks geometrically described by a fractal algorithm as in
60
4.2. The structure equations
Sect. 2.2 on each side with cd1 /f = 0.7. We use the parameters reported here below.
We set the network inlet pressure to 42 mmHg and the outlet pressure to 18 mmHg,
respectively. The inlet arteriole has radius equal to 62µm, the outlet venule equal to
72.5µm. We set γ equal to 0.32 for arterioles and 0.05 for venules. The first half length
of each capillary is considered to belong to the arteriolar network, and thus described
as a thick–walled ring, with γ = 0.2. The second half length is considered to belong to
the venular network, and thus is described as a thin–walled ring, with γ = 0.08. Blood
viscosity is described according to equation (3.9), assuming plasma viscosity equal to
1cP and blood hematocrit equal to 45%. We choose to account for non-circular shapes
by using the concept of hydraulic diameter defined as four times the ratio between the
vessel cross sectional area and the wetted perimeter of the cross-section [16].
In the following, we present the results obtained in several simulations which high-
light significant network behaviours.
Assessment of the number of elements in vessels. We study the influence on the
results of the number of element into which each vessel is partitioned. This difference
is extremely small when considering the physiological range of typical external and
internal blood pressures. It is clear that, were the external pressure be significantly
higher than the internal one, using a greater number of elements would allow to obtain
a more accurate description of the buckling phenomenon. In the following simulations,
we set Ne = 3 in each vessel.
Hemodynamical consequences of vessel buckling. We consider a single thin–walled
vessel (venule) with inlet pressure pin = 40 mmHg and pe = 18 mmHg and we study the
flux for outlet pressure pout decreasing monotonically in the range [20, 10] mmHg. The
vessel has undeformed radius equal to 30 µm and length equal to 370 µm. Simulations
are run dividing the vessel into Ne = 800 consecutive segments, with progressively
smaller segments as the end of the tube is approached. In Fig. 4.13(left), we show
the flux as a function of the outlet pressure. When this latter is decreased, blood flow
increases till pout > pe . When pout = pe , the downstream portion of the tube enters
into buckling. The flow reaches then a plateau value and it does not depend any more
on pout . This trend is in qualitative accordance with the predictions of the Starling resis-
tor model. However, in this latter model only two situations are possible, fully patent
or fully closed vessel cross section. The distensible behaviour simulated in the present
work is more complex, since the conductivities are consistently coupled with the trans-
mural pressure. In Fig. 4.13(right), we show as an example the 3D configuration of the
tube when pout = 10 mmHg. Notice the narrow, deformed cross sections in the very
downstream (where the low pressure outlet is) portion of the vessel. A similar configu-
ration was also observed in [93], where a more complex structural shell model coupled
with fluid lubrication theory were used to simulate the experimental setting of the Star-
ling resistor device (notice that in [93] the upstream and downstream cross sections of
the tube are maintained fixed).
Analysis of flow in a bifurcation. In this simulations, we study the influence of the
thin-walled model in vessel bifurcations, both for arteries and veins. We analyse two
different geometries:
(Aaa) : one high pressure inflow and two low pressure outflows, simulating an arteriole
diverging into two smaller branches;
61
Chapter 4. Models of Blood Flow in a Distensible Network
Figure 4.13: Buckling of a single thin–walled distensible vessel. Left: blood flow in the vessel as
a function of pout . As long as pout > pe (dark gray region), the flux in the tube increases as pout is
decreased. When pout < pe (light grey region), the downstream portion of the tube enters into buckling
instability and the flow reaches a plateau, becoming independent of pout . Right: 3D configuration
assumed by the tube for pout = 10 mmHg. A selected number of cross section shapes (black lines) are
highlighted. Notice the very small dumbbell-shaped cross sections formed at the end of the vessel.
(vvV) : two high pressure inflows and one low pressure outflow, simulating two venules
converging into one larger branch.
Observe that the thin-walled model is not appropriated for the arteries, but it is used
here with the purpose to simulate to simulate the buckling phenomenon in an con-
verging bifurcation (typically present in the arterial tree). Moreover, for simplicity, we
assume the Young modulus to be a constant value equal to the baseline values and the
blood viscosity to be a constant set to µ = 3cP.
In each bifurcation, we set the reference radius of the large vessel to 30µm and the radii
of the small vessels to 24µm. Simulations are run dividing each vessel into N = 50
elements. The length is determined by the fractal relationship proposed in [199] (see
Sect.2.2). For each test case discussed below, the figures reporting the corresponding
results are organized as follows: in the first row we show a schematic sketch of the bi-
furcation along with the imposed data, in the second row we show the flow in the three
vessels as a function of the outlet pressure and in the third row we show the pressure in
the common node NC (see Fig. 4.14) as a function of the outlet pressure.
According to the value of the outlet and external pressure in each branch, buckling may
take place or not. The two columns of Fig. 4.15 represents the results in a reference
configuration for different outlet pressures: the first column correspond to a diverging
(arterial) bifurcation and the second to a converging (venous) bifurcation. These sim-
ulations represent the case where no buckling occurs, since po ≥ pe . In this condition,
the flow rate always increases and pressure in node NC decreases, with a nonlinear
trend, as outlet pressure po decreases, according to a generalized Ohm’s law.
We consider then combinations of parameters for which buckling will occur. First we
study the behaviour of the Aaa bifurcation for different outlet and external pressures,
keeping fixed the inlet pressure pin = 40 mmHg. In test Aaa1, first we investigate the
effect of an increase of the external pressure in both small vessels (results in Fig. 4.16,
first column). Buckling of the small vessels choke the flow in the large upstream vessel
62
4.2. The structure equations
Figure 4.14: Schematic diagram of a generic bifurcation, where a larger vessel and two smaller vessels
converge in node NC . The arrows (here arbitrarily chosen) indicate the direction of the flux in each
vessel, which determines the sign in the Kirchhoff conservation equation.
(in absence of alternative pathways). Should the external pressure be further increased
in both small vessels, the plateau, both in flow and pressure values, would be reached
for higher po . Moreover, the plateau for the flow (pressure) would move in the south–
east (north–east) direction, with respect to the condition for pe = 15mmHg (data not
shown). Then, we investigate the effect of an increase of the external pressure in just
one of the small vessels (results in Fig. 4.16, second column). As the tubes assume dif-
ferent cross sections and resistances, the curves for the flow in the two branches do not
superpose. Diversion (“steal”) of flow toward the right branch occurs. A more marked
uneven flow repartition arises when the left branch enters into buckling instability. In
this condition, the flow in the left branch does not reach a plateau but decreases, while
flow in the right branch increases. Notice that in a Starling model, the fluxes of the
two small branches would increase along the same curve as long as buckling does not
occur, while a qualitatively analogous behaviour to the present one would be observed
after buckling (cf. the linear curves in Figs.3, 4 in [162]).
63
Chapter 4. Models of Blood Flow in a Distensible Network
Figure 4.15: Blood flow and blood pressure (in the node NC ) computed in the reference states for an
arterial bifurcation (first column) and venous bifurcation (second column) with different values of
the outlet pressure. Observe that the flow rate always increases and pressure in node NC decreases,
with a nonlinear trend, as the outlet pressure po decreases.
64
4.2. The structure equations
Figure 4.16: Test case Aaa1 (arterial bifurcation): effect of an increase of the external pressure in both
(first column) or just one (second column) of the daughter vessels. In the first case, the collapse of
the daughter vessels chokes the flow even in the father vessel, while in the second, diversion of flow
toward the right branch occurs due to the different transmural pressures in the two daughter tubes.
is so small that there is almost no flow. The second column of Fig. 4.18 shows a
behaviour analogous to the first column, with unequal fluxes in the small vessels due
to the different external loads. With the present inlet pressure conditions equal in both
small vessels, no flow diversion (counter-current flow with respect to the reference case)
takes place.
Assessment of the full network: comparison with experimental measures. We as-
sess the fact that the network we consider can compute physiologically coherent fluid-
65
Chapter 4. Models of Blood Flow in a Distensible Network
Figure 4.17: Test case Aaa2 (arterial bifurcation): effect of an increase of an equal amount of the
external pressure in the daughter branches in a system where the outlet pressure is fixed in the left
branch, while it varies in the range [20,10]mmHg in the right branch. When no buckling occurs (first
column), flow is diverted in advantage of the right branch which has a higher pressure drop. The
flow in the parent vessel is always increasing. When the external pressure is increased to 15mmHg
(second column), flow steal occurs till buckling of the right vessel, after which all the branches reach
a flow plateau.
dynamical fields. We compare the blood flow we obtain from model simulation with
the experimental measures in the human retina microcirculation performed by different
authors in the same diameter range. The external pressure is set to pe = 15 mmHg,
corresponding to the intraocular pressure of a healthy subject. The results are shown
in Fig. 4.19, which favorably compares volumetric blood fluxes. Blood velocities (not
66
4.2. The structure equations
Figure 4.18: Test case vvV1 (venous bifurcation): effect of an increase of the external pressure just in the
larger vessel (first column) or both in the larger vessel and in the left small vessel (second column).
In both cases, after an initial increase of flow occurring as long as po ≥ pe in the large vessel, the
flux goes to zero in all branches, since the the large vessel is entering buckling. In the third column,
the curves of the fluxes of the small vessels do not superpose due to the different external loads.
67
Chapter 4. Models of Blood Flow in a Distensible Network
Figure 4.19: Comparison of blood flow computed by the present model on a network generated by fractal
branching and experimental data measured in humans by different authors.
are obtained by partitioning the range between the minumin and maximum diameter
into 20 classes using log10 spacing. Class 1 corresponds to larger vessel, 20 to smaller
ones. For each class, we plot the bars whose heights represent the relative number
(with respect to the total number of arteries and veins, respectively) of vessels display-
ing an average pressure in the color–coded range. Colored continuous curves link the
height of the bars belonging to the same pressure range. Fig. 4.20 shows the histogram
plots for arteries, Fig. 4.21 for veins. Observe as, for increasing external pressure,
peaks for each pressure range shift towards smaller vessels (larger vessels in the venu-
lar network), thus indicating a global increase in blood pressure. Lower pressure ranges
progressively tend to disappear. This phenomenon is to be ascribed to the presence of
buckled vessels in veins. In the two bottom panels of Fig. 4.21 we also observe a much
more sharp compartmentalization of lower pressure ranges in class 1. Again, this phe-
nomenon is to be ascribed to the presence buckled vessels. This is also the cause of
the larger spreading in pressure ranges attained over the diameter classes for increasing
pe . These differences are more evident in the step pe =20 to 22 mmHg (pe =18 to 20
for the venous network) than in the previous steps. Elbow–like turns or “holes” in the
enveloping curves are instead due to the asymmetry of the network and the specific bin
classification of each vessel.
68
4.2. The structure equations
Figure 4.20: Histogram plots for relative frequencies of blood pressure in the arterial part of the net-
work. Colored continuous curves link the height of the bars belonging to the same pressure range.
a strong increase in resistance is observed in veins due to the combination of area vari-
ation and consequent high rise of blood viscosity, strongly diameter depended in that
region. Observe also how blood pressure and flow variations are instead similar, this
being connected to the continuity conditions enforced at the artero-venous interfaces.
This generates strong flux diversion and “flux stealing” from other vessels [36, 162],
localized in the four or five generations surrounding the sphere. Observe that, whilst the
no arterial vessel is actually included in the sphere, there exists a region of the arterial
network which is also affected by perturbations.
69
Chapter 4. Models of Blood Flow in a Distensible Network
Figure 4.21: Histogram plots for the relative frequencies of pressures in the venous part of the network.
Colored continuous curves link the height of the bars belonging to the same pressure range.
arterioles venules
mean blood pressure +6.46 % +7.10 %
flow rate -15.07 % -15.14 %
cross section area -15.20 % -55.47 %
conductance -31.84 % -91.64 %
Table 4.3: Maximal (with sign) percentage variations of the arteriolar and venular networks in the
specified vessel parameters subsequent to a local increase of external pressure in the spherical region
depicted in Fig. 4.22.
70
4.2. The structure equations
Figure 4.22: Effect of a local increase of external pressure. The gray shaded region represents the
sphere inside which the external pressure is set to pe = 20 mmHg, whereas pe elsewhere is equal to
15 mmHg. Colors code the net variation ∆p of blood pressure in the arterial and venous networks.
Insets on the right part show zoomed view of the affected network regions.
71
CHAPTER 5
Nitric oxide - mediated autoregulatory control
73
Chapter 5. Nitric oxide - mediated autoregulatory control
coupling this autoregulation model with a model similar to that discussed in Chap.3 for
O2 .
While in Chap.3 we represent the vessel wall as a simple transfer condition, in this
chapter the vessel wall is described as a three-layered structure with different charac-
teristic in each region. A radial solution for the solute transport, both for oxygen and
nitric oxide, along the vessel wall is obtained using Bessel-type function.
The O2 and NO concentration influence the level of calcium ions in the smooth muscle
connected to the actin-myosin motor, giving rise to muscle contraction. The new diam-
eter of the vessel lumen is then obtained from a mechanical balance of the forces acting
on the wall element, among which internal SMCs contraction and passive transmural
pressure.
Remark 4. The NO concentration and production in the intima strongly depends on the
oxygen level in the blood [81]. As a matter of fact, an increase of oxygen concentration
in blood causes an increase of NO concentration and then an increase of vessel radius.
This process seems to be in conflict with the mechanism of autoregulation, which modu-
lates the vessel radius to maintain a constant oxygen concentration in blood. Therefore
the hyperoxia condition should provoke a vessel reduction. In reality, the autoregu-
lation mechanism is very complex, and results from the delicate balanced action of
several vaso-dilator and vaso-constriction substances. While the dilation mechanism
is due to the NO, the constriction mechanism is mainly provoked by the endothelin-1
(ET-1) produced in the endothelial layer. These substances, reaching the SMCs layer,
are transduced by the ET-B and ET-A receptors which provoke the vessel constriction.
The ET-B receptors may be also present in the entotelial surface, where their stimula-
tion with the ET-1 results in release of NO. These mechanisms are very complex, and
deserve further future analysis.
Very few mathematical works in literature address the NO-induced SMCs relax-
ation till the consequent vascular response. In [216], Yang et al. propose a mathe-
matical model to investigate the role of NO in the regulation system of vascular SMCs
under various physiological and pathological conditions. They focus on two effects de-
scribed above, interaction of NO with cytosolic Ca2+ concentration, Ca2+ channel and
actin-myosin cycles. In [120] and [81], it is proposed a mathematical description of
biotransport of NO and of O2 in the entire vessel wall. These chemicals are inherently
interdependent, since nitric oxide synthase (NOS) isoforms require O2 to produce NO
and, furthermore, tissue O2 consumption is reversibly inhibited by NO. They simulated
steady-state radial NO and O2 gradients in the bloodstream, plasma layer, endothelium,
vascular wall, and surrounding tissue for different conditions. In [27], [118] and [171],
a mathematical study of myogenic and shear stress-dependent autoregulation in arterial
wall is proposed where it is computed the radial displacement under variations of blood
flow in vessel lumen. Unlike in [171] where Ca2+ and f concentrations are directly re-
lated to circumferential stress, in [27] and [118] the role of NO in the radius regulation
is also taken into account and the dynamic of this chemical is studied in a multilayer
structure of vessel wall (et and SMCs layers). References [11] and [50] propose a
mathematical model that considers, at the same time, the three types of autoregula-
tory mechanisms, described above, in the retinal and cerebral vasculature respectively.
They highlight the importance of the metabolic contribution since their models show
74
5.1. Oxygen and Nitric Oxide transport through vessel wall
that the myogenic mechanism does not provide enough variation in the vascular resis-
tance to support constancy of blood flow to tissue under variable perfusion pressure.
Unlike [11] where the auoregulation dynamics are described via use of empirical cor-
relations, in [50] interactions between NO and Ca2+ resulting in radial variation are
described in more detailed way.
Figure 5.1: Schematic representation of endothelium-derived smooth muscle relaxation via NO effect,
from Abbey Research Ltd.
75
Chapter 5. Nitric oxide - mediated autoregulatory control
model is given by the fact that the NO concentration and production strongly depends
on the oxygen level in the blood [81] and at the same time it inhibits the O2 consump-
tion in the vessel wall.
In the endothelial layer, molecules of O2 are consumed by the endothelial cells to syn-
thesize NO. The dynamic of oxygen partial pressure pO2 and nitric oxide concentration
cN O in endothelium can be described by the following kinetic equations [81]:
∂pO2 ,et 1 ∂ ∂pO2 ,et
αet = rDet,O2 αet − Ret,O2 − Qet,O2 (5.1)
∂t r ∂r ∂r
∂cN O,et 1 ∂ ∂cN O,et
= rDN O,et + Ret,N O − Qet,N O (5.2)
∂t r ∂r ∂r
where αet is the Bunsen solubility coefficient, Det,O2 and Det,N O are diffusion coeffi-
cients of O2 and NO, respectively. The consumption and production terms are defined
as:
where QM AX,et is the maximum O2 consumption rate. The term A takes into
cN O,et
account the O2 consumption inhibition by NO and is defined as A = (1 + ),
KN O
where the value of KN O = 27nM is determined from experimental data [28]. The
constant Ket is the M-M constant under no inhibition by the NO.
The SMCs layer is primarily responsible for the consumption of NO and O2 . In par-
ticular, the concentration of the latter is responsible for the control of the muscle tone.
The O2 and NO profiles across the SMCs layers are described by the following kinetic
equations [81]:
∂pO2 ,smc 1 ∂ ∂pO2 ,smc
αsmc = rDsmc,O2 αsmc − Qsmc,O2 , (5.3)
∂t r ∂r ∂r
∂cN O,smc 1 ∂ ∂cN O,smc
= rDN O,smc − Rsmc,N O , (5.4)
∂t r ∂r ∂r
where αsmc is the Bunsen solubility coefficient in SMCs layer and Dsmc,O2 and Dsmc,N O
are the diffusion coefficients of O2 and NO, respectively. The reaction terms are defined
as:
77
Chapter 5. Nitric oxide - mediated autoregulatory control
78
5.2. Smooth muscle cell contraction
Figure 5.3: Schematic representation of Ca2+ influx and outflux into the cytosol and the sarcoplasmic
reticulum (ER).
where Km is Michaelis Menten constant and nH = 2.5 is the Hill coefficient which
provides a measure of the cooperativity.
The solution of system (5.9) is a sum of two decaying exponential functions:
+ −
cCa2+ (t) = α+ e−t/τ + α− e−t/τ + c0 (5.12)
+ −
cs,Ca2+ (t) = αs+ e−t/τ + αs− e−t/τ + cs,ss (5.13)
±
where c0 and cs,ss are solutions of the steady state associated to system (5.9), τ is the
lifetime constant (see [63] for details).
79
Chapter 5. Nitric oxide - mediated autoregulatory control
d2 R
4m = Fint + Fext = −σθ 2πhLe + P0 2πRLe , (5.15)
dt2
where 4m = ρw 2πRLe h, Fint is the internal force proportional to the circumferential
stress σθ and Fext is the external force due to the blood pressure (pb ) and interstitial
pressure (pe ) loads. We set Fext = P0 , being P0 = pb − pe the transmural pressure, as
done in Sect.4.1. The stress σθ comprises
E R − R0
• passive elastic force: ,
1 − ν2 R0
which is analogous of that considered in the Sect.4.2.2, where R0 indicates the ra-
dius measured in the reference/unloaded configuration. The Young modulus E can be
80
5.4. Dimensional analysis: equation scaling
where Ret,smc and Rsmc,t are the inlet and outlet radius of the SMCs layer.
In addiction, the stress σθ also includes:
• active force due muscular tonus (active muscular stress): k2 F .
Using the hypothesis of incompressibility, i.e. R0 h0 = Rh, and of small deformations
R − R0
(R/R0 ≈ 1), Eq.(5.15) written in terms of the radial relative displacement η =
R0
reduces to:
d2 η dη h0 E(F ) h0 k2
ρw h0 R0 2
+ λh0 + 2
[η + κ1 η 2 ] = P0 − F. (5.17)
dt dt R0 (1 − ν ) R0
In Eq.(5.17), following [118], we also added a damping term λdη/dt and a non-linear
stress term k1 η 2 .
Observe that in this work, we have described changes in muscular tone F in response to
only chemical stimuli. A description of mechanical solicitation of SMCs is presented,
e.g., in [118, 170].
For convenience in the following discussion, Eq.(5.17) is rewritten as a system of two
order differential equations:
dη
= vη
dt
dvη λ κ0 (F0 + F ) P0 k2 (5.18)
2
=− vη − [η + κ1 η ] + − F.
dt ρw R0 ρw h0 F0 R0 ρw h0 R0 R02 ρw
pO2 ,i = p0 p∗O2 ,i , cN O,i = c0N O c∗N O,i , where i = et, smc, t (5.19)
81
Chapter 5. Nitric oxide - mediated autoregulatory control
and
cCa2+ = c0Ca2+ c∗Ca2+ , f = f 0f ∗, t = t0 t∗ , r = R0 r∗ (5.20)
where p0 , c0 , c0Ca2+ , f 0 , t0 and R0 are reference values reported in Tab.5.3.
After substituting (5.19)-(5.20) into model equations , we obtain:
∗
∂pO2 ,et Det,O2 t0 1 ∂ ∂p∗O2 ,et RN OM AX,et t0 p∗O2 ,et
∗
= r − +
∂t∗ R02 r∗ ∂r∗ ∂r∗ p0 αet p∗O2 ,et + Km,et /p0
QM AX,et t0 p∗O2 ,et
− ∗ ∗ 0
(5.21a)
P α p + K (1 + c )/p
0 et m,et
O ,et N O,et
2
∗
∂cN O,et ∂c∗N O,et RN OM AX,et t0 p∗O2 ,et
DN O,et t0 1 ∂
∗
∗
= 2 ∗ ∂r ∗
r ∗
+ 0 ∗ 0
+
∂t R 0 r ∂r C N O p O 2 ,et + K m,et /p
−kN O,et CN0 O P0 αet (c∗N O,et )2 p∗O2 ,et ,
(5.21b)
∂p∗O2 ,smc 0 ∂p∗O2 ,smc
D smc,O2 t 1 ∂ ∗
= r
∂t∗ R0 r∗ ∂r∗ ∂r∗
QM AX,smc t0 p∗O2 ,smc
− , (5.21c)
p0 αsmc p∗O2 ,smc + Km,smc (1 + c∗N O,et /p0 )
∂c∗N O,smc DN O,smc t0 1 ∂ ∂c∗N O,smc
∗
− kN O,smc CN0 O t0 (c∗N O,smc )2 ,
∗
= 2 ∗ ∗
r
∂t R0 r ∂r ∂r
(5.21d)
∗ 0 ∗
∂pO2 ,t ∂pO2 ,t
Dt,O2 t 1 ∂ ∗
= r
∂t∗ R0 r∗ ∂r∗ ∂r∗
0 p∗O2 ,t
QM AX,t t
− , (5.21e)
p0 αt p∗O2 ,t + Km,t (1 + c∗N O,et /p0 )
∂c∗N O,t DN O,t t0 1 ∂ ∂c∗N O,t
∗
− kN O,t c0N O t0 (c∗N O,t ),
∗
= 2 ∗ ∂r ∗
r
∂t R r ∂r
0
(5.21f)
∗ 0
∂cCa2+ ϕt
= −α1 t0 c∗Ca2+ − k1 CN0 O t0 c∗N O,smc c∗Ca2+ + 0 ,
∂t CCa2+
∂f ∗ 0 0
0 ∗ γCCa 2+ t
(c∗Ca2+ − 1)Θ[c∗Ca2+ − 1],
∗
= −α2 t f + 0
(5.21g)
∂t f
dη ρ R d2 η t0 κ0 (F0 + F ) P0 t0 k2 t0
∗ = − w 00 ∗ 2 − [η + κ1 η 2 ] + −
F. (5.21h)
dt λt d(t ) λh0 F0 λh0 R0 λ
In Tab.5.1, we show the numerical values of adimensional coefficients of equation
system (5.21). These values point out that the problem at hand is very stiff: time scales
of different processes differ from two to three orders of magnitude. Except for reaction
term connected to interaction between NO and O2 , diffusion processes are processes
faster then others.
82
Eq. Diffusion term Value Decay term Value Source term Value Inhibition term Value
R02 P0 αet
(5.21a) 1.97 20.1
DO2 ,et QM AX,et
P0 αet
0.134
RN OM AX,et
0
R02 1 CN O
(5.21b) 0.167 0
5.18·104 6.67·10−4
DN O,et kN O,et p0 CN O αet RN O,max
R02 αsmc P0
(5.21c) 1.97 20.1
DO2 ,smc QM AX,smc
R02 1
83
(5.21d) 0.167 0
2 ·102
DN O,smc kN O,smc CN O
1 CCa2+ ,th 1
(5.21g) 20 17.2 6.67·103
α1 ϕ0 k1 C 0
1 f0
(5.21g) 2 0
2.6
α2 γCCa 2+
Eq. Acceleration term Value Elastic Force term Value External Force term Value Active Force term Value
Table 5.1: Indicative numerical value of adimensional coefficient of equation system (5.21)
5.4. Dimensional analysis: equation scaling
Chapter 5. Nitric oxide - mediated autoregulatory control
In the following, we present the results of the numerical simulation of the mathemat-
ical model of the active autoregulation mechanism described above. First we consider
the case of "passive" vessel, where the phosporilation mechanism is assumed to have
no effect in the vessel regulation. Then we present the results of the complete model
and, eventually, we discuss the case of impaired vasodilator effect, which is reproduced
by turning off the effect of NO on calcium and myosin.
dη κ0 P0
+ [η + κ1 η 2 ] = . (5.22)
dt λh0 λh0
1 κ0 a a
η(t) = −( − tanh(atanh( ) + t ) ) (5.23)
2k1 a 2λh0 2κ1 κ0
where
q
a= κ20 + 4κ1 κ0 P0
setting:
r
4κ1 P0 25κ0
κ0 = , λ= ρw R0 .
3 3ρw h0 R0
Eq.(5.23) (see Fig.5.4) represents the transition of the system from one stationary state
1
(η = 0) to another (η∞ = ), reached after about half a second when the wall elastic-
2κ1
ity forces balance a constant transmural pressure. Characteristic parameter values used
in the simulations are given in Tab.5.3.
84
5.5. Simulation results
Figure 5.4: Evolution in time of η in the passive case when the vessel is loaded with P0 ' 37mmHg.
Observe that after about half a second, the solution reaches a stable equilibrium point.
1 3
η1 = and η2 = −
2k1 2k1
but η2 solution is not admissible since its value is negative and greater then one. If we
linearize Eq.(5.22) around the steady state η1 , we obtain:
dη κ0 κ1 κ0 2 P0
=− (1 + 2κ1 η1 )η + η1 + . (5.24)
dt λh0 λh0 λh0
We can observe that the eigenvalue associated with the homogeneous part is real and
negative and so η1 represents a stable equilibrium point.
In order to test the effect of hyperoxia on the radial displacement, we suppose that the
value of the O2 partial pressure at r = RL is represent by a function of time varying
from 30 to 70 mmHg in T second and having a trend in time (see Fig.5.5) given by:
8t
PO2 ,L = 50 + 20tanh( − 4).
T
85
Chapter 5. Nitric oxide - mediated autoregulatory control
Figure 5.5: Boundary condition relative to O2 partial pressure in vessel lumen as a function of time.
Initial condition
In order to define a reasonable initial value for numerical methods, we calculate the
stationary solution of the autoregulation problem by means of numerical and analytical
techniques.
First,we assume that the distribution of N O is known and we consider the general
steady state differential equation for the O2 concentration in layer Ωi , i = et, smc, t, of
the vessel wall. It can be written as:
∂pO2 ,i 1 ∂ 2 pO2 ,i
Di,O2 αi + Di,O2 αi − Ri,O2 − Qi,O2 (cN O,i ) = 0 (5.25)
∂r r ∂r2
where we have set Rsmc,O2 = Rt,O2 = 0. Observe that the Michaelis-Menten term
makes the equation non-linear. Since the analytical solution is not known, we rely on a
fixed point approach so that, at each iteration, we obtain an equation whose analytical
solution can be computed. Namely, we rewrite Eq.(5.25) as:
86
5.5. Simulation results
where, we have denoted by the subscript k the value of a variable at the k-th iteration
of the fixed point map and we have defined:
Z Z
k−1 1 k−1 1
pO2 ,i = p dΩi , cN O,i = cN O,i dΩi . (5.27)
|Ωi | Ωi O2 ,i |Ωi | Ωi
Multiplying Eq.(5.26) by r2 and dividing by Di,O2 αi , we obtain a Kelvin’s equation
of zero-th order, whose solution is a linear combination of Bessel functions [209]:
pO2 ,i = Ai I0,i (βi r) + Bi J0,i (βi r) (5.28)
where I0,i and J0,i are the modified Bessel function, and we have set
v !
u
u 1 R N OM AX,i Q M AX,i
βi = t + k−1 . (5.29)
DO2 αi pk−1 O2 ,i + K m,i p O2 ,i + K m,i (1 + c̄N O,i /c0N O )
where we have set RN OM AX,smc = RN OM AX,t = 0. The constants Ai and Bi are then
saturated by the enforcement of boundary and interface conditions.
We consider now the case of NO transport equations. We assume that the O2 con-
centration is a known function. The NO dynamics in the tree layers of the vessel wall
in steady state conditions is given by:
1 ∂ ∂cN O,et
rDN O,et + Ret,N O (5.30a)
r ∂r ∂r
−kN O,et (cN O,et )2 αet pO2 ,et = 0
1 ∂ ∂cN O,smc
rDN O,smc − kN O,smc c2N O,smc = 0 (5.30b)
r ∂r ∂r
1 ∂ ∂cN O,t
rDN O,t − kN O,t cN O,t = 0 (5.30c)
r ∂r ∂r
Observe that, as in the case of O2 concentration, also the NO transport equation is
a non-linear equation, whose solution is not analytically known. Again, we overcome
this obstacle using a fixed-point method. We rewrite the system of equation (5.30) as:
!
1 ∂ ∂ckN O,et RN OM AX,et pO2 ,et
rDN O,et + (5.31a)
r ∂r ∂r pO2 ,et + Km,et
!
∂ckN O,smc
1 ∂
rDN O,smc − kN O,smc ckN O,smc ck−1
N O,smc = 0 (5.31b)
r ∂r ∂r
!
k
1 ∂ ∂c N O,t
r ∂r rDN O,t ∂r − kN O,t ckN O,t = 0 (5.31c)
where in the first equation, relying on the observations obtained by the scaling of
Sect5.4, we have neglect the second decay term for cN O,et due to its very long time
scale. The solution of Eq.(5.31a) is:
−r2 RN OM AX,et pO2 ,et log(r)
cN O,et = + Cet + Det (5.32)
4DN O,et pO2 ,et + Km,et DN O,et
87
Chapter 5. Nitric oxide - mediated autoregulatory control
while the Eqs.(5.31b) and (5.31c) are, again, the zero-th order Kelvin equation. The
general solution is:
where I0,i and J0,i are the modified Bessel function, and
s s
KN O,smc ck−1
N O,smc KN O,t
βsmc = , βt = . (5.34)
DO2 ,smc DO2 ,t
Again the constants Ci and Di are saturated by the enforcement of boundary and inter-
face conditions.
Once the decoupled analytical solutions of the individual equations of the steady
state problem are known, we can set up an iterative procedure to find numerically the
solution of the full steady-state problem for O2 and NO. We then iterate the following
steps until convergence:
1) solve equations related to O2 pressure distribution in et, smc, t− layers consider-
ing a fixed NO concentration distribution (cN O ). An additional internal fixed point
iteration is necessary due to the non-linearity of the equations;
2) solve equations related to NO concentration distribution in et, smc− layers con-
sidering a fixed O2 pressure distribution (p̄O2 ) calculated at previous step. Now
a new distribution c̄N O is defined. An additional internal fixed point iteration is
necessary due to the non-linearity of the equations;
3) repeat step 1) with the new c̄N O value.
In Fig.5.6 we show the steady state solution of the distribution of O2 (top, in mmHg)
and NO (bottom, in nM) along the vessel wall (in µm). The characteristic parameter
values used in simulations along with their nomenclature and definition are given in
Tabs.5.2 - 5.3.
For the calcium and phosphorylated actin-myosin, the stationary solutions is sim-
ply obtained by setting the temporal derivative to zero and then solving the resulting
equation. The corresponding results are reported in Fig.5.7. Moreover, we suppose that
the reference state of the system represents an equilibrium state for the system itself.
Therefore we set
P0 R0
k2 = in Ωsmc ,
F 0 h0
dη
such that = 0.
dt
88
5.5. Simulation results
Figure 5.6: Steady state solution of the complete coupled system along the vessel wall. Top: equilibrium
distribution of the O2 . Bottom: equilibrium distribution of the NO.
89
Chapter 5. Nitric oxide - mediated autoregulatory control
Figure 5.7: Initial concentration of calcium (top) and phosphorilated myosin (bottom) in the SMCs
layer. The calcium and phosphorilated myosin increase along the SMCs layer due to the decrease of
NO along that layer (see Fig. 5.6).
90
5.5. Simulation results
part of the time slot due to the effect of diffusion and inhibition terms. Then the NO
concentration progressively decreases, while the O2 concentration increases, due to
the overcoming of the decay term on the other term. The Ca2+ and f concentration
decreases due to the high concentration of the NO, while in the last part of the time slot
the concentrations start to increase. Observe that in the second half of the time slot the
NO, Ca2+ and f concentrations in the SMCs are flat curves along the SMCs thickness.
In Fig.5.10 we show the vessel radius variation as a function of time. This result is in
accordance with that computed by the model of [118] for an arterial vessel.
ET-Inactivated Case
Figure 5.11: Schematic figure of chemical pathway in the arterial wall in healthy conditions (left) and
with endothelial dysfunction (right).
91
Chapter 5. Nitric oxide - mediated autoregulatory control
Figure 5.8: Oxygen (top) and nitric oxide (bottom) concentration computed with the full model
92
5.5. Simulation results
Figure 5.9: Calcium (top) and myosin (bottom) concentration in SMCs layer computed with the full
model
93
Chapter 5. Nitric oxide - mediated autoregulatory control
duced by ET-1 is determined on the receptor localization and the balance between ETA
and ETB receptors.
In our studies we try to reproduce these results neglecting the influence of nitric
oxide system on calcium and myosin.
Considering Eq.(5.9), the κL1 coefficient changes as follows:
κL1 = κL1,b .
Since κL1 is bigger than in the full model case, more vasoconstriction is produced.
In Fig.5.12, 5.13 we show the behavior of calcium and myosin concentrations, respec-
tively. The Ca2+ and f concentration increase due to the deactivation of the NO influ-
ence on the calcium channels. This increasing concentration leads to SMCs contraction
and then to vessel radius reduction (see Fig.5.14 ).
In Fig.5.15 we compare vessel dilation in the full case and the ET-inactivaed case.
In Fig.5.16 we also add, for comparison, the vessel dilation in the passive case. Observe
the high effect of the autoregulation mechanism on the vessel (observe the different or-
ders of magnitude for η obtained with the three models). The presence of the SMCs
layer makes the vessel wall more stiff and then leads to a less intensive variation of
vessel radius.
As shown from the results, we obtain a very weak variation of the vessel radius. This
suggests that the NO causes only a small part of the effect of the autoregulation. An
expansion of the model can be done, e.g., by including the effect of the ET-A and ET-B.
This issue needs further future analysis.
94
5.5. Simulation results
Figure 5.12: Calcium concentration in SMC layer computed with the ET-Inactivated model.
Figure 5.13: Myosin concentration in SMC layer computed with the ET-Inactivated model.
95
Chapter 5. Nitric oxide - mediated autoregulatory control
Figure 5.14: Time variation of the radial displacement computed with the ET-Inactivated model.
Figure 5.15: Comparison of radial displacements computed with the Full model and ET-Inactivated
Model
96
5.5. Simulation results
97
Chapter 5. Nitric oxide - mediated autoregulatory control
Name Definition
RL Radius of interface lumen-et
Ret,smc Radius of interface et-smc
Rsmc,t Radius of interface smc-tissue
RL External radius of vessel
Det,O2 Diiffusion coefficient for O2 in et layer
Dsmc,O2 Diffusion coefficient for O2 in SMCs layer
Dt,O2 Diffusion coefficient for O2 in tissue layer
Det,N O Diffusion coefficient for N O in et layer
Dsmc,N O Diffusion coefficient for NO in SMCs layer
Dt,N O Diffusion coefficient for NO in tissue layer
hN O,w NO permeability at external membrane of vessel wall
hO2 ,w O2 permeability at external membrane of vessel wall
αet O2 solubility in et layer
αsmc O2 solubility in SMCs layer
αt O2 solubility in tissue layer
RN OM AX,et Rate of N O production from O2
kN O,et First order rate constant of N O comsumption in et layer
kN O,smc First order rate constant of N O comsumption in SMCs layer
kN O,t First order rate constant of N O comsumption in tissue layer
Km,et Michaelis-Menten constant for et layer
Km,smc Michaelis-Menten constant for SMCs layer
Km,smc Michaelis-Menten constant for tissue layer
QM AX,et O2 consumption rate from et layer
QM AX,smc O2 consumption rate from SMCs layer
QM AX,t O2 consumption rate from tissue layer
α1 rate of Ca2+ ATPase pumping
α2 rate of myosin dephosphorylation
k1 NO-dependent Ca2+ ATPase pumping feed-back coeficient
k2 proportionality of SMCs tonus response to the f concentration
φ0 passive diffusion rate for Ca2+
CCAth threshold concentration of Ca2+
f0 reference concentration of phosphorylated myosin
E average Young’s elastic modulus of an arterial wall
σ Poisson’s ratio
ρw arterial wall density
P0 mean transmural blood pressure
λ viscous coefficient of arterial wall
κ0 elacticity coefficient of the relaxed arterial wall
κ1 nonlinear-elasticity coefficient of the arterial wall
98
5.5. Simulation results
99
Part II
101
CHAPTER 6
Overview of the eye retina anatomy
The retina is a complex matrix of neural cells lining the innermost surface of the eye
globe (see Figure 6.1a) between the sclera and the choroid. In adult humans, the retina
is approximately 60% of the eye globe of about 24mm in diameter, yielding a surface
of about 1090 mm2 . The purpose of the retina is to convert the light into neural signals.
Light enters in the retina from the lens, then travels through the thickness of the retina
before striking and activating the photoreceptors (rods and cones). Visual information
travels to the brain via the axons of the retinal ganglion cells, the output neurons of
the retina. From the retina, these informations is sent to various visual centers of the
brain through the fibers of the optic nerve. In vertebrate embryonic development, the
retina and the optic nerve originate as outgrowths of the developing brain, so the retina
is considered part of the central nervous system and is actually brain tissue. That is why
often their vasculature share very similar structures.
Histologically, the retinal tissue complex is composed of multiple layers, as reported
in Figure 6.1b.
These layers are usually collected into two different regions:
• outer retina (OR), proximal to the sclera, an avascular region that includes the
retinal pigmented epithelium, the outer and inner segments of the photoreceptors
layer and the outer nuclear layer;
• inner retina (IR), proximal to the vitreous, a vascularized region that goes from
the outer plexiform layer up to the nerve fiber /ganglion cell layer.
103
Chapter 6. Overview of the eye retina anatomy
Figure 6.1: a) Anatomical sketch of the eye; b) section of the retina through its thickness. The image
highlights the neural structures. Image from [35].
104
6.2. Experimental oxygen profiles
Figure 6.2: Subdivision into functional layers along the retinal thickness Correspondence between
anatomical structures and numbering of the layers (see also Fig.6.1): 1=retinal pigmented epithe-
lium and outer segments of the photoreceptor layer, 2=inner segments of the photoreceptor layer, 3
= outer nuclear layer, 4 = outer plexiform layer, 5 = inner nuclear layer and inner plexiform layer, 6
= ganglion cell/nerve fibre layer. The numbering of the layers follows the convention generally used
in modeling studies of O2 profiles in the retina, which starts from the choroid and goes towards the
vitreous [87, 127, 181].
within the inner nuclear layer (see Fig. 6.2) [8, 192]. The larger vessels lay in the
innermost layer, whereas two plexi of capillaries occupy the other two layers with pre-
capillary arterioles and postcapillary venules linking them to larger vessels. According
to the anatomical studies of the inner retina circulation, a fraction equal to 30% (resp.
60%) of the final branches of the arteriolar/venous network is connected via vertical
vessel to the intermediate (resp. deep) capillary plexus. The remaining 10% of arteri-
ole/venule pairs lie in the plane adjacent to the vitreous.
Capillaries exhibit a mesh-like structure with a characteristic size of 30-50µm [8]. The
capillaries either form a ”loop” shape if they are distributed in the same layer, or move
transversely to connect vessels in the other layers.
105
Chapter 6. Overview of the eye retina anatomy
observed between individuals: in the work of [2], the choroidal O2 tension measured in
two different macaques had a difference as high as 30 mmHg. Moreover, the presence
of a fovea in the retina of primates suggests that oxygenation in these species, even in
non-human type primates, is certain to be somewhat different from that of others, more
investigated, mammals [2].
In Fig. 6.3 we show the oxygen profile in the retinal thickness predicted by our
model for humans (the one discussed in Chap.7) compared against the profiles mea-
sured in rats by [45] and [122], in cats by [87] and in macaques by [23]. In order to
compare data, the x-axis is normalized with respect to the effective average thickness
of the retina of each specie. Similarly, data on the y-axis have been normalized with re-
spect to the maximum O2 tension, in such a way that the value at the choroid boundary
corresponds to 100% for each specie.
In general, the macaque retina is expected to be the most similar to the human retina.
However, a direct comparison with our model predictions is not straightforward (see
Fig.6.3). The discrepancies between the results of the present model and the profiles in
the macaque are to be found in:
i) the thickness of the outer retina layers. In the work of [2], data in the fovea of
the macaque are collected, which is an avascular region and could be, in princi-
ple, considered to be a good benchmark to calibrate parameters in the outer retina;
however, the macaque foveal region has a thickness which is significantly smaller
(172 ± 24 µm) than the one of the perifoveal vascularized region, and its particu-
larity makes it not completely suitable to represent the entire retinal tissue.
ii) like other central nervous structures, the primate retina has local variations in vas-
cularization that may reflect local variations in metabolism, rather than simply
tissue thickness or volume [192]. The local effect of the presence of vessels may
thus have a significant effect when directly measuring oxygen profiles. This is
especially true when these profiles are compared with the present model, which
performs an average of the O2 tension over each x − y cross–section of the retinal
tissue.
106
6.2. Experimental oxygen profiles
Figure 6.3: Comparison between O2 profiles predicted by our model (discussed in Chap.7) for the
human retina and those experimentally measured in rats by [45] and [122], in cats by [87] and in
macaques by [2] and [23]
107
CHAPTER 7
Coupled problem 1
The scientific community continues to accrue evidence that blood flow alterations and
ischemic conditions in the retina play an important role in the pathogenesis of ocular
diseases. For example, reduced retinal blood flow and retinal tissue hypoxia have been
associated with diabetic retinopathy [73], retinopathy of prematurity [210] and glau-
coma [140, 147, 183]. Many factors influence retinal hemodynamics and tissue oxy-
genation, including blood pressure, blood rheology, oxygen arterial permeability and
tissue metabolic demand. Even though the scientific community continues to accrue
evidence on the relationship between hemodynamic alterations, ischemic conditions
and ocular pathologies, the mechanisms governing this relationship, as well as the main
factors influencing it, still remain not completely understood (see, e.g., [32,211]), since
they are difficult to isolate in vivo, we propose here a novel mathematical and com-
putational model describing the coupling between blood flow mechanics and oxygen
(O2 ) transport in blood and throughout the retinal tissue. To the best of our knowl-
edge, this represents the first attempt to consistently couple a description of the retinal
microcirculation with the surrounding metabolically active tissue.
Despite the several existing mathematical and computational models on blood flow
and solute delivery in the circulatory system (see, e.g., [19, 62, 149, 203], to date, very
few theoretical models specifically address retinal microcirculation. In [129], a com-
putational study on a realistic retinal arterial network was performed to analyse blood
flow and oxygen profiles in the vessels using a 2D model based on the Navier-Stokes
equations. In this model, the O2 level in the tissue enters only as a boundary condition
for O2 delivery through arteriole walls and is assumed to be a given value not depending
on time and space. Various diffusion-reaction models have been proposed to describe
1 This chapter is part of the article "Blood flow mechanics and oxygen transport and delivery in the retinal microcirculation:
multiscale mathematical modeling and numerical simulation" by Causin, P., Guidoboni, G., Malgaroli, F., Sacco, R., Harris, A.
appeared on Biomechanics and modeling in mechanobiology, 15(3), 525-542(2016).
109
Chapter 7. Coupled problem
O2 profiles along the retinal tissue. However, blood flow was either absent [45, 208] or
prescribed as a given fixed value [181]. In [11], a model coupling blood flow and solute
(O2 and carbon dioxide) levels in the retinal microcirculation and tissue was developed
to theoretically investigate the mechanisms underlying retinal blood flow autoregula-
tion. However, in [11] a very simplified description of the vascular network based on
a representative segment model is adopted, including only “large” and “small” sets of
parallel pipes.
Our model extends these previous works by consistently coupling retinal blood flow
in a realistic vascular network and O2 transport in blood and tissue. Albeit in a simpli-
fied manner, the model accounts for the three-dimensional anatomical structure of the
retina, and comprises two main building blocks, depicted in Fig. 7.1: i) a model for the
retinal microcirculation, consisting in a Vascular Tree Network (VTN) formed by ar-
terioles, venules and capillaries organized in a graph. Capillary plexi, originating from
terminal arterioles and converging into smaller venules are embedded in two distinct tis-
sue layers. Arteriolar and venular networks are represented by fractal trees by a DLA
algorithm (see Sect.2.3), whereas capillary plexi are represented using a simplified
lumped description. Blood flowing in the VTN is modeled as a two-constituent mixture
of Newtonian incompressible fluids, plasma and red blood cells (RBCs) [129,189,218].
Oxygen is transported by blood and filters out from the vessel walls into the surround-
ing tissue, where the amount of permeating solute depends on the tissue metabolic
demand [159]. Blood pressure and velocity fields in the arteriolar/venular branches, as
well as solute transport, are described by 1D equations obtained by averaging the 3D
equations for mass conservation and momentum balance over each vessel cross sec-
tion. The abundance of capillaries prevents from applying a similar discrete approach
to the capillary plexi. Therefore, we define a simplified spatial arrangement of capillar-
ies consisting in the connection of each terminal arteriole and venule through a set of
parallel pipes; ii) a model for the retinal tissue, consisting in 6 layers characterized by
specific O2 consumption rates. The capillary plexi in layers 4 and 6 provide O2 sources
within the tissue.
The coupling between blocks i) and ii) forms the Fully Coupled Model (FCM),
which consists in a nonlinear system of partial differential equations, here solved using
numerical methods.
The model is validated against available experimental results to identify baseline
conditions. Then, a sensitivity analysis is performed to quantify the influence of blood
pressure, blood rheology, oxygen arterial permeability and tissue oxygen demand on
the O2 distribution within the blood vessels and in the tissue.
The chapter is organized as follow. A general overview of the model along with
the geometrical description is given in Sect. 7.1. The mathematical model is described
in Sect. 7.2, together with the numerical methods used in the simulations. Simulation
results are reported in Sect. 7.3 and discussed in Sect. 7.3.4, along with comments on
the limitations of the present model.
110
7.1. Model concept
Figure 7.1: Model concept: a) schematization of the retina: VTN graph model (arterioles and venules,
obtained from the DLA algorithm) embedded in the retinal thickness; b) subdivision into functional
layers along the retinal thickness. The retinal circulation nourishes the inner retina via the central
retinal artery (CRA) [86]. In humans, the CRA travels inside the optic nerve as it pierces the sclera,
and then it branches into arterioles that supply the so–called inner retina (layers 4 to 6 in the model).
The major arterial and venous branches and the successive divisions of the retinal vasculature lay in
the nerve fiber layer close to the internal limiting membrane. Vascular branches dive deep into the
retina, forming two distinct capillary plexi, a sclerad network that brackets the inner nuclear layer
(deep capillary plexus) and a vitread network that brackets the ganglion cell layer or lies within the
ganglion cell and nerve fiber layers (intermediate capillary plexus) [8, 192]. Retinal veins, which
follow a similar pattern, complete the vascular circuit through the retina by draining the capillary
plexi fed by the arterial system. The retinal veins drain into the central retinal vein (CRV). The outer
part of the retina (layers 1 to 3 in the model) is avascular and is mainly nourished by diffusion from
the choroid, which is not explicitly modelled here.
z direction. The z axis is oriented from the vitreous to the choroid (see Fig. 7.1). The
total retinal surface corresponds to the area of a circular disc whose diameter usually
ranges between 30 and 40 mm [158]. Here, we assume that the area of Ωt,xy , denoted
by |Ωt,xy |, amounts to 10.9 cm2 , corresponding to a disc diameter of 33 mm. The height
Lt of the parallelepipedon is set equal to 250 µm, corresponding to the average retinal
thickness in humans [3]. In the following, we provide details of the representation of
each component of the model.
The human retinal vascular network has a branching pattern which, at low resolution,
can be considered as a fractal structure. Experimental results have shown a complex de-
pendence of the estimated fractal dimension on a number of factors involved, including
diversity of subjects, image acquisition, type of image and processing, fractal analysis
methods and possible ocular diseases. A well-accepted average value of the estimated
fractal dimension of normal human retinal vascular network is approximately 1.7. This
is the same fractal dimension that is found for a Diffusion-Limited Aggregation (DLA)
111
Chapter 7. Coupled problem
process in two dimensions [214]. In this work, we exploit such a similarity to obtain
realizations of retinal vessel trees, shown in Fig.7.1, by generating DLA clusters via a
custom Matlab code, (see Sect. 2.3 for more details). The major branches representing
arterioles stemming from the cluster center are identified and assigned a radius that de-
creases from 54 to 10 µm moving away from the center. A smaller value in the range
from 15 to 5 µm is assigned to branches departing from the main vessels. The geomet-
rical structure of the venular network is analogous to that of the arteriolar network, the
only difference being in the radius of each venule which is assumed to be 35% larger
than that of the corresponding arteriole [199]. Each segment of networks has constant
radius R along its length L. The vessel lengths are assigned in such a way to obtain
obtain a corresponding surface on the planar network. Namely, we use the scaling rule:
1 lattice site= 0.162 cm. Extra segments are added to connect the arteriolar and venular
networks to the systemic circulation. These segments are located at the cluster center
and are oriented perpendicularly to the network plane. They represent the intraocular
portions of the CRA and CRV (see Fig. 7.1). In order to check the biophysical consis-
tency of the vessel area distribution in the network, the sum of the cross sectional areas
of the arterioles located within two disc diameters from the cluster center has been com-
puted and found to be of the order of 10−3 cm2 , which agrees with the value obtained
by [177] from experimental measures in humans within the same retinal region.
In the model, terminal arterioles can dig into either the intermediate or deep cap-
illary plexi, located in layers l = 4 and l = 6, or remain on the vitreous plane (see
Fig. 7.1(b)). Each terminal arteriole is paired with its corresponding terminal venule
in the self–similar networks. According to the anatomical studies of the inner retina
circulation, a fraction equal to 30% (resp. 60%) of the final branches of the VTN is
connected via vertical vessel segments to the intermediate (resp. deep) capillary plexus.
The remaining 10% of arteriole/venule pairs lie in the plane adjacent to the vitreous.
Terminal branches are randomly assigned to either the intermediate or deep capillary
plexi or the vitreous plane.
Capillary plexi
112
7.2. Mathematical model
Figure 7.2: Schematic representation of a capillary district joining the pre–capillary arteriole node A
with the post–capillary venule node V . The capillaries are represented as Nu equivalent pipes in
parallel.
113
Chapter 7. Coupled problem
following, pb denote the oxygen partial pressure (in mmHg), Qb the blood flux (in
cm3 /s) and HD the blood hematocrit (dimensionless).
Inlet and outlet boundary conditions network. Flux as well as pressure boundary con-
ditions can be enforced at the inlet and outlet of the arteriolar/venular tree in a suit-
able manner. In this work, pressure conditions are specified. An inlet value pb,in =
40 mmHg - estimated by considering hydrostatic and frictional pressure losses from
the aorta to the CRA - is assigned at the inlet of the segment coming from the cen-
ter of the arteriolar network representing the CRA output [79]. A blood pressure
pb,out = 15 mmHg, equivalent to the intraocular pressure (IOP) in normal condition,
is prescribed at the outlet of the segment exiting from the center of the venular network
representing the CRV input [11, 199]. An inlet standard hematocrit value HD,in = 0.45
is fixed at the entrance of the CRA.
Artero-capillary-venous (ArCVe) passage. Blood flow in each capillary district is stud-
ied according to the geometrical representation discussed in Sect. 7.1.1, where each
artero-venous junction, pertaining to either intermediate or deep capillary plexus, is de-
scribed by a system of Nu equivalent capillaries in parallel. We let scap ∈ (0, Lcap )
be the curvilinear abscissa defined on the capillary: scap = 0 corresponds to the arte-
riolar side of the capillary while scap = Lcap to the venous side. Equations analogous
to (3.12) hold in each capillary of the parallel system. As a consequence, the ArCVe
passage nodes are treated as “standard” junction nodes, where the following conditions
are enforced at junction j, for j = 1, . . . , Nl , with l = 4, 6:
Qb,j + Nu Qb,cap = 0, at nodes j = Ar, Ve
H Q + N H
D,j b,j u D,cap Qb,cap = 0, at nodes j = Ar, Ve
(7.1)
pb,A = pb,cap (scap = 0), at node Ar,
pb,V = pb,cap (scap = Lcap ), at node Ve,
where Ar and Ve represent arterial and venous sides of the passage node, respectively.
For each artero-venous junction that remains on the superficial plane without digging
into any of the two capillary plexi, we simply assume equations analogous to (3.12).
114
7.2. Mathematical model
Eventually, for each arteriolar terminal segment belonging to the fraction which ends
in the superficial plane, directly attached to the corresponding terminal venules, it holds
equations analogous to (3.31).
115
Chapter 7. Coupled problem
where O2 flux through the capillary walls is assumed to be proportional to the transmu-
ral O2 concentration drop, to the lateral surface of capillaries and to the permeability
coefficient of the wall. Upon multiplication by the number of capillaries in each dis-
trict, the resulting term Sj represents the oxygen flux through the j-th capillary district,
where pt is an average value of O2 tension in the district. The total O2 source (O2 vol-
ume rate per unit volume, in [mlO2 /cm3 /s]) associated with layers l = 4, 6 is obtained
by adding all the district contributions in that layer and dividing by the layer volume
1 PNl
|Ωt,l |, leading to S = Sj .
|Ωt,l | j=1
Collecting consumption and source terms yields the following expression for the reac-
tion term in the tissue
0 l = 1, 3,
Q(pt , pc ) = −C(pt ) l = 2, 5, (7.6)
S(pt , pc ) − C(pt ) l = 4, 6.
Model reduction of the problem for O2 in tissue. Let us average Eqn. (7.3) in the x − y
plane. To this end, we introduce the quantity
Z
1
pt (z, t) = pt (x, y, z, t) dxdy, (7.7)
|Ωt,xy | Ωt,xy
representing the average O2 partial pressure over the x − y plane. Recalling the bound-
ary conditions on the lateral surface of the domain, Eqn. (7.3) becomes
∂pt ∂ 2p
αt − Dt αt 2t = Q(pt , pc ), l = 1, . . . , 6. (7.8)
∂t ∂z
The nonlinear term in the reaction rate is treated in a similar manner as done for the
treatment of O2 saturation in blood The numerical simulations reported later are carried
out in steady state conditions and for metabolic rates which correspond to the light–
adapted condition as defined in [208].
116
7.3. Simulation results
elements in primal-mixed form on the network curvilinear axial coordinate are used to
discretize the PDE model for VTN-blood and O2 [182].
The PDE system for O2 , in the physiological range of parameters, is strongly dom-
inated by convection, and this leads to a large local Péclet number if the spatial dis-
cretization parameter is not extremely small. An exponential fitting approach [15] is
used to guarantee numerical stability even in the limit of a vanishing diffusion coef-
ficient. The overall algorithm is depicted in the block diagram of Fig. 7.3. All the
simulations are performed using an in-house developed Matlab code. The reference
(baseline) values of the parameters used in all the simulations are reported in Tab. 7.1.
The VTN geometry is generated via the DLA algorithm discussed in Sect. 2.3.
Symbol Value Unit Description Reference
|Ωt,xy | 10.9 cm2 retinal surface [158]
Lt 25·10−3 cm retinal tissue thickness estimated
N4 235 - # vessels digging in tissue layer 4 estimated
N8 118 - # vessels digging in tissue layer 6 estimated
Lcap 5·10−2 cm capillary length estimated
Rcap 3·10−4 cm capillary radius estimated
Slat 3π · 10−5 cm2 capillary lateral surface estimated
|Ω|t,l 2.725·10−2 cm3 volume of l–th tissue layer estimated
VTN - fluid
HD,in 45 % hematocrit [101]
µp 11·10−2 g/cm/s viscosity of plasma [68]
pb,in 40 mmHg arteriole inlet blood partial pressure estimated
pb,out 15 mmHg venule outlet blood pressure estimated
VTN - O2
DHb 1.5·10−7 cm2 /s diffusivity of Hb in RBCs [141]
Dc 2.18·10−5 cm2 /s diffusivity of free O2 in plasma [125]
CHb 0.2 ml O2 /ml O2 carrying capability of Hb in blood [141]
α1 3.38·10−5 ml O2 /ml/mmHg O2 solubility coefficient in RBCs [75]
α2 2.82·10−5 ml O2 /ml/mmHg O2 solubility coefficient in plasma [75]
p50 26.6 mmHg half saturation constant of O2 saturation in Hb [141]
nHill 2.7 - Hill exponent [141]
Dw 1·10−5 cm2 /s diffusivity of free O2 in vessel wall estimated
tcap 5·10−5 cm capillary wall thickness estimated
pc,in 100 mmHg arteriole inlet O2 partial pressure estimated
TISSUE - O2
αt 2.4·10−5 ml O2 /ml/mmHg O2 solubility coefficient in tissue [128]
Dt 1·10−5 cm2 /s diffusivity of free O2 in tissue [157]
Qmax
OR 90 mmHg/s maximal rate of O2 consumption in the outer retina [181]
Qmax
IR 26 mmHg/s maximal rate of O2 consumption in the inner retina [181]
K1/2 2 mmHg partial pressure of O2 at half maximal consumption [70]
pt,choroid 80 mmHg O2 partial pressure at choroid level [128]
pt,vitreous 20 mmHg O2 partial pressure at vitreous level [5]
117
Chapter 7. Coupled problem
Start
VTN - fluid
(0) (0)
HD , pb
Solve for HD and pb
by algorithm1
(fixed point procedure)
VTN fluid field
HD , pb
k=0
VTN - O2
(k) (k)
pc , pt
Solve for pc field
(fixed point procedure)
(k+1)
pc
VTN- O2 pressure
tissue O2 pressure
TISSUE - O2
(k+1) (k)
pc , pt
Solve for pt fields
(fixed point procedure)
(k+1)
pt
check convergence
End
Figure 7.3: Flow chart of the computational algorithm used to numerically solve the FCM. First, we
compute the blood flow field, with internal iterations (represented in the box on the right) required
to solve for the coupling between hematocrit and blood viscosity. Then, we compute the O2 partial
pressure field in the VTN and tissue, by an outer fixed point iteration (light blue box); internal it-
erations are required to address nonlinearities in each sub–problem (saturation terms in the VTN,
Michaelis-Menten consumption terms in the tissue, boxes on the right). The overall convergence of
the algorithm is measured by checking the relative infinite norm of the difference of tissue pressure
between successive iterative steps.
118
7.3. Simulation results
VTN - fluid
πR4
∂Qb ∂(HD Qb ) ∂pb Q
= 0, = 0, Qb = − , vs = ,
∂s ∂s 8µb ∂s A
pb = pb,in at CRA, pb = pb,out at CRV, HD = HD,in at CRA.
Junction conditions
Artero-venous bifurcations
X X
pb,f = pb,d1 = pb,d2 , Qb,j = 0, HD,j Qb,j = 0.
j=f,d1 ,d2 j=f,d1 ,d2
Table 7.2: Summary of the model equations for blood flow along the VTN.
the results obtained with our model with those obtained with the model [199], we
adopt the same simplifying assumption of constant hematocrit as in [199]. We
also remark that both our model and the one in [199] are solved using the simpli-
fied expression for blood viscosity given in [88]. In [199], a recursive formula is
utilized (Eqs. (3),(4) of [199]) based on a prescribed inlet blood flow to compute
flow and velocity in each network generation, and then pressure values are post–
calculated from the flow. The solution procedure of our model is different. First,
blood pressures are computed by solving a linear system of algebraic equations,
and then flow and velocity are post–calculated from the pressures. The results of
the comparison between the two models are reported in Fig. 7.4a),b), where blood
velocity and pressure are represented along the tree generations. The results show
that our numerical solver correctly reproduce Figs.2A,B in [199].
• Blood velocity and flow in the VTN. Fig. 7.5 shows the distribution of the mean ve-
locity and flow rate as a a function of the vessel diameter in the arteriolar and venu-
lar trees. The model predicted values of blood velocity and flow along the network
are compared with clinical measurements in humans reported by [177], [212]
and [77] Fig. 7.5 shows that the model predicted velocities and flows are in good
agreement with experimental data.
• Oxygen distribution in the VTN. Fig. 7.6 shows the O2 saturation in the VTN,
organized in 4 bins for both arterioles and venules. The diameter range corre-
sponding to each bin is reported in the figure. Numerical data are compared with
119
Chapter 7. Coupled problem
VTN-O2
∂Js 2 ∂Jw 2
+ αc Lp pc fc (R) = − + αc Lp βt pt,w fc (R),
∂s R ∂s R
∂ D Hb
Js = v s ωc (αc + CHb HD SeO2 )pc − Dc p (αc + CHb HD SeO2 ) ,
∂s c Dc
Jw = v s (1 − ωc )(αc + CHb HD SeO2 )pt,w ,
pc = pc,in at CRA,
∂ DHb
−Dc p (αc + CHb HD SO2 ) es · nout = 0 at CRV.
e
∂s c Dc
Junction conditions
Artero-venous bifurcations
X
pc,f = pc,d1 = pc,d2 , Aj Js,j = 0.
f,d1 ,d2
Table 7.3: Summary of the model equations for O2 transport and delivery along the VTN.
available clinical measurements in humans. Fig. 7.6 shows that the model pre-
dicted O2 saturation is within the same range as the corresponding measurements
for every considered diameter range. Data are clinically obtained in an annulus
around the optic disk. The exact definition of the investigation area may slightly
vary from author to author. Here, we consider an annulus with inner and outer ra-
dius of 1 and 1.5 optic disc diameters, where for this latter we consider the average
value between the mean vertical and horizontal diameter (1.88 and 1.77 mm, re-
spectively [169]). Fig. 7.6 shows that the model predicted O2 saturation is within
the same range as the corresponding measurement for every considered diameter
range. The model under-predicts venous oxygen saturation (Fig. 7.6). This might
be ascribed to the simplified approach we adopted to describe the delicate phe-
nomena occurring in the capillaries or to the averaging procedure we utilized to
describe the phenomena occurring in the retinal tissue in the xy plane.
• Oxygen profile in the retinal tissue under acute retinal occlusion. The model pre-
dicted O2 profile through the depth of the retinal tissue is more difficult to validate.
This is due to the fact that direct reliable measurements of retinal O2 profiles are
available only under specific conditions and direct estimates of metabolic con-
120
7.3. Simulation results
TISSUE-O2
∂pt ∂ 2p
αt − Dt αt 2t = Q(pt )
∂t ∂z
0 l = 1, 3,
Q(pt ) = −C(pt ) l = 2, 5,
S(p , p ) − C(p ) l = 4, 6
t c t
Qmax
l pt
C(pt ) = αt
pt + K1/2
1 PNl
S(pc , pt ) = (Nu Lcap Slat αt (βt−1 pc,j − pt ).)
|Ωt,l | j=1
pt = pt,ch at z = 0, pt = pt,vitreous at z = Lt
Junction conditions
∂pt,j ∂pt,j+1
pt,j = pt,j+1 , −Dj = −Dj+1 .
∂z ∂z
Table 7.4: Summary of the model equations for O2 transport and delivery in the retinal tissue.
sumption rates are currently not available. We validate the description of the O2
profile in the tissue under acute retinal occlusion. This phenomenon is mathe-
matically reproduced by artificially blocking the delivery of O2 through retinal
microcirculation so that the retinal oxygenation relies only on the choroidal O2
supply. Setting the O2 consumption rates as in Tab.7.1, we calibrate the thick-
ness of the metabolically active layers in the outer retina in order to obtain a good
agreement between the O2 profiles (i) predicted by our model, (ii) experimentally
measured in cats by [4], and (iii) simulated in humans by [181]. The comparison
is reported in Fig. 7.7. The thicknesses of the metabolically active layers resulting
from these calibration are in good agreement with those of [208].
This validation, performed in the case of retinal arterial occlusion, shows that
the O2 profile computed for humans in the modelling work by [181] is in good
agreement with the profile predicted by our model when (i) thicknesses measured
in the human outer retina by [9] are used, and (ii) the outer retina is assumed to
occupy half of the total retinal thickness. Thus, we can conclude that the model
description of O2 transport and delivery along the retinal tissue is consistent with
the currently available data.
The mathematical model is then used to simulate the distribution of bio–physical vari-
ables of clinical interest in the vascular tree which are not easy (or possible) to ob-
tain experimentally. The experimental/direct evaluation of such distributions in animal
models as well as in human patients is still very challenging. Specifically, we present
here the model predicted distributions of blood pressure, blood hematocrit and viscosity
121
Chapter 7. Coupled problem
Figure 7.4: Validation of the numerical code. Solution of blood flow on the structured dichotomic tree
proposed by [199]: a)blood velocity; b) blood pressure. Both quantities are plotted against vessel
diameters. These results correctly reproduce those reported in Figs.2A,B of [199].
122
7.3. Simulation results
Figure 7.5: Distribution of the mean velocity in the arteriolar (a) and venular (b) trees and distribution of
the flow rate in the arteriolar (c) and venular (d) trees as a function of the vessel diameter compared
with experimental data in humans by [177], [212] and [77].
for larger outlets. [169] estimated that the pressure drop from the disc to a vessel of 30-
40 microns in diameter is of the order of 15 mmHg. [10] estimated a pressure drop of
the order of 16 mmHg in arterioles and of 4 mmHg in venules. The results obtained
with the present model, namely pressure drops of approximately 15 mmHg in arterioles
and 4 mmHg in venules, are thus in quantitative agreement with those reported in the
literature. The distributions of blood hematocrit and viscosity in the VTN predicted by
our model show trends similar to those reported in [68], but with a more pronounced
spreading of the data. The difference in data spreading could be due to the topology of
the vascular network and the portion of the vascular tree used to display the results.
123
Chapter 7. Coupled problem
Figure 7.6: Oxygen saturation in the VTN as a function of the vessel diameter for arterioles (a) and
venules (b) as predicted by our model and clinically measured by [18], [187], [85] and [72]. The
height of the bars indicates average of the data, error bars indicate standard deviation.
model outputs.
124
7.3. Simulation results
Figure 7.7: Retinal tissue O2 profiles under acute retinal occlusion conditions as predicted by our model,
experimentally measured obtained from the model and from experimental data by [4] and simulated
by [181].
125
Chapter 7. Coupled problem
Diameter ranges
Arterioles 1 2 3 4 5
Pb [mmHg] 20.79 ± 0.19 23.53 ± 3.72 26.71 ± 1.82 31.33 ± 1.67 35.81 ± 1.24
HD 0.23 ± 0.10 0.45 ± 0.10 0.53 ± 0.06 0.51 ± 0.03 0.48 ± 0.02
µb 103 [Pa s] 2.93 ± 0.89 2.94 ± 0.70 2.69 ± 0.22 2.63 ± 0.08 2.58 ± 0.04
Venules 1 2 3 4 5
Pb [mmHg] 20.77 ± 0.19 20.01 ± 1.13 19.23 ± 0.58 17.77 ± 0.54 16.34 ± 0.39
HD 0.37 ± 0.13 0.47 ± 0.09 0.52 ± 0.04 0.49 ± 0.02 0.46 ± 0.01
3
µb 10 [Pa s] 2.43 ± 0.65 2.56 ± 0.52 2.8 ± 0.23 2.79 ± 0.09 2.73 ± 0.04
Table 7.5: Mean values and standard deviations of blood pressure Pb , hematocrit HD and blood
viscosity µb along the VTN, binned in 5 diameter ranges. Arterioles: 1: 0 ≤ D < 22, 2 :
22 ≤ D < 43, 3 : 43 ≤ D < 65, 4 : 65 ≤ D < 86, 5 : 86 ≤ D < 108; venules:
0 ≤ D < 2, 2 : 29 ≤ D < 58, 3 : 58 ≤ D < 87, 4 : 87 ≤ D < 117, 5 : 117 ≤ D < 146. The
vessel diameter D is expressed in µm.
Changes in inlet arterial blood pressure (pb,in ). Arterial blood pressure has a strong
influence on all the output variables. These results are supported by clinical observa-
tions. Low blood pressure (i.e. arterial hypotension) is associated with low perfusion
of ocular tissues [32, 161] and is identified as a risk factor for many ocular diseases,
including glaucoma [38, 44, 52, 92] and nonarteritic anterior ischemic optic neuropathy
(NAION) [89, 107]. High blood pressure (i.e. arterial hypertension) is also an impor-
tant clinical condition and has been identified as a risk factor for retinal artery and vein
occlusion and macular edema [22, 108, 112, 134], diabetic retinopathy [22, 215] and
NAION [17]. Our model predicts that a sudden increase in arterial pressure leads to an
increase in blood flow and, consequently, an increase of the O2 levels within the vessels
and inside the retinal tissue. These results are consistent with the experimental obser-
vations in cats [146], where acute severe elevation in systemic blood pressure resulted
in sudden increases in blood velocity and retinal blood flow.
Changes in plasma viscosity (µp ). Plasma viscosity has a strong influence on the O2
tension at the level of the retinal ganglion cells, pt -RGC. More precisely, our model
predicts that an increase in µp from 0.6 cP to 3 cP, would lead to a decrease in blood
flow and, consequently, to a decrease in pt -RGC from 16.3 mmHg to 13.8 mmHg.
Thus, our model suggests that when µp is abnormally high, the O2 delivery to the
retinal ganglion cells might become dangerously low. These findings support clinical
studies reporting elevated blood viscosity as a risk factor for retinal artery and vein
occlusion [112], glaucoma [137] and diabetic retinopathy [154].
Changes in the Hill’s law parameters (p50 , nHill ), inlet blood O2 level (pc,in ), choroid
O2 level (pt,choroid ). Several clinical and experimental studies have shown that the
retinal vasculature is extremely responsive to changes in systemic O2 levels [161].
When breathing 100% O2 (hyperoxia), retinal arterioles are observed to markedly con-
strict [55, 95, 103, 178, 179, 195], leading to a decrease in blood flow [131]. In contrast,
a decrease in O2 tension (hypoxia) induces a dilation of the retinal arterioles and an
increase in blood flow [55, 161]. Several model parameters are altered when breath-
ing air with different O2 levels. With increases in the systemic blood oxygenation, O2
126
7.3. Simulation results
Qmax
IR [mmHg/s] fA [-]
-50% 100% -100% -20%
O2 -wall [mmHg] +1.66 -0.34 +6.67 +3.98
SO2 [%] +37.22 -16.02 +0.05 +0.07
pt -RGC [mmHg] +0.79 -0.10 +186.28 +10.55
Table 7.6: Sensitivity analysis. Changes in O2 transmural tension (O2 -wall), mean O2 saturation in
arterioles (SO2 ) and O2 tension in the retinal ganglion cells (pt -RGC), due to changes in blood
pressure in the VTN at the inlet of the arterial network, pb,in , plasma viscosity, µp , constant of O2
tension in the Hill’s law, p50 , exponent in the Hill’s law, nHill ,O2 tension in the VTN at the inlet of the
arterial network, pc,in , O2 tension at the choroid, pt,choroid , maximal O2 consumption rate, Qmax IR
and arterial permeability, Lp =fA Lp,ref . The first row of the table reports the percentage variation
of each model parameter. The influence of individual parameter variations on the model outputs is
measured by the relative condition number r. The absolute value of r is defined as |r| = kδxk/kxk kδdk/kdk
where d is the baseline parameter value and x the related baseline output, whereas δd and δx are
the variations in the parameter and in the output, respectively. The sign of the condition number
indicates whether the variable decreases (-) or increases (+) with respect to the reference value. The
color scale white-yellow-red used in the table visually grades the influence of individual parameter
variations on the model outputs (white=small, red=large variations). Baseline parameter values are:
Pb,in = 40 mmHg, µp = 1 cP, p50 = 26.6 mmHg, nHill = 2.7, pc,in = 100 mmHg, pt,choroid = 80
mmHg, QmaxIR = 26 mmHg/s, fA = 1, O2 -wall=41.17 mmHg, SO2 =68.39 %, pt -RGC=15.03 mmHg.
127
Chapter 7. Coupled problem
Figure 7.8: Box–plot representation of blood pressure (top row), viscosity (middle row) and hematocrit
(bottom row) in arteries and veins.
levels rise in the arterial blood (pc,in ) and in the choroid (pt,choroid ). Moreover, during
hyperoxia, hyperoxygenated hemoglobin carries less carbon dioxide from the tissues in
the carbamino form, a phenomenon also known as Haldane effect [20, 42, 49], leading
to a decrease in the half–saturation constant for O2 tension in the Hill’s law (p50 ). Thus,
we have varied pc,in , pt,choroid , p50 and nHill independently in our model to single out
their contribution to the retinal oxygenation. Our model predicts that the O2 level in
the retinal ganglion cells is mildly influenced by alterations in pc,in , pt,choroid , p50 and
nHill . The model predicts that changes in pt,choroid mostly affect the O2 levels in the
outer retina. However, changes in pc,in , p50 and nHill strongly affect the O2 transmu-
ral tension, O2 -wall, which, in turn might lead to remarkable blood flow changes via
vasoactive diameter alterations. Even though our model does not currently account for
vessel vasoactivity, these results suggest that changes in O2 -wall should be included in
a model for vascular autoregulation simulating hyperoxic and hypoxic conditions.
128
7.3. Simulation results
Figure 7.9: Model predicted O2 profiles in the retinal tissue due to: a) changes in inner retina consump-
tion rate (Qmax max
IR ); b) changes in outer retina consumption rate (QOR ).
rameter QmaxIR represents the maximum consumption rate of the inner retinal layers,
and, in particular, of the retinal ganglion cells. Our model predicts that a variation of
Qmax
IR results in a remarkable change in the O2 tension at the level of the retinal ganglion
cells, but only slight changes in arterial O2 saturation and O2 transmural tension. The
metabolic needs of the inner retina, and therefore QmaxIR , can be altered experimentally
via flicker light stimulation, where the retinal neuronal activity is increased by expo-
sure to a series of intermittent light flashes occurring at a certain rate [211]. Several
studies have shown that the vasodilation response to flicker stimulation is impaired in
pathological conditions, including glaucoma [172], diabetic retinopathy [84] and va-
sospastic propensity [78]. Our model suggests that the observed vasodilation in the
retinal arterioles might be due to an O2 -mediated response [10, 161].
129
Chapter 7. Coupled problem
face between the retinal tissue and the vitreous. As a consequence, the O2 level in the
tissue might remain higher than the O2 level in the vasculature, possibly leading to lo-
cal reverted O2 fluxes from the tissue to the vasculature [217]. It is also important to
emphasize that we are using a simplified model for the retinal tissue, where the diffu-
sion/reaction problem for O2 transport is occurring only in the direction of the retinal
depth, whereas inhomogeneities in the transversal plane have been neglected. Indeed,
many questions regarding the physiology and the modeling of arterial permeability to
O2 are still open. This is a very challenging problem which constitutes an interesting
future direction of research.
130
7.3. Simulation results
changes in local and systemic conditions, while meeting the metabolic needs of the tis-
sue. Dysfunctional autoregulation has been recognized as an important factor contribut-
ing to the pathogenesis of several ocular diseases, see, e.g., the discussions in [140]
and [211]. In [11], a first relevant model has been presented, in which a simplified
retinal vascular network is capable of autoregulating on the basis of phenomenological
relations connecting tone and diameter of retinal arterioles with metabolic conditions.
Linear fits derived from the compartmental model of [217] are used to couple the levels
of carbon dioxide in the vasculature and in the tissue. In [37], a biophysically moti-
vated model of an actively autoregulating blood vessel has been proposed, including
the wall dynamics of O2 and nitric oxide and their effect on calcium within smooth
muscle cells of the arterial wall. Its application has been tested on a single retinal arte-
riole. It would be extremely interesting, but definitely not trivial, to extend the present
model to include changes in vessel diameters due to autoregulatory mechanisms using,
for example, the approaches in [11] and [37].
131
CHAPTER 8
Study of solute transport through the retinal tissue
133
Chapter 8. Study of solute transport through the retinal tissue
OR IR Consumption Term
Model Blood Source
# layers # layers Constant MM
[87] 1/2/3 X X
[45] 3X 5X X
[180] 3X 1X X X
[12] 3X 5X X X
Table 8.1: Schematic prospect of existing mathematical models in recent literature on oxygen profile in
the retinal tissue. Symbols: IR= inner retina, OR= outer retina, MM = Michaelis-Menten. Numbers
in columns OR and IR indicate the number of layer considered in the corresponding domains.
prescribed blood flow and arterial O2 values from arterial microcirculation are used to
model sources in the diffusion-reaction PDE system describing the O2 profile. In [180],
a numerical solution of the problem is pursued and the possible effects of retinal de-
tachment are analysed.
In all the above cited models, the choroidal supply is mathematically represented by a
Dirichlet boundary condition.
In Tab.8.1 we presents at the best of our knowledge, a schematic survey of the existing
most recent mathematical models for oxygen profile in the retinal tissue.
In this chapter, we propose mathematical models which describe the oxygen pres-
sure profile along the whole retinal depth, including sources from blood circulation and
tissue metabolic consumption. The oxygen transport along the retinal depth is described
by a 3D-diffusion-reaction equation, where both constant and non-linear (Michaelis-
menten type, in particular) reaction terms are considered. The mathematical models,
described in the following, differ from the different geometrical model settings consid-
ered based on the tissue metabolic characteristics. All the models consider a one-layer
structure for the inner retina, with a blood source uniformly distributed, assuming dif-
ferent numbers of layers in the outer retina (from 1 to 3 layers).
∂ 2 pt
− Dt αt = Q(pt , pc ), (8.1)
∂z 2
where pt is the average O2 partial pressure over the x − y plane, pc the capillary oxygen
pressure.
In Eq.(8.1), the reaction term Q is constituted by two different components: the
sources (S) of O2 , provided by the capillary plexi and the consumptions (C) of O2 , due
to the metabolic activity of neural cells in the tissue so that Q = S −C. The inclusion of
a blood source in a mathematical model for the retinal oxygen profile is limited to very
few work (see Tab.8.1). In these works the oxygen source is described as a function
of constant parameters, which are blood flow, free and bound oxygen concentration in
134
8.1. Model of O2 delivery to tissue: analytical approach.
blood.
As for the consumption term, several authors (see again Tab.8.1) propose mathematical
models with a constant consumption term. With this latter choice, Eq.(8.1) naturally
turns out to be linear with solution:
C−S 2
pt = z + βz + γ (8.2)
2Dt αt
where the values of β and γ are determined by the imposition of the boundary con-
ditions. It is obvious that the knowledge of an analytical solution allow to perform a
mathematical analysis of such solution, e.g. a sensitivity study.
An alternative to the constant consumption rate, is the Michaelis-Menten kinetics.
However this choice leads Eq.(8.1) to be non-linear and its analytical solution is un-
known. In the following we apply the analytical strategy of the homotopy perturbation
method to obtain a semi-analytical solution of Eq.(8.1) with Michaelis-Menten kinet-
ics [91, 188].
8.1.1 Homotopy perturbation method for tissue oxygen transport with Michaelis-
Menten kinetics
The homotopy perturbation method (HPM) is a semi-analytical technique for solving
nonlinear partial differential equations. This method is based on the topological concept
of the homotopy from topology to generate a convergent series solution. For further de-
tails, see [13, 90, 126] and the work of [188] for an application similar to the one of this
work. The goal of this section is the derivation of an approximate analytical solution
by the homotopy perturbation method of the following steady state one-dimensional
diffusion-reaction problem:
Find u(z) such that
d2 u u(z)
−D 2
(z) = −γ , −L ≤ z ≤ L , (8.3)
dz u(z) + k
equipped with Dirichlet boundary conditions (see Fig.8.1)
0
u1 u2
−L z L
z u(z) γL2 k
zb = , u
b(b
z) = , γ
b= , k=
b . (8.5)
L pref Dpref pref
135
Chapter 8. Study of solute transport through the retinal tissue
where pref is a reference value for the oxygen pressure variable. Substituting the above
quantities in Eqs(8.3) and (8.4), we obtain the following dimensionless equation:
find u
b(b
z ) such that
d2 u u
b(b
z)
, −1 ≤ zb ≤ 1 ,
b
2
(b
z) = γ
b (8.6)
db
z u
b(b
z) + b
k
with dimensionless boundary conditions
u
b(−1) = pb1 , u
b(1) = pb2 . (8.7)
where pb1 = u1 /pref and pb2 = u2 /pref . Now we introduce the differential operator A[·]
so that Eq.(8.6) can be rewritten as
d2 v v
A[b
u(b
z )] = 0, with A[v] := 2 − γ
b . (8.8)
db
z v+bk
According to the HPM, we introduce a simple and easy-to-handle linear auxiliary op-
erator L[·], related to the original operator A[·], defined as
d2 v v
L[v] := 2
−γ
b . (8.9)
db
z 1+bk
Observe that the choice for the expression of L[·] is arbitrary. Then, letting q ∈ [0, 1] be
an embedding parameter, we construct the function φ = φ(b z ; q) and the corresponding
family of homotopy equations
(1 − q) (L[φ(b
z ; q) − u0 (b
z )]) + q (A[φ(b
z ; q)]) = 0. (8.10)
136
8.1. Model of O2 delivery to tissue: analytical approach.
equations for u
b0 , u
b1 , u
b2 :
d2 u
b0 γ
bub0
(SP )0 q0 : 2
− = 0, (8.13)
dbz 1+b k
d2 u
b1 γ
bub1 d2 u
b0 d2 u
b0 γ
bub0
(SP )1 q1 : 2
− + u
b0 2
− (1 − k)
b
2
+
dbz 1+b k dbz db
z 1+b k
−γbu
b0 = 0, (8.14)
d2 u
b2 γ
bub2 d2 u
b1 d2 u
b1 d2 u
b0
(SP )2 q2 : 2
− − (1 − k)
b
2
+ u 0 2
+ u1
z2
b b
dbz 1+b k dbz db
z db
γ
bub1
+ −γ
bub1 = 0 , (8.15)
1+b k
with boundary conditions
u
b0 (−1) = pb1 , u
b0 (1) = pb2 , (8.16)
u
b1 (−1) = ub1 (1) = 0 , (8.17)
u
b2 (−1) = ub2 (1) = 0 . (8.18)
Thus, we have to solve the subproblems (SP )0 , (SP )1 , (SP )2 given by the Eqs(8.13)-
(8.15) equipped with boundary conditions (8.16)-(8.18), respectively.
Equation 2
du b0 γ
z) − for −1 ≤ zb ≤ 1
b
(b u
b0 (b
z) = 0
db
z 2
1 + k
b (8.19)
u
b0 (−1) = u b0 (1) = 1
where s
γ
b e2m − 1
m= and a= .
1+b
k e3m − e−m
Equation
2
du b1 γ
bub1 d2 u
b0 d2 u
b0
2
− = −bu 0 2
+ (1 − k)
b
db
z dbz dbz2
1+b k (8.21)
+(bγ − m2 )bu0 for −1 ≤ zb ≤ 1
u
b1 (−1) = ub1 (1) = 0
is a second-order linear non homogeneous differential equation. We look for a solution
u
b1 (b
z ) of the type:
u
b1 (b
z) = u
b1h (b
z) + u
b1p (b
z) ,
137
Chapter 8. Study of solute transport through the retinal tissue
where ub1h (b
z ) is a general solution of the homogeneous equation associated to the orig-
inal one and u b1p (b
z ) is a particular solution of the non homogeneous equation.
The homogeneous equation
d2 u
b1 γ
−
b
(b
z ) u
b1 (b
z) = 0 ,
dbz2 1+bk
has solution
u z ) = A1h embz + B1h e−mbz ,
b1h (b A1h , B1h ∈ R .
The values of the constants A1h , B1h will be determined by the enforcement of the
boundary conditions. We look now for a particular solution of
d2 u
b1 γ
bub1 d2 u
b0 d2 u
b0
2
− = −b
u0 2
+ (1 − k)
b
2
γ − m2 )b
+ (b u0 . (8.22)
dbz 1+k b dbz dbz
Observe that the right-hand side of Eq.(8.22) is a known function determined by the
solution of the problem (SP )0 . The solution of Eq.(8.22) is:
with
A2 B2 Am Bm
A1 = − , B1 = − , C1 = , D1 = − , E1 = 2AB ,
3 3 2 2
where
p2 e2m − pb1 )
A = a(b p1 e2m − pb2 ).
B = a(b (8.24)
Equation
2
du b2 γ
bub2 d2 u
b1 d2 u
b0
2
− = (1 − u0 − k) 2
− u 1
z2
b b
db
z dbz db
1+b k
+(b γ − m2 )bu1 for −1 ≤ zb ≤ 1
u
b1 (−1) = ub1 (1) = 0
is obtained by the same procedure used for subproblem (SP )1 . The solution u b2h (b
z)
of the homogeneous equation associated to the original one and the particular solution
u
b2p (b
z ) read
138
8.1. Model of O2 delivery to tissue: analytical approach.
1 2 2 2
By setting b = a cosh(2m) − am sinh(m) − 2a , we obtain
2 cosh(m) 3
u
b(b
z) = u
b0 (b
z) + u
b1 (b
z) + u
b2 (b
z) =
+ a embz (pB e2m − pA ) + e−mbz (pA e2m − pB )
139
Chapter 8. Study of solute transport through the retinal tissue
the photoreceptors mitochondria. The inner retina, instead, has an average oxygen con-
sumption rate that is approximately one-fifth of that of the outer region [23]. Based
on these facts, we can then identify different subdomains (or layers) along the retinal
thickness with a different metabolic activity. We consider two different model settings
which differ in the number of layers composing the outer retina, namely (see Fig.8.2
for the nomenclature):
• model (SM1 ): the outer retina is divided into two layers, one with constant con-
sumption rate and the other with Michaelis-Menten consumption rate;
• model (CM ): the outer retina is divided into three layers, one with Michaelis-
Menten consumption rate and the others with constant consumption rate.
In all the models, the inner retina is assumed to be, for simplicity, single layered.
Therefore, even if the oxygen sources in the inner retina are located only in two specific
layers (where the two capillary layers are), in this context, we consider a constant source
term uniformly distributed along the entire inner retina. The subdivision of the inner
retina into layers have been taken into account in Chap.7.
In each model, we set the following boundary conditions: fixed pressure at the interface
with the choroid (left boundary of the OR), which furnishes a constant and large oxygen
supply, and zero flux at the interface with the vitreous (right boundary of the IR).
(SM1 )
0 Ω1,2 L2 Ω3 L3 Ω4 Lt
(CM )
0 Ω1 L1 Ω2 L2 Ω3 L3 Ω4 Lt
Figure 8.2: Schematic representations of the domains of the simplified model (SM1 ) and of the complete
model (CM ). Red subdomains are characterized by a Michaelis-Menten consumption term, orange
subdomains by a constant consumption and cyan ones by a null consumption.
140
8.1. Model of O2 delivery to tissue: analytical approach.
ψ : Ω −→ Ω
b (8.28)
z − L1
z 7−→ zb = ψ(z) = 2 − 1.
L2 − L1
such that ψ(Ω2 ) = (−1, 1).
We also introduce the dimensionless variables:
p1 1 Qmax
2 (L2 − L1 )2 k0.5
pbi = , γ
b= , k0.5 =
b ,
pch pch D 4 pch
1 Q4 − S4 (Lt − L3 )2
R4 = .
pch D 4
We denote by pA , pB and pC the unknown values of oxygen pressure at z = L1 , L2 , L3 ,
respectively.
In order to compute the solution of equation system (8.25), we perform two (not
straightforward) steps. First, we compute the solution of differential equations of oxy-
gen profile independently in each different layer assuming parametric Dirichlet condi-
tions at the interface of each domain. Namely, we solve:
• Oxygen pressure profile in layer 1
∂ 2 pb1
= 0 for zb ∈ [ψ(0), −1] , (8.29)
∂b z2
with boundary conditions
pb1 (ψ(0)) = 1 , pb1 (−1) = pA . (8.30)
141
Chapter 8. Study of solute transport through the retinal tissue
Ω1 Ω2 Ω3 Ω4
z (cm)
0 L1 L2 L3 Lt
ψ ψ
zb
ψ(0) −1 1 ψ (L3 ) ψ (Lt )
This last step is carried with the symbolic engine of Matlab® . The analytical solution
of the oxygen profile in Ω is obtained by coming back to the original dimensional
variables.
In Fig.8.4 we show the computed oxygen pressure profiles obtained with the (CM )
model considering different value of the oxygen source in the inner retina.
The curves are computed with a total retina thickness of 250 µm, equally divided in
outer retina and inner retina. The abscissa of the boundary of the layer 2 are chosen so
that it occupies the 20% of the outer retina, with the extremes located at the 75% and
85% of the retinal thickness. The other values of parameters are reported in Tab.8.2.
From Fig.8.4, we observe that the oxygen partial pressure is maximal at the choroid
and then decreases sharply due to the high oxygen consumption located in Ω2 . If the
blood flow supplies an adequate amount of nutrients the oxygen profile, after attaining
its minimum in Ω2 , increases along Ω3 and Ω4 . Otherwise, the oxygen profile decreases
along the layers Ω3 and Ω4 till reaching a very low value (near to 0), since the choroid
is not able to feed all the layers of the retina.
For completeness, in Fig.8.5 we compare the oxygen profiles obtained with the (CM )
and (SM )1 models (parameters values are reported in Table 8.3). In the (SM )1 we
have joined Ω1 and Ω2 into the domain Ω1,2 (see Fig.8.2). The maximal consumption
terms in domain Ω1,2 , where a Michaelis-Menten kinetics is assumed, is determined as
follow:
Z L2
max 1
Q1,2 = Qmax dz . (8.37)
L2 L1 2
Slight variations of the source term in the inner retina are considered in order to obtain
as much as possible comparable profiles. Observe that (SM )1 model gives a trend
pretty similar to that of the complete model.
143
Chapter 8. Study of solute transport through the retinal tissue
80
Ω1 Ω2 Ω3 Ω4
70
oxygen partial pressure p (mmHg)
60
50
40
30
20
10
0
0 0.005 0.01 0.015 0.02 0.025
distance from the choroid z (cm)
Figure 8.4: Oxygen partial pressure profiles along the whole retina depth for source term S4 (from
highest to lowest profile): 27, 26.5, 26, 25.5 and 25.25 mmHg/s.
Figure 8.5: Comparison of the profiles obtained with original model CM (blue profile)and simplified
model SM1 (aquamarine profile). Parameters values are specified in Table 8.3.
144
8.1. Model of O2 delivery to tissue: analytical approach.
CM SM1
Qmax
2 Q4 S4 Qmax
1,2 Q4 S4
90 26 25.45 18 26 25.95
Table 8.3: Values of maximal consumption rates and source terms (expressed in mmHg/s) employed to
produce the profiles of Figure 8.5.
diseases.
In the following, we employ the analytical solution to the retina oxygenation problem
in order to explore the sensitivity of the oxygen pressure profile with respect to changes
due to variation of consumption rates and intensity of the oxygen blood source.
We perform the sensitivity analysis using the analytical solution obtained in models
(SM )1 , since it has a lower complexity in the analytical expression and in its deriva-
tives.
We define the set of the model parameters P = {P1 , P2 , P3 }, where
P1 = Qmax
1,2 , P2 = Q4 , P3 = S4 . (8.38)
We indicate with P ∗ = {P1∗ , P2∗ , P3∗ } the set of the corresponding baseline values
(listed in Tab.8.3).
We define the sensitivity index Spk (z) with respect to the model parameter Pk as
k ∂p(z; P)
Sp (z) := k = 1, 2, 3. (8.39)
∂Pk P=P ∗
In Fig.8.6, we report the indexes curves of the oxygen pressure profiles with respect
to parameter P obtained from model SM1 . The sensitivity index curve with respect
to source term (data not reported) is symmetric to that of the consumption term Q4
with respect to the domain axis, since both Q4 and S4 are constant values with oppo-
site signs. All the indexes show the largest variation in layer Ω4 (inner retina region).
Moreover Qmax 1,2 influence in a uniform way the inner retina.
Eventually, we observe that the oxygen profile is most sensitive to the model parameters
of the inner retina. As a matter of fact, we notice a difference of one order of magni-
tude between the variation with respect to changes in the maximal consumption rate
Qmax
1,2 and the variation with respect to changes in the consumption and source terms
of the inner retina. This is in accordance with the clinical evidence that retinal tissue
pathologies are importantly related to impaired retinal circulation
145
Chapter 8. Study of solute transport through the retinal tissue
0
Ω 1,2 Ω3 Ω4
−0.1
−0.2
se ns it iv ity inde x S 1p ( z )
−0.3
−0.4
−0.5
−0.6
−0.7
−0.8
−0.9
0 0.005 0.01 0.015 0.02 0.025
dist anc e f r om t he chor oid z ( c m)
0
−2 Ω 1,2 Ω3 Ω4
−4
se ns it iv ity inde x S 2p ( z )
−6
−8
−10
−12
−14
−16
−18
0 0.005 0.01 0.015 0.02 0.025
dist anc e f r om t he chor oid z ( c m)
146
CHAPTER 9
Mathematical assessment of drug build–up in
the posterior eye following transscleral delivery 1
The understanding of drug delivery mechanisms in the posterior segment of the eye
(PSE) - including sclera, choroid and retina - is one of the most challenging tasks in
the pharmaceutical industry [100]. The efficiency of drug delivery to the PSE is hin-
dered by several barriers. Static barriers consist of physical obstacles to drug diffu-
sion such as the sclera itself, the retinal pigment epithelium (RPE, the so–called outer
blood retinal, oBRB) and the retinal vessels (the so–called inner blood retinal barrier,
iBRB). Dynamic barriers include drug clearance mechanisms through blood and lym-
phatic vessels and degradation processes. Drug solubility, lipophilicity, charge, degree
of ionization, molecular size and shape affect the penetration rate of the drug across the
various barriers [54]. Convection by interstitial fluid filtration can play a certain role
especially when considering low–diffusible molecules [113]. There is also consider-
able evidence suggesting that active transport (“pumping”) across the RPE can induce
significant effects, sucking out fluid, and thus dissolved drugs, from the retina towards
the choroid [57, 204].
Literature on mathematical models of the pharmacokinetics of ocular and periocu-
lar delivery systems is relatively sparse (see Ch.4 of [200] for a recent review on this
subject). Tab. 9.1 presents –at the best of our knowledge – a schematic survey of the
existing most recent mathematical models of ocular drug delivery.
There is a general agreement in distinguishing between small (molecular weight
< 1 kDa, as fluoroscein) and large (molecular weight > 10 kDa, as albumin, anti-
bodies) molecules, since they display different diffusivities. Interstitial fluid filtration
1 This chapter is part of the article "Mathematical assessment of drug build-up in the posterior eye following trans-scleral
delivery" by Causin, P., Malgaroli, appeared on Journal of Mathematics in Industry 6.1 (2016): 9.
147
Chapter 9. Mathematical assessment of drug build–up in the posterior eye following
transscleral delivery
Table 9.1: Schematic prospect of existing mathematical models in recent literature on ocular drug deliv-
ery. Symbols: nD, n = 0, 1, 3: spatial dimensions of the model; V=vitreous, R=retina, C=choroid,
S=sclera. a : treated with an addictional fictitious advective term; b : treated with different inward and
outward membrane permeabilities. Merged cells in a row indicate that the corresponding domains
are treated as a sole entity. Interface boundary conditions are checked when a Robin–type bound-
ary conditions with a specific membrane permeability is used for drug at interface; c : a partition
coefficient is considered at the interface; d : Navier–Stokes equations are considered.
148
9.1. Mathematical model of the posterior segment of the eye
is also kept into account in several papers by Darcy equations, under the hypothesis of
steady flow. Interface boundary conditions for the drug concentration between different
subdomains (vitreous, retina, choroid, sclera or combinations of them) are assigned via
simple continuity of normal fluxes or via more complex Robin–type conditions mod-
elling the presence of a permeable membrane. More controversial, less investigated,
points are represented by the inclusion of active transport mechanisms (see [14, 194])
and the interaction of the drug with blood flow in circulatory vessels (see [7,14]). These
latter aspects turn out to be very relevant when computing drug levels in the retina after
periocular administration. The retina is an important therapeutic target but only a few
studies address it specifically (refer to Tab. 9.1).
In this chapter, we present a 1D model including vitreous, retina, choroid and sclera
subdomains with a blood phase in the retina. We compute the filtration velocity of
the interstitial fluid and we solve convective–diffusive–reactive equations for drug con-
centration. Permeability parameters and partition coefficients simulate the presence of
internal membranes and barriers. We carry out numerical simulations based on transs-
cleral drug delivery, which is an attractive alternative to the intravitreous mode of ad-
ministration. We investigate the problem sensitivity with respect to the above men-
tioned parameters, which allow to evaluate the importance of tuning these properties
when devising a new drug and their relative importance.
The chapter is organized as follows. In Sect. 9.1, we present the geometrical assump-
tions and the mathematical model of the PSE. In Sect.9.2, we show the results obtained
from numerical solution of the mathematical model and sensitivity analysis. Based on
these results, we discuss the significance of the model and the main results relevant
for devising new drug formulations and delivery techniques. A theoretical analysis is
carried out in Sect.9.3 to establish lower and upper bounds for the drug concentration
in the retina, an important target for drug delivery.
149
Chapter 9. Mathematical assessment of drug build–up in the posterior eye following
transscleral delivery
• choroid (C), a dense network of large and small blood vessels with a relatively
sparse population of cells;
• retina (R), composition of several layers of densely packed neuronal and glial
cells, vascularized by arterioles and venules which run superficially along the reti-
nal inner surface and supply/drain the embedded capillary plexi (see Fig. 9.2, top
panel);
• vitreous(V), a clear, avascular, gelatinous body which accounts for about 80% of
the volume of the eye.
Figure 9.1: Globe of the eye and schematics of the PSE regions considered in the mathematical model.
RPE: retinal pigmented epithelium, oBRB: outer blood retinal barrier, iBRB: internal blood retinal
barrier, ILM: internal limiting membrane.
150
9.1. Mathematical model of the posterior segment of the eye
• velocity continuity
vi · ni = vj · ni
Drug mass balance in the PSE. Drug mass balance is enforced in each layer according
to the characteristic features of the layer itself.
∂Cj ∂ 2 Cj ∂Cj
= Dj 2
− vj − kj Cj , j = S, V. (9.2)
∂t ∂x ∂x
where vj is computed from Eq.(9.1) in layer j.
Drugs in the choroid bloodstream rapidly equilibrate with the extravascular space, due
to the fenestrated structure of choriocapillaris [41, 71]. For this reason, a unique con-
centration value is considered both for tissue and blood domains, yielding
∂CC ∂ 2 CC ∂CC
= SC (t) + DC − v C − kC CC , (9.3)
∂t ∂x2 ∂x
where SC = SC (t) is a systemic drug source rate, vC is computed from Eq.(9.1) solved
in the choroid and kC the clearance rate due to choriocapillaries.
Drug concentration in the retina is modeled assuming the domain to contain a mixture
of tissue and blood vessels. We denote by φ = φ(x) ∈ (0, 1) the retinal vascular poros-
ity, which represents the volumetric fraction occupied by the vascular space, the re-
maining volumetric fraction being extravascular space. We have that φ → 1 in the inner
regions where the vascular beds are located and φ → 0 in the outer regions. We consider
a mathematical representation of the function φ as in Fig. 9.2 (bottom panel) which re-
sults from the superposition of Gaussian functions centered in the anatomical location
of each vessel plexum. The standard deviation of the Gaussian is chosen in such a way
that the function is significantly greater than zero in a region roughly corresponding to
the thickness of the vessel plexum itself (superficial layer and two capillary beds). We
let β = β(φ) [1/s] be the rate of drug transport across the blood vessels representing
the effect of the iBRB (see also [14, 106]). We set β = Lvw Avessel ρ, where Lvw ' 10−6
[cm/s] is the permeability of the blood vessel wall [124], Avessel ' 9 · 10−6 [cm2 ] the
average lateral surface of a vessel (arteriole, capillary, venule [66]) and ρ = ρmax φ the
number of vessels per cm3 of tissue, with ρmax ' 4.55 · 106 [1/cm3 ] [41]. Denoting
by CRt = CRt (t, x) and CRb = CRb (t, x) the drug concentration in tissue and vascular
151
Chapter 9. Mathematical assessment of drug build–up in the posterior eye following
transscleral delivery
Figure 9.2: Top: retinal tissue with blood vessel plexi. Vessels nourish the inner retinal region, while
the outer retina is avascular. The rightmost plexum is the arteriolar/venular network, middle and
leftmost plexi are constituted by capillary networks. In the retinal tissue, different types of neural
cells are represented. Bottom: mathematical representation of the function φ, volumetric fraction
of blood in the multiphase model. The maximal values are attained in correspondence of the vessel
plexi.
152
9.2. Numerical simulations
The complete model for amounts to solve Eq. (9.1) along with Eqs. (9.2)-(9.4) with
the respective interface conditions. We enforce at the external boundary of the vitre-
ous a hydrostatic pressure equal to 15 mmHg, corresponding to a normal intraocular
pressure (IOP), and at the external boundary of the sclera a hydrostatic pressure equal
to 10 mmHg, corresponding to the episcleral venous pressure [14]. For the drug, if
not differently specified, we prescribe at the sclera external boundary a concentration
exponentially decreasing in time from the initial value 1 mg/cm3 fitting the trend ob-
tained from a model of drug release in posterior eye gel implants [106] (blue curve in
Fig. 9.3 of the present paper). The vitreous is assumed to be thick enough so that there
is no normal diffusive and convective flux of drug across its external boundary when
considering episcleral drug delivery. The drug concentration problem is supplied by
the initial conditions Cj (t = 0, x) = Cj,0 (x) = 0, j = S, C, R, V (exogenous drug).
The fluid filtration equations are solved numerically by means of the function bvp4c
of MATLAB® . The drug concentration equations are solved using for time integration
an explicit Runge–Kutta scheme with adaptive time step implemented by the function
ode23t of MATLAB® . Spatial discretization is performed by an in–house developed
code implementing 1D linear finite elements with stabilization techniques for possibly
advection–dominated problems. The parameters used in the simulations are reported
in Tab. 9.2. Notice that the external source terms SC and SRb may be non-zero due to
(a minimal) systemic drug absorption into circulation and successive release into the
tissues. Being unable to quantify such sources, it is reasonable in first approximation
to set SC = SR = 0.
153
Chapter 9. Mathematical assessment of drug build–up in the posterior eye following
transscleral delivery
Parameter Value Description Ref.
tS 600 µm S thickness [133]
tC 300 µm C thickness [133]
tR 246 µm R thickness [133]
tV 15000 µm V thickness [139]
(K/µ)S 8.4·10−7 cm2 /s Hydraulic conductivity in S [14]
(K/µ)C 2.35·10−11 cm2 /s Hydraulic conductivity in C [14]
(K/µ)R 1.5 ·10−11 cm2 /s Hydraulic conductivity in R [14]
(K/µ)V 1.5−11 cm2 /s Hydraulic conductivity in V [14]
RSC 10−7 cm/s/mmHg Hydraulic permeability at ΓSC [82, 194]
RCR 10−7 cm/s/mmHg Hydraulic permeability at ΓCR [82, 194]
RRV 10−7 cm/s/mmHg Hydraulic permeability at ΓRV [82]
DS 4·10−7 cm2 /s Drug Diffusivity coefficient in S [133]
DC 1.6·10−7 cm2 /s Drug Diffusivity coefficient in C [133]
DR 1.17 ·10−7 cm2 /s Drug Diffusivity coefficient in R [133]
DV 6 10−6 cm2 /s Drug Diffusivity coefficient in V [200]
kS 3 ·10−4 1/s Drug Clearance coefficient in S [133]
kC 3 ·10−4 1/s Drug Clearance coefficient in C [133]
kRt 3 ·10−4 1/s Drug Clearance coefficient in Rt [133]
kRb 3 ·10−4 1/s Drug Clearance coefficient in Rb [133]
kV 8 ·10−5 1/s Drug Clearance coefficient in V [174]
LSC 10−4 cm/s Permeability coefficient at ΓSC [115]
LCR 10−5 cm/s Permeability coefficient at ΓCR [71, 174]
LRV 10−5 cm/s Permeability coefficient at ΓRV [71, 115, 174]
PCS 1 Partition coefficient at ΓSC [115]
PCR 1/1.33 Partition coefficient at ΓCR [133]
PRV 1/10 Partition coefficient at ΓRV [115]
Table 9.2: Value of the model parameters used in the numerical simulations (if not specified otherwise)
embedding blood vessels (function φ > 1) and the second one the avascular portion.
We have also proved that the numerical solution, in the case of no convective field,
is bounded from below and above. The following theorem holds:
C̄V λ(t)
Theorem 1. Let CN := max max {PCR C̄C , , }, max {CRb,0 ,
t∈(0,T ) PRV kRb x∈ΩR
CRt,0 } . Then, the solution CR = CR (t, x) = [CRt (t, x), CRb (t, x)] of problem (9.7)-
(9.8) satisfies 0 ≤ CR (t, x) ≤ CN a.e. in ΩR × (0, T ).
The proof of Theorem1 is reported in Sect.9.3.
In Figs. 9.4 and 9.5, we report results from simulations in different literature models
for molecules comparable to fluoroscein and for similar drug inlet boundary conditions
with episcleral plug or injection. The peak concentration value found in our simulations
is comparable to the one of the models of [106] and [133] (scaled inlet concentration)
in the retina and to the ones of the models of [14] and [133] (scaled inlet concentration)
in the choroid. Time-to peak values of our model, of the order of 1-2 [h], compa-
rable to the results of [115]. Such a scattering of the results may be ascribed to i)
154
9.2. Numerical simulations
Figure 9.3: Drug peak concentration as a function of time in the different layers considered in the model.
The curves relative to the retina are specified into tissue and blood phases and further divided into
inner and outer region, the first indicating the portion of the retina embedding blood vessels (function
φ > 1) and the second one the avascular portion. The inset shows a zoom for short times.
different dimensionality of the models; ii) different values of the parameters (see the
below sensitivity study); iii) different biophysical mechanisms included (active pump-
ing, filtration velocity, membrane permeabilities, membrane partition coefficient). The
following sensitivity analysis is aimed at clarifying the role of the different parameters
and mechanisms.
155
Chapter 9. Mathematical assessment of drug build–up in the posterior eye following
transscleral delivery
Figure 9.4: Drug peak concentration values for the present model and literature models (data repre-
sented are where available and upon conversion of units of measure). Comparable values of pa-
rameters and inlet conditions are used. Values denoted by the symbol ‘*’ have been obtained by
multiplying the values by 100 to scale drug concentration at the boundary (assuming a linear be-
haviour as suggested by the results of [115]. Values from [14] have been interpreted as percentage
of the enforced boundary concentration, in lack of indications. Notice that log scale is used for the y
axis.
permeability. Both for peak concentration and time-to-peak, the curves results rela-
tively flat even varying the parameter of 1 or 2 decades in log scale. This is probably
due to permeability not being a limiting phenomenon in these regions. The iBRB drug
permeability does not seem to have a major role except for a region of diameter span-
ning about 1 decade at the left and at the right of the baseline value, where the slope
of the curve is comparable to the other ones. The mild influence of iBRB might due
to having neglected the complexity of active/carried–mediated transport mechanisms
across these interfaces and having considered a single partition coefficient across the
vessel walls in both (inward and outward) directions. Representing active pumping as
a fictitious velocity and not as a membrane effect, hinders the role of RPE (see [115]
for this choice).
In Fig.9.9, we show the drug peak concentration as a function of time with active
(‘+’) or inactive (‘-’) selected components of the convective field in baseline conditions.
In Fig.9.8 we show the sensitivity of the drug peak concentration and time-to-peak
values in tissue and blood phases of the retina with respect to the different components
of the convective field (filtration velocity and active pumping). The choroid (data not
reported) results to be very weakly sensitive to the variations in the convective field
values. From the results, it is apparent that the filtration velocity is too low to induce
156
9.2. Numerical simulations
Figure 9.5: Drug time-to-peak values for the present model and literature models (data are represented
where available and upon conversion of units of measure). Comparable values of parameters and
inlet conditions are used. Notice that log scale is used for the y axis.
An important result of the present analysis is the way in which a 30% increase of
the drug peak concentration in choroid and retina can be obtained from the baseline
values, for example to meet a therapeutic threshold. Assuming continuous dependence
of the solution on data - a property which can be inferred via a mathematical analysis
not much different to the one performed in Sect.9.3 - it is found that such an increase
can be obtained acting on i) scleral biophysical parameters and then on choroidal and
retinal ones; ii) active pumping. As for i), in the neighborhood of baseline conditions,
diffusion and clearance appear to play a comparable role, because comparable varia-
tions in the scleral diffusion (increase) or scleral clearance (decrease) are required to
increase drug peak concentration (notice the slope of the curves in Fig. 9.6). The sclera
is permeable to hydrophilic compounds, even macromolecules, but the permeability
in the RPE/choroid/Bruch’s membrane is 1-2 orders of magnitude lower than in the
sclera [71]. Moreover, the trend line of decreasing choroid-Bruch’s membrane perme-
ability with increasing solute lipophilicity and/or molecular radius appears to be steeper
than the sclera. However, the simulations suggest that the slope of the sensitivity curves
is always higher for the scleral parameters that for the choroidal ones. Moreover, sensi-
tivity in a certain layer is higher to properties of layers located at its left side than at its
157
Chapter 9. Mathematical assessment of drug build–up in the posterior eye following
transscleral delivery
Figure 9.6: Drug peak concentration in the choroid (left column) and retina (right column) as a function
of diffusion coefficient (top row), clearance rate (middle row) and membrane permeability (bottom
row) in the various layers (specified in the legends). The dotted vertical lines indicate baseline values
of the parameters. Arrows indicate the value of concentration in baseline conditions.
158
9.2. Numerical simulations
Figure 9.7: Drug time-to-peak in the choroid (left column) and retina (right column) as a function of
diffusion coefficient (top row), clearance rate (middle row) and membrane permeability (bottom row)
in the various layers (specified in the legends). The dotted vertical lines indicate baseline values of
the parameters. Arrows indicate the value of time-to-peak in baseline conditions.
159
Chapter 9. Mathematical assessment of drug build–up in the posterior eye following
transscleral delivery
Figure 9.8: Drug peak concentration and time-to-peak as a function of time with active (‘+’) or inactive
(‘-’) selected components of the convective field in baseline conditions.
right side. This is a natural consequence of the positioning of the source. In particular,
referring to the theoretical analysis carried out for the retinal domain, this implies that
the drug concentration in the choroid is always responsible for the upper bound. As for
ii), pro–drugs have been envisaged as carriers able to favorably enhance drug delivery.
This a very advanced issue in pharmacokinetics, we refer to [132] for a general review.
The present model does not allow to provide data regarding the 3D spatial distribu-
tion of the drug on the eye globe. While this aspect is very important and has been re-
cently considered in a few mathematical models [14, 106, 115], it remains very difficult
to compare results from such models with experiments. Unknown boundary conditions
in the 3D models, on the one hand, and tissue homogenization after explant with loss
of spatial dependence, on the other, are just examples of such problems.
160
9.3. Bounds for drug concentration in the retina
Figure 9.9: Sensitivity study of the drug peak concentration and time-to-peak values in tissue and blood
phases of the retina with respect to the different components of the convective field (filtration velocity,
top row and active pumping, bottom row). The choroid (data not reported) results to be very weakly
sensitive to variations of the convective field values.
main ΩR become:
∂CRt
−DRt · nRt = LCR (PCR CRt − C̄C ) on ΓCR , (9.5)
∂x
∂CRt
−DRt · nRt = LRV (PRV CRt − C̄V ) on ΓRV . (9.6)
∂x
In the sequel, L2 (ΩR ) will denote the space of measurable functions whose square is
Lebesgue integrable in ΩR . The space L2 (ΩR ) is equipped with the norm || · ||L2 (ΩR ) .
We denote with (·, ·) its scalar product both for scalar and vector functions. We intro-
duce the Sobolev space H 1 (ΩR ), which denotes functions that belong to L2 (ΩR ) along
with their first order distributional derivative in space. For v ∈ H 1 (ΩR ), we denote by
∂x v such a derivative. The space H 1 (ΩR ) is equipped with the norm || · ||H 1 (ΩR ) .
We also introduce the space L2 (0, T ; V ) = {v : (0, T ) → V | v(t) is measurable,
RT
||v||2L2 (0,T ;V ) := 0 ||v(t)||2V dt < +∞}, with V = H 1 (ΩR ) or L2 (ΩR ), and the
space L∞ (0, T ; L2 (ΩR )) = {v : (0, T ) → L2 (ΩR | v(t) is measurable, ||v(t)||2L2 (ΩR ) is
essentially bounded in (0, T )}, endowed with the norm ||v||L∞ (0,T ;L2 (ΩR )) := inf{M >
0 : kv(t)k2L2 (ΩR ) ≤ M a.e. in (0, T )}. For v ∈ L2 (0, T ; V ), we denote by ∂t v its
derivative with respect to time, such that ∂t v ∈ L2 (0, T ; V 0 (ΩR )), V 0 being the dual
161
Chapter 9. Mathematical assessment of drug build–up in the posterior eye following
transscleral delivery
R
space of V , and ∀ u ∈ L2 (0, T ; V ), h∂t v(t), u(t)i = ΩR ∂t v u dx represents a duality
pairing. We here assume for simplicity that DR , kRt , kRb , LCR , LRV are positive con-
stants and that β ∈ L∞ (ΩR ) and CRb,0 , CRt,0 ∈ L2 (ΩR ). We let φ ∈ C 0 (Ω̄R ) with
φm = minx∈ΩR φ(x) and φM = maxx∈ΩR φ(x).
The weak counterpart of the multiphase system (9.4) reads:
- in the tissue phase, find CRt = CRt (t, x) ∈ L2 (0, T ; H 1 (ΩR )) with ∂t CRt ∈
L2 (0, T ; H −1 (ΩR )), such that:
h(1 − φ)∂t CRt , vi + DR ((1 − φ)∂x CRt , ∂x v) + (kR (1 − φ)CRt , v)
+(β(CRt − CRb ), v) + LCR ((1 − φ)(CRt − PCR C̄C ), v)Γ (9.7)
CR
+LRV ((1 − φ)(PRV CRt − C̄V )v)Γ = 0 ∀v ∈ H 1 (ΩR );
RV
- in the blood phase, find CRb = CRb (t, x) ∈ L2 (0, T ; L2 (ΩR )) with ∂t CRb ∈
L2 (0, T ; L2 (ΩR )), such that, ∀w ∈ L2 (ΩR ),
hφ∂t CRb , wi + (kRb φCRb , w) + (β(CRb − CRt ), w) = (SRb , w) (9.8)
We prove that the drug concentration in the multiphase model is non–negative and
bounded above by a maximal value CN depending on the initial conditions, boundary
data and external sources.
C̄V λ(t)
Theorem 2. Let CN := max max {PCR C̄C , , }, max {CRb,0 ,
t∈(0,T ) PRV kRb x∈ΩR
CRt,0 } . Then, the solution CR = CR (t, x) = [CRt (t, x), CRb (t, x)] of problem (9.7)-
(9.8) satisfies 0 ≤ CR (t, x) ≤ CN a.e. in ΩR × (0, T ).
Proof of Theorem 1. We face the multiphase problem in the retina in a decoupled
manner by introducing the following Gauss-Seidel abstract iterative procedure:
0
given 0 ≤ CRb k
≤ CN a.e. in ΩR × (0, T ), ∀k = 1, 2, . . . , find CRk = [CRt k
, CRb ]∈
2 1 2 2
L (0, T ; H (ΩR )) × L (0, T ; L (ΩR )) such that it yields:
- in the tissue phase
k k k
h(1 − φ)∂t CRt , vi + DR ((1 − φ)∂x CRt , ∂x v) + (kR (1 − φ)CRt , v)
k k−1 k
+(β(CRt − CRb ), v) + LCR ((1 − φ)(CRt − PCR C̄C ), v)Γ (9.9)
CR
k
1
+LRV ((1 − φ)(PRV CRt − C̄V )v)Γ = 0 ∀v ∈ H (ΩR );
RV
We verify by a recursive argument that the sequence CRk satisfies 0 ≤ CRk ≤ CN . The
key point is given by the following lemma, for k = 1, 2, . . . .
162
9.3. Bounds for drug concentration in the retina
k−1 k
Lemma 9.3.1. Given 0 ≤ CRb ≤ CN in ΩR × (0, T ), then 0 ≤ CRt ≤ CN in
k k
ΩR × (0, T ). Conversely, given 0 ≤ CRt ≤ CN in ΩR × (0, T ), then 0 ≤ CRb ≤ CN in
ΩR × (0, T ),
Proof of Lemma 9.3.1. Let ` be a constant. For any u ∈ V , we introduce the non-
negative functions [u − `]+ = max{u − `, 0}, [u − `]− = max{` − u, 0}, such that
[u − `]+ , [u − `]− ∈ V . We refer to [74] for the properties of such functions.
k−1
We start proving that CRk ≥ 0 under the hypothesis that CRb ≥ 0. We set ` = 0 and
−
choose v = −[CRt ] as a test function in (9.9), yielding
1 d k − 2 k − 2
(1 − φM ) k[CRt ] kL2 (ΩR ) + DR (1 − φM )k∂x [CRt ] kL2 (ΩR )
2 dt
k − 2
+(kR (1 − φM ) + β(φm ))k[CRt ] kL2 (ΩR )
k − 2 k − 2
+LRV PRV ((1 − φ)([CRt ] ) ) + LCR ((1 − φ)([CRt ] ) ) ≤
ΓRV ΓCR (9.11)
Z
k−1 k − k −
−β(φm ) CRb [CRt ] dx − LCR PCR C̄C ((1 − φ)[CRt ] )
ΩR ΓCR
k −
−LRV C̄V ((1 − φ)[CRt ] ) .
ΓRV
The right-hand of side of (9.11) is non-positive for a.e. t ∈ (0, T ). Integrating in time
k
over (0, t) and using the fact that [CRt,0 ]− = 0, we obtain
Z t
k − 2 k − 2
k[CRt (t)] kL2 (ΩR ) + 2DR k∂x [CRt ] kL2 (ΩR )
0
Z t
β(φm ) k − 2
+2 kR + k[CRt ] kL2 (ΩR ) (9.12)
1 − φM 0
Z t Z t
k − 2 k − 2
+2LRV PRV ([CRt ] )
+ 2LCR ([CRt ] ) ≤ 0.
0 ΓRV 0 ΓCR
We evaluate the integrals at the left-hand side of (9.12) for t = T and integrate once
again, this time on [0, T ], obtaining:
β(φm ) k − 2
min{1 + 2T kR + , 2DR T }k[CRt ] kL2 (0,T ;H 1 (ΩR ))
1 − φM
Z T Z T (9.13)
k − 2 k −
+2T LRV PRV ([CRt ] ) + 2T LCR ([CRt ] )2 ≤ 0.
0 ΓRV 0 ΓCR
k
From (9.13) we deduce that [CRt (t)]− = 0 and thus CRt
k
≥ 0 a.e. in ΩR × (0, T ). Now,
−
we choose v = −[CRb ] as a test function in (9.10), yielding
1 d k − 2 k − 2
φm k[CRb ] kL2 (ΩR ) + (β(φm ) + φm kRb )k[CRb ] kL2 (ΩR ) ≤
2 dt Z Z (9.14)
k k − k −
−β(φm ) (CRt )[CRb ] dx − SRb (t, φm )[CRb ] dx.
ΩR ΩR
Thanks to the previous result and to the hypotheses on the external source, the right-
hand of side of (9.14) is non-positive for a.e. t ∈ (0, T ). Integrating in time over (0, t)
163
Chapter 9. Mathematical assessment of drug build–up in the posterior eye following
transscleral delivery
k
and using the fact that [CRt,0 ]− = 0, we obtain:
Z t
k − 2 β(φm ) k − 2
k[CRb (t)] kL2 (ΩR ) + 2 + kRb k[CRb ] kL2 (ΩR ) ≤
φm 0
(9.15)
β(φm ) t
Z Z Z tZ
k k − SRb (t, φm ) k −
−2 (CRt )[CRb ] dx − 2 [CRb ] dx.
φm 0 ΩR 0 ΩR φm
We evaluate the integrals at the left-hand side for t = T and integrate once again, this
time on [0, T ], obtaining:
β(φm ) k − 2
1 + 2T ( + kRb ) k[CRb ] kL2 (0,T ;L2 (ΩR )) ≤ 0. (9.16)
φm
k
From (9.16) we deduce that [CRb (t)]− = 0 and thus CRb
k
≥ 0 a.e. in ΩR × (0, T ).
k−1
Now we prove that CRk ≤ CN under the hypothesis CRb ≤ CN . We set ` = CN and
k +
we choose v = [CRt − CN ] as a test function in (9.9), yielding
β(φm ) k
min 1 + 2T kR + , 2DR T k[CRt − CN ]+ k2L2 (0,T ;H 1 (ΩR ))
1 − φM
Z TZ
k
+CN kR T [CRt − CN ]+ dx dt
0 ΩR
Z T Z
β(φm ) k k−1
+ T [CRt − CN ]+ (CN − CRb ) dx dt
1 − φM 0 ΩR
Z T Z T (9.17)
k + 2 k
+LRV PRV T ([CRt − CN ] )
dt + LCR T ([CRt − CN ]+ )2 dt
0 ΓRV 0 ΓCR
Z T
k
− CN ]+
+LRV PRV CN − C̄V T [CRt dt
0 ΓRV
Z T
+LCR T k
(CN − PCR C̄C )[CRt − C N ]+
dt ≤ 0.
0 ΓCR
k
Proceeding in a similar way as described above, we deduce that k[CRt −CN ]+ k2L2 (0,T ;H 1 (ΩR )) =
k
0, which directly implies that CRt ≤ CN a.e. in ΩR × (0, T ). Now, we choose
k +
v = [CRb − CN ] as a test function in (9.10), yielding
1d k
φm k[CRb k
− CN ]+ k2L2 (ΩR ) + (φm kRb + β(φm ))k[CRb − CN ]+ k2L2 (ΩR ) ≤
2 dt Z
k k
−β(φm ) (CN − CRt )[CRb − CN ]+ dx (9.18)
ΩR
Z
k
− φ(kRb CN − λ(t)) [CRb − CN ]+ dx.
ΩR
Upon observing that all the terms at the right hand side are non-positive, we integrate
k
in time over (0, T ) using the fact that [CRb,0 − CN ]+ = 0 and we integrate once again
on [0, T ], obtaining
k
(φm + 2T (β(φm ) + φm kRb ))k[CRb − CN ]+ k2L2 (0,T ;L2 (ΩR )) ≤ 0.
164
9.3. Bounds for drug concentration in the retina
k
We deduce that k[CRb − CN ]+ k2L2 (0,T ;L2 (ΩR )) = 0, that directly implies that CRb
k
≤ CN
a.e. in ΩR × (0, T ).
To conclude the theorem, we prove the following lemma.
Lemma 9.3.2. The sequence CRk converges to the solution of system (9.7)-(9.8), pre-
cisely
k k
lim kCRt − CRt kL2 (0,T ;H 1 (ΩR )) = 0, lim kCRb − CRb kL2 (0,T ;L2 (ΩR )) = 0,
k→∞ k→∞
Taking the limit for M → +∞ and noting that the sum is bounded above, we have that
lim kekj kL2 (ΩR ) = 0 lim kekj kL2 ((0,T );V ) = 0, j = Rb, Rt.
k→∞ k→∞
165
Conclusions
In the first part of the thesis, we have presented mathematical and numerical models
for blood flow and oxygen transport in large rigid/distensible networks. These latter
are represented by a graph of 1D tubes generated, on a computer, by fractal algorithms
which reproduce anatomical features of a real network.
A detailed representation of the blood rheology in the vessels is taken into account
describing blood as a mixture of two components, red blood cells and plasma, respec-
tively. Upon averaging the Navier-Stokes equations on the vessel cross section, blood
flow and pressure drop in each vessel element result to be connected by a generalized
Ohm’s law including a conductivity parameter. This latter is function of the area and
shape of the tube cross section and biophysical properties of blood and, in the case of
distensible vessels, is a dynamical parameter. The cross section shape is described by a
simplified phenomenological structure equations or by more sophisticated relations de-
termined in a consistent way by thin/thick-wall structural models, loaded by pressure.
In the case of a thin-wall model, we also have include the effect of chemical stimuli
in the vessel wall due to presence of vadilator/vasoconstriction substances. A buckling
model is considered in the case of venules, which can experience low/possibly negative
values of transmural pressure.
We also proposed a novel 1D averaged version of the oxygen transport equation, in-
cluding RBCs-bounded and free parts. The resulting equation is modulated by moment
parameters which describe the effect of the oxygen profile in the vessel.
In Part II, we have employed the previous described mathematical models to pro-
167
Conclusions
pose a novel study of the coupling between blood flow mechanics and oxygen transport
in the eye retina. We also included in the model the mathematical description of oxygen
delivery in the retinal tissue, a strongly metabolically compartmentalized region. The
complete model results into a multiscale model which is solved via numerical tech-
niques. The model is validated against experimental data on humans and is used to
carry out predictive results, among which a sensitivity analysis. Eventually, we have
focused our attention on drug delivery in the posterior segments of the eye, a delicate
and difficult target even for modern pharmacology.
Future research work should deal with the inclusion in the proposed model of:
• a model of capillary beds, which form a mesh deeply embedded in the surrounding
tissue. If, on the one side, this capillary mesh-like structure is pretty regular, on the
other, it is not feasible to face its complete description. Homogenized models for
porous media can be used effectively to face this problem as done, for example,
in [56] for pulmonary alveoli circulation;
• a more detailed description of autoregulation mechanisms. Possible regulatory
laws could be borrowed from the works by Ursino and co-authors (see, e.g., [67]
and references therein) in the case of compartmental models, and from the works
by Arciero and co-authors (see, e.g., [10, 33]) in the case of a simplified repre-
sentative 1D/segment model of large/small arteries/veins and capillaries. In these
works, the vasoactive response of the arterioles is modelled via "regulatory vari-
ables" which depend on myogenic and metabolic stimuli according to phenomeno-
logical laws;
• inclusion of phenomenon of acute/chronic vessel remodelling. Acute remodelling
is recently taken into account in the work of David and co-authors (see, e.g., [51]
and references therein) for realistic network of the cerebral microcirculation in
the case of circle of Willis, and by Secomb and co-authors (see, e.g., [64] and
references therein) in the case where the capillaries are recruited according to the
increased metabolic requirements of the surrounding tissue;
• the presence of multiple inputs and outputs in the network. This is a very delicate
point, in view of the fact that imposing conditions at these boundaries requires the
availability of physiological data or the numerical coupling with enlarged circula-
tory models. This aspect is dealt with for example in [150];
• a detailed description of the 3D geometry of the network and of the surrounding
tissue. A 1D-3D model for tissue perfusion is presented in the works of [76]
and [47], where the vessels are described as 1D segments, coupled to form a 3D
graph network immersed in a 3D structure for the tissue.
168
Bibliography
[1] M. Abramowitz and I. A. Stegun. Handbook of mathematical functions. Washington DC: NBS, 1964.
[2] J. Ahmed, R.D. Braun, R. Dunn, and R.A. Linsenmeier. Oxygen distribution in the macaque retina. Invest.
Ophthalmol. Vis. Sci., 34(3):516–521, 1993.
[3] B. Alamouti and J. Funk. Retinal thickness decreases with age: an oct study. Br. J. Ophthalmol., 87(7):899–
901, 2003.
[4] V. A. Alder, J. Ben-Nun, and S.J. Cringle. PO2 profiles and oxygen consumption in cat retina with an occluded
retinal circulation. Invest. Ophthalmol. Vis. Sci., 31(6):1029–1034, 1990.
[5] V.A. Alder and S.J. Cringle. Vitreal and retinal oxygenation. Graefes. Arch. Clin. Exp. Ophthalmol.,
228(1):151–157, 1990.
[6] M. Aletti, J.F. Gerbeau, and D. Lombardi. Modeling autoregulation in three-dimensional simulations of
retinal hemodynamics. J. Model. Ophthalmol., 1, 2015.
[7] A.C. Amrite, H.F. Edelhauser, and U.B. Kompella. Modeling of corneal and retinal pharmacokinetics after
periocular drug administration. Invest. Ophthalmol. Vis. Sci., 49(1):320, 2008.
[8] B. Anand-Aptea and J. G. Hollyfielda. Developmental Anatomy of the Retinal and Choroidal Vasculature.
Cleveland, OH, Elsevier, 2010.
[9] R. Arai, I. Kimura, Y. Imamura, K. Shinoda, C.S. Matsumoto, K. Seki, M. Ishida, A. Murakami, and A. Mi-
zota. Photoreceptor inner and outer segment layer thickness in multiple evanescent white dot syndrome.
Graefes. Arch. Clin. Exp. Ophthalmol., 252(10):1645–1651, 2014.
[10] J. Arciero, A. Harris, B. Siesky, A. Amireskandari, V. Gershuny, A. Pickrell, and G. Guidoboni. Theoret-
ical analysis of vascular regulatory mechanisms contributing to retinal blood flow autoregulation. Invest.
Ophthalmol. Vis. Sci., 54(8):5584–5593, 2013.
[11] J. Arciero, A. Pickrell, B. Siesky, and A. Harris. Theoretical analysis of myogenic and metabolic responses
in retinal blood flow autoregulation. Invest. Ophthalmol. Vis. Sci., 53(14):6847–6847, 2012.
[12] R. Avtar and D. Tandon. Mathematical modelling of intraretinal oxygen partial pressure. Trop. J. Pharm.
Res., 7(4):1107–1116, 2008.
[13] E. Babolian, A. Azizi, and J. Saeidian. Some notes on using the homotopy perturbation method for solving
time-dependent differential equation. Math. Comp. Model., 50(1):213–224, 2009.
[14] R.K. Balachandran and V.H. Barocas. Computer modeling of drug delivery to the posterior eye: effect of
active transport and loss to choroidal blood flow. Pharmaceut. Res., 25(11):2685–2696, 2008.
[15] R. E. Bank, J. F. Buergler, W. Fichtner, and R. K. Smith. Some upwinding techniques for finite element
approximations of convection-diffusion equations. Numer. Math., 58(1):185–202, 1990.
[16] O.K. Baskurt, M.R. Hardeman, M.W. Rampling, and H.J. Meiselman. Handbook of hemorheology and
hemodynamics, volume 69 of Biomedical and health research. IOS Press, 1 edition, 2007.
[17] A. Bawazir, R. Gharebaghi, A. Hussein, and W.H.W. Hitam. Non-arteritic anterior ischaemic optic neuropa-
thy in Malaysia: a 5 years review. Int. J. Ophthalmol., 4(3):272–274, 2011.
169
Bibliography
[18] J. M. Beach, K. J. Schwenzer, S. Srinnivas, D. Kim, and J. S. Tiedeman. Oxymetry of retinal vessels by
dual-wavelength imaging: calibration and influence of pigmentation. J. Appl. Physiol., 86(2):748–758, 1999.
[19] D. A. Beard and J. B. Bassingthwaighte. Mathematical modelling of intraretinal oxygen partial pressure.
Ann. Biomed. Eng., 29(4):298–310, 2001.
[20] H.F. Becker, O. Polo, S.G. McNamara, M. Berthon-Jones, and C.E. Sullivan. Effect of different levels of
hyperoxia on breathing in healthy subjects. J. Appl. Physiol., 81(4):1683–1690, 1996.
[21] Y. Behzadi and T.T. Liu. An arteriolar compliance model of the cerebral blood flow response to neural
stimulus. Neuroimage, 25(4):1100, 2005.
[22] M. Bhargava, M.K. Ikram, and T.Y. Wong. How does hypertension affect your eyes? J. Hum. Hypertens.,
26(2):71–83, 2012.
[23] G. Birol, S. Wang, E. Budzynski, N.D. Wangsa-Wirawan, and R.A. Linsenmeier. Oxygen distribution and
consumption in the macaque retina. Am. J. Physiol. Heart Circ. Physiol., 293(3):H1696–H1704, 2007.
[24] J.J. Bishop, P.R. Nance, A.S. Popel, M. Intaglietta, and P.C. Johnson. Diameter changes in skeletal muscle
venules during arterial pressure reduction. Am. J. Physiol.: Heart Circ. Physiol., 279(1):H47, 2000.
[25] J. Bols, J. Degroote, B. Trachet, B. Verhegghe, P. Segers, and J. Vierendeels. A computational method to
assess the in vivo stresses and unloaded configuration of patient-specific blood vessels. J. Comput. Appl.
Math., 246:10, 2013.
[26] R.D Braun, R.A. Linsenmeier, and T.K. Goldstick. Oxygen consumption in the inner and outer retina of the
cat. Invest. Ophth. Vis. Sci., 36(3):542–554, 1995.
[27] V.A. Buchin and N.Kh. Shadrina. Regulation of the lumen of a resistance blood vessel by mechanical stimuli.
Fluid Dynamics, 45(2):211–222, 2010.
[28] D.G. Buerk, K.A. Barbee, and D. Jaron. Nitric oxide signaling in the microcirculation. Crit. Rev. Biomed.
Eng., 39(5):397–433, 2011.
[29] C. Cancelli and T.J. Pedley. A separated-flow model for collapsible-tube oscillations. J. Fluid Mech.,
157:375–404, 1985.
[30] S. Čanić, C. J. Hartley, D. Rosenstrauch, J. Tambaca, G. Guidoboni, and A. Mikelic. Blood flow in compliant
arteries: An effective viscoelastic reduced model, numerics and experimental validation. Ann. Biomed. Eng.,
34:575–592, 2006.
[31] S. Čanić and E. H. Kim. Mathematical analysis of the quasilinear effects in a hyperbolic model blood flow
through compliant axi-symmetric vessels. Math. Method Appl. Sci., 26(14):1161–1186, 2003.
[32] J. Caprioli and A. L. Coleman. Blood pressure, perfusion pressure, and glaucoma. Am. J. Ophthalmol.,
149(5):704–712, 2010.
[33] B.E. Carlson, J.C. Arciero, and T.W. Secomb. Theoretical model of blood flow autoregulation: roles of
myogenic, shear-dependent, and metabolic responses. Am. J. Physiol. : Heart Circ. Physiol., 295(4):H1572,
2008.
[34] S. Cassani. Blood circulation and aqueous humor flow in the eye: multi-scale modeling and clinical applica-
tions. PhD thesis, Indiana University-Purdue University at Indianapolis, 2016.
[35] P. Causin, G. Guidoboni, F. Malgaroli, R. Sacco, and A. Harris. Blood flow mechanics and oxygen transport
and delivery in the retinal microcirculation: multiscale mathematical modeling and numerical simulation.
Biomech. Model Mechanobiol., 15(3):525–542, 2016.
[36] P. Causin and F. Malgaroli. Blood flow repartition in distensible microvascular networks: Implication of
interstitial and outflow pressure conditions. J. Coupled Syst. Multiscale Dyn., 4(1):14–24, 2016.
[37] P. Causin, F. Malgaroli, and F. Zorzan. A mathematical model for metabolic autoregulation of bio-mechanical
processes in microvessels: Application to the inner retinal circulation. J. Mech. Med. Biol., 15(02):1540027,
2015.
[38] M.E. Charlson, C.G. de Moraes, A. Link, M.T. Wells, G. Harmon, J.C. Peterson, R. Ritch, and J.M. Liebmann.
Nocturnal systemic hypotension increases the risk of glaucoma progression. Ophthalmology, 121(10):2004–
2012, 2014.
[39] KW Chow and CC Mak. A simple model for the two dimensional blood flow in the collapse of veins. J.
Math. Biol., 52(6):733–744, 2006.
[40] G. Cicco, F. Giorgino, and S. Cicco. Wound healing in diabetes: hemoreological and microcirculatory aspects.
In Oxygen Transport to Tissue XXXII. Springer, 2011.
170
Bibliography
171
Bibliography
[65] J.M. Fullana and S. Zaleski. A branched one-dimensional model of vessel networks. J. Fluid Mech., 621:183–
204, 2009.
[66] Y.C. Fung. Biomechanics: circulation. Springer Science & Business Media, 2013.
[67] Giacomo Gadda, Angelo Taibi, Francesco Sisini, Mauro Gambaccini, Paolo Zamboni, and Mauro Ursino. A
new hemodynamic model for the study of cerebral venous outflow. Am. J. Physiol. : Heart Circ. Physiol.,
308(3):H217, 2015.
[68] P. Ganesan, S. He, and H. Xu. Analysis of retinal circulation using an image-based network model of retinal
vasculature. Microvasc. Res., 80(1):99 – 109, 2010.
[69] P. Ganesan, S. He, and H. Xu. Development of an Image-Based network model of retinal vasculature. Ann.
Biom. Eng., 38:1566–1585, 2010.
[70] R.A. Ganfield, P. Nair, and W.J. Whalen. Mass transfer, storage, and utilization of o2 in cat cerebral cortex.
An. J. Physiol.–Legacy Content, 219(3):814–821, 1970.
[71] R.J. Gaudana, M. Barot, A. Patel, V. Khurana, and A.K. Mitra. Barriers for posterior segment ocular drug
delivery. In Mitra, editor, Treatise on Ocular Drug Delivery, pages 68–95. Bentham Science, 2013.
[72] A. Geirsdottir, S.H Palsson, O.and Hardarson, O.B. Olafsdottir, J.V. Kristjansdottir, and E. Stefánsson. Retinal
vessel oxygen saturation in healthy individuals. Invest. Ophth. Vis. Sci., 53(9):5433–5442, 2012.
[73] A.L. Gerber, A. Harris, B. Siesky, E. Lee, T. J Schaab, A. Huck, and A. Amireskandari. Vascular dysfunction
in diabetes and glaucoma: A complex relationship reviewed. J. Glaucoma, 24(6):474–479, 2014.
[74] D. Gilbarg and N.S. Trudinger. Elliptic partial differential equations of second order, volume 224. Springer
Science & Business Media, 2001.
[75] D. Goldman and A.S. Popel. A computational study of the effect of capillary network anastomoses and
tortuosity on oxygen transport. J. Theor. Biol., 206(2):181–194, 2000.
[76] I.G. Gould and A.A. Linninger. Hematocrit distribution and tissue oxygenation in large microcirculatory
networks. Microcirculation, 22(1):1–18, 2015.
[77] J.E. Grunwald, C.E. Riva, J. Baine, and A.J. Brucker. Total retinal volumetric blood flow rate in diabetic
patients with poor glycemic control. Invest. Ophthalmol. Vis. Sci., 33(2):356–363, 1992.
[78] K. Gugleta, C. Zawinka, I. Rickenbacher, A. Kochkorov, R. Katamay, J. Flammer, and S. Orgul. Analysis
of retinal vasodilation after flicker light stimulation in relation to vasospastic propensity. Invest. Ophthalmol.
Vis. Sci., 47(9):4034–4041, 2006.
[79] G. Guidoboni, A. Harris, L. Carichino, Y. Arieli, and B. A. Siesky. Effect of intraocular pressure on the
hemodynamics of the central retinal artery: a mathematical model. Math Biosci. Eng., 11(3):523–546, 2014.
[80] G. Guidoboni, A. Harris, S. Cassani, J. Arciero, B. Siesky, A. Amireskandari, L. Tobe, P. Egan, I. Janulevi-
ciene, and J. Park. Intraocular pressure, blood pressure and retinal blood flow autoregulation: a mathematical
model to clarify their relationship and clinical relevance. Invest. Ophthalmol. Vis. Sci., pages IOVS–13, 2014.
[81] S. I. Gundersen, G. Chen, and A. F. Palmer. Mathematical model of NO and O2 transport in an arteriole
facilitated by hemoglobi based O2 carriers. Biophys Chem, 143(1–2):1–17, 2009.
[82] N. Haghjou, M.J. Abdekhodaie, Y.L. Cheng, and M. Saadatmand. Computer modeling of drug distribution
after intravitreal administration. World Acad. Sci. Eng. Technol., 53:706–716, 2011.
[83] N.B. Hamilton, D. Attwell, and C.N. Hall. Pericyte-mediated regulation of capillary diameter: a component
of neurovascular coupling in health and disease. Front. Neuroenergetics, 2, 2010.
[84] M. Hammer, T. Heller, S. Jentsch, J. Dawczynski, D. Schweitzer, S. Peters, K.U. Schmidtke, and U.A.
Muller. Retinal vessel oxygen saturation under flicker light stimulation in patients with nonproliferative
diabetic retinopathy. Invest. Ophthalmol. Vis. Sci., 53(7):4063–4068, 2012.
[85] S.H. Hardarson and E. Stefánsson. Retinal oxygen saturation is altered in diabetic retinopathy. Br. J. Oph-
thalmol., 96:560–563, 2012.
[86] A. Harris, C.P. Jonescu-Cuypers, L. Kagemann, T.A. Ciulla, and G.K. Krieglstein. Atlas of ocular blood
flow: vascular anatomy, pathophysiology, and metabolism. Butterworth-Heinemann (Elsevier), Philadelphia,
2003.
[87] L.M. Haugh, R.A. Linsenmeier, and T K. Goldstick. Mathematical models of the spatial distribution of retinal
oxygen tension and consumption, including changes upon illumination. Ann. Biomed. Eng., (1):19–36, 1989.
[88] R. H. Haynes. Physical basis of the dependence of blood viscosity on the tube radius. Am. J. Physiol.,
198(6):1193–1200, 1960.
172
Bibliography
[89] S.S. Hayreh. Role of retinal hypoxia in diabetic macular edema: a new concept. Graefes Arch. Clin. Exp.
Ophthalmol., 246(3):356–361, 2008.
[90] J.H. He. Homotopy perturbation technique. Comput. Methods Appl. Mech. Eng., 178:257 – 262, 1999.
[91] J.H. He. Homotopy perturbation method: a new nonlinear analytical technique. Appl. Math. Comput., 135(73
- 79), 2003.
[92] Z. He, A.J. Vingrys, J.A. Armitage, and B.V. Bui. The role of blood pressure in glaucoma. Clin. Exp. Optom.,
94(2):133–149, 2011.
[93] M. Heil and T.J. Pedley. Large post-buckling deformations of cylindrical shells conveying viscous flow. J.
Fluid Struct., 10(6):565, 1996.
[94] J.D. Hellums, P.K.Nair, N.S. Huang, and N.Ohshima. Simulation of intraluminal gas transport process in the
microcirculation. Ann. Biomed. Eng., 24:1–24, 1996.
[95] J.B. Hickam and R. Frayser. Studies of the retinal circulation in man: Observation on vessel diameter,
arteriovenous oxygen difference and mean circulation time. Circulation, 33:302–316, 1966.
[96] H. Ho, K. Mithraratne, and P. Hunter. Numerical simulation of blood flow in an anatomically-accurate
cerebral venous tree. IEEE Trans. Med. Imaging, 32(1):85–91, 2013.
[97] N.M. Holekamp, Y.B. Shui, and D.C. Beebe. Vitrectomy surgery increases oxygen exposure to the lens: a
possible mechanism for nuclear cataract formation. Am. J. Ophthalmol., 139(2):302–310, 2005.
[98] G.A. Holzapfel, G. Sommer, C.T Gasser, and P. Regitnig. Determination of layer-specific mechanical proper-
ties of human coronary arteries with nonatherosclerotic intimal thickening and related constitutive modeling.
Am. J. Physiol. Heart. Circ. Physiol., 289(5):H2048–H2058, 2005.
[99] S.D. House and P.C. Johnson. Diameter and blood flow of skeletal muscle venules during local flow regula-
tion. Am. J. Physiol. Heart Circ. Physiol., 250(5):H828–H837, 1986.
[100] P.M. Hughes, O. Olejnik, J.-E. Chang-Lin, and C.G. Wilson. Topical and systemic drug delivery to the
posterior segments. Adv. Drug Deliver Rev., 57(14):2010–2032, 2005.
[101] N.V. Iftimia, D.X. Hammer, C.E. Bigelow, D.I. Rosen, T. Ustun, A.A. Ferrante, D. Vu, and R.D. Ferguson.
Toward noninvasive measurement of blood hematocrit using spectral domain low coherence interferometry
and retinal tracking. Opt. Express, 14(8):3377–3388, 2006.
[102] R.K. Jain. Transport of molecules in the tumor interstitium: a review. Cancer Res., 47(12):3039–3051, 1987.
[103] S. Jean-Louis, J.V. Lovasik, and H. Kergoat. Systemic hyperoxia and retinal vasomotor responses. Invest.
Ophthalmol. Vis. Sci., 46:1714–1720, 2005.
[104] G.S. Kassab, J. Berkley, and Y.C.B. Fung. Analysis of pig’s coronary arterial blood flow with detailed
anatomical data. Ann. Biomed. Eng., 25(1):204–217, 1997.
[105] M. Kavdia and A.S. Popel. Wall shear stress differentially affects no level in arterioles for volume expanders
and hb-based o2 carriers. Microvasc. Res., 66(1):49–58, 2003.
[106] M.E. Kavousanakis, N.G. Kalogeropoulos, and D.T Hatziavramidis. Computational modeling of drug deliv-
ery to the posterior eye. Chem. Eng. Sci., 108:203–212, 2014.
[107] N.M. Kerr, S.S. Chew, and H.V. Danesh-Meyer. Non-arteritic anterior ischaemic optic neuropathy: a review
and update. J. Clin. Neurosci., 16(8):994–1000, 2009.
[108] T. Kida, S. Morishita, K. Kakurai, H. Suzuki, H. Oku, and T. Ikeda. Treatment of systemic hypertension
is important for improvement of macular edema associated with retinal vein occlusion. Clin. Ophthalmol.,
8:724780, 2014.
[109] H. Kim, M.J. Lizak, G. Tansey, K.G. Csaky, M.R. Robinson, P. Yuan, N.S. Wang, and R.J. Lutz. Study
of ocular transport of drugs released from an intravitreal implant using magnetic resonance imaging. Ann.
Biomed. Eng., 33(2):150–164, 2005.
[110] N. Kizilova, M. Hamadiche, and M. Gad-el Hak. Mathematical models of biofluid flows in compliant ducts.
Arch. Mech., 64(1):65–94, 2012.
[111] B. Koeppen and B. Stanton. Berne & Levy Physiology. Mosby Elsevier, Philadelphia, 2009.
[112] P. Kolar. Risk factors for central and branch retinal vein occlusion: a meta-analysis of published clinical data.
J. Ophthalmol., 2014:724780, 2014.
[113] U.B. Kompella and H.F. Edelhauser. Drug product development for the back of the eye, volume 2. Springer
Science & Business Media, 2011.
173
Bibliography
[114] T.E. Kornfield and E.A. Newman. Regulation of blood flow in the retinal trilaminar vascular network. J.
Neurosci., 34(34):11504–11513, 2014.
[115] S. Kotha and L. Murtomäki. Virtual pharmacokinetic model of human eye. Math Biosci., 253:11–18, 2014.
[116] P. Kozlovsky, U. Zaretsky, A.J. Jaffa, and D. Elad. General tube law for collapsible thin and thick-wall tubes.
J. Biomech., 47(10):2378–2384, 2014.
[117] G Krenz and C Dawson. Flow and pressure distributions in vascular networks consisting of distensible
vessels. Am. J. Physiol. : Heart Circ. Physiol., 284(6):H2192, 2003.
[118] N.A. Kudryashov and I.L. Chernyavskii. Numerical simulation of the process of autoregulation of the arterial
blood flow. Fluid dynamics, 43(1):32–48, 2008.
[119] W.D. Lakin, S.A. Stevens, B.I. Tranmer, and PL Penar. A whole-body mathematical model for intracranial
pressure dynamics. J. Math. Biol., 46(4):347–383, 2003.
[120] K.A. Lamkin-Kennard, D.G. Buerk, and D. Jaron. Interactions between no and O2 in the microcirculation: a
mathematical analysis. Microvasc. Res., 68(1):38–50, 2004.
[121] P. Lanzer. Mastering endovascular techniques: a guide to excellence. Lippincott Williams & Wilkins, 2007.
[122] J. Lau and R.A. Linsenmeier. Oxygen consumption and distribution in the long–evans rat retina. Exp. Eye
Res., 102:50–58, 2012.
[123] J.S. Lee and L.P. Lee. A density method for determining plasma and red blood cell volume. Ann. Biomed.
Eng., 20(2):195–204, 1992.
[124] J.R. Levick. An introduction to cardiovascular physiology. Butterworth-Heinemann, 2013.
[125] Z. Li, T. Yipintsoi, and J.B. Bassingthwaighte. Nonlinear model for capillary-tissue oxygen transport and
metabolism. Ann. Biomed. Eng., 25(4):604–619, 1997.
[126] S.J. Liao. Homotopy Analysis Method in Nonlinear Differential Equations. Springer, 2012.
[127] R.A. Linsenmeier. Effects of light and darkness on oxygen distribution and consumption in the cat retina. J.
Gen. Physiol., 88(4):521–542, 1986.
[128] R.A. Linsenmeier and R.D. Braun. Oxygen distribution and consumption in the cat retina during normoxia
and hypoxemia. J. Gen. Physiol., 99:177–197, 1992.
[129] D. Liu, N. B. Wood, N. Witt, A. D. Hughes, S. A. Thom, and X. Y. Xu. Computational analysis of oxygen
transport in the retinal arterial network. Curr. Eye Res., 34(11):945–956, 2009.
[130] I.A. Lubashevsky and V.V. Gafiychuk. Analysis of the optimality principles responsible for vascular network
architectonics. arXiv preprint adap-org/9909003, 1999.
[131] A. Luksch, G. Garhofer, A. Imhof, K. Polak, E. Polska, G.T. Dorner, S. Anzenhofer, M. Wolzt, and
L. Schmetterer. Effect of inhalation of different mixtures of O2 and CO2 on retinal blood flow. Br. J.
Ophthalmol., 86:1143–1147, 2002.
[132] Barot M., Bagui M., Gokulgandhi M.R., and Mitra A.K. Prodrug strategies in ocular drug delivery. medicinal
chemistry. Med. Chem., 8(4):753–768, 2012.
[133] F. Mac Gabhann, M. Demetriades, A, T. Deering, J.D. Packer, S.M. Shah, E. Duh, P.A. Campochiaro, and A.S.
Popel. Protein transport to choroid and retina following periocular injection: theoretical and experimental
study. Ann. Biomed. Eng., 35(4):615–630, 2007.
[134] F. Martinez, E. Furió, M.J. Fabià, A.V. Pérez, Gonzalez-Albert, G. Rojo-Martinez, M.T. Martinez-Larrad, F.J.
Mena-Martin, F. Soriguer, M. Serrano-Rios, F.J. Chaves, J.C. Martin-Escudero, J. Redòn, and M.J. Garcia-
Fuster. Risk factors associated with retinal vein occlusion. Int. J. Clin. Pract., 68(7):871–881, 2014.
[135] B. R. Masters. Fractal analysis of the vascular tree in the human retina. Annu. Rev. Biomed. Eng., 6:427–452,
2004.
[136] H. Metzger. Distribution of oxygen partial pressure in a two-dimensional tissue supplied by capillary meshes
and concurrent and countercurrent systems. Math. Biosci., 5(1):143–154, 1969.
[137] K. Michalska-Malecka, L. Slowinska-Lozynska, and W. Romaniuk. Influence of rheological factors on the
development of primary open angle glaucoma. Klin. Oczna, 114(2):135–137, 2012.
[138] A. Mikelic, G. Guidoboni, and S. Canic. Fluid-structure interaction in a pre-stressed tube with thick elastic
walls i: the stationary stokes problem. Netw. Heterog. Media, 2(3):397, 2007.
[139] P.J. Missel. Simulating intravitreal injections in anatomically accurate models for rabbit, monkey, and human
eyes. Pharm. Res., 29(12):3251–3272, 2012.
174
Bibliography
[140] D. Moore, A. Harris, D. WuDunn, N. Kheradiya, and B. Siesky. Dysfunctional regulation of ocular blood
flow: a risk factor for glaucoma? Clin. Ophthalmo.l, 2(4):849–861, 2008.
[141] JA Moore and CR Ethier. Oxygen mass transfer calculations in large arteries. Journal of biomechanical
engineering, 119(4):469–475, 1997.
[142] L. Müller and E.F. Toro. Enhanced global mathematical model for studying cerebral venous blood flow. J.
Biomech., 47(13):3361, 2014.
[143] L. Müller and E.F. Toro. A global multiscale mathematical model for the human circulation with emphasis
on the venous system. Int. J. Numer. Method Biomed. Eng., 30(7):681, 2014.
[144] C.D. Murray. The physiological principle of minimum work i. the vascular system and the cost of blood
volume. Proc. Natl Acad. Sci., 12(3):207–214, 1926.
[145] P.K. Nair, N.S. Huang, J.D. Hellums, and J.S. Olson. A simple model for prediction of oxygen transport rates
by flowing blood in large capillaries. Microvasc. Res., 39(2):203–211, 1990.
[146] S. Nakabayashi, T. Nagaoka, T. Tani, K. Sogawa, T.W. Hein, L. Kuo, and A. Yoshida. Retinal arteriolar
responses to acute severe elevation in systemic blood pressure in cats: role of endothelium-derived factors.
Exp. Eye Res., 103:63–70, 2012.
[147] M.T. Nicolela, S.M. Drance, S.J. Rankin, A.R. Buckley, and B.E. Walman. Color Doppler imaging in patients
with asymmetric glaucoma and unilateral visual field loss. Am. J. Ophthalmol., 121(5):502–510, 1996.
[148] P.R. Ninawe, D. Hatziavramidis, and S.J. Parulekar. Delivery of drug macromolecules from thermally re-
sponsive gel implants to the posterior eye. Chem. Eng. Sci., 65(18):5170–5177, 2010.
[149] M. S. Olufsen, C. S. Peskin, W. Y. Kim, E. M. Pedersen, A. Nadim, and J. Larsen. Numerical simulation and
experimental validation of blood flow in arteries with structured-tree outflow conditions. Ann. Biomed. Eng.,
28(11):1281–99, 2000.
[150] Qing Pan, Ruofan Wang, Bettina Reglin, Guolong Cai, Jing Yan, Axel R Pries, and Gangmin Ning. A one-
dimensional mathematical model for studying the pulsatile flow in microvascular networks. J. Biomech. Eng.,
136(1):011009, 2014.
[151] J. Park, P.M. Bungay, R.J. Lutz, J.J. Augsburger, R.W. Millard, A.S. Roy, and R.K. Banerjee. Evaluation of
coupled convective–diffusive transport of drugs administered by intravitreal injection and controlled release
implant. J. Controlled Release, 105(3):279–295, 2005.
[152] T.J. Pedley. The fluid mechanics of large blood vessels. 1980.
[153] T.J. Pedley, B.S. Brook, and R.S. Seymour. Blood pressure and flow rate in the giraffe jugular vein. Phil.
Trans. R. Soc. B, 351(1342):855–866, 1996.
[154] D.J. Pepple and H.L. Reid. Alterations in hemorheological determinants and glycated hemoglobin in black
diabetic patients with retinopathy. J. Natl Med. Assoc., 101(3):258–260, 2009.
[155] K. Perktold, M. Prosi, and P. Zunino. Mathematical models of mass transfer in the vascular walls. In
Cardiovascular Mathematics. Modeling and simulation of the circulatory system. Springer, 2009.
[156] R. Pittman. Regulation of Tissue Oxygenation. Morgan & Claypool Life Sciences, California, 2011.
[157] B.W. Pogue, J.A. O’Hara, C.M. Wilmot, K.D. Paulsen, and H.M. Swartz. Estimation of oxygen distribution in
rif-1 tumors by diffusion model-based interpretation of pimonidazole hypoxia and eppendorf measurements.
Radiation Research, 155(1):15–25, 2001.
[158] S.L. Polyak. The retina. Chicago: University of Chicago Press, 1941.
[159] A.S. Popel. Theory of oxygen transport to tissue. Crit. Rev. Biomed. Eng., 17(3):257–321, 1988.
[160] A.S. Popel and P.C. Johnson. Microcirculation and hemorheology. Annual review of fluid mechanics, 37:43,
2005.
[161] CJ Pournaras, E Rungger-Brandle, CE Riva, SH Hardarson, and E Stefansson. Regulation of retinal blood
flow in health and disease. Prog Retin Eye Res, 27(3):284–330, 2008.
[162] M. Pranevicius and O. Pranevicius. Cerebral venous steal: blood flow diversion with increased tissue pressure.
Neurosurgery, 51(5):1267, 2002.
[163] A. R. Pries, K. Ley, M. Claassen, and P. Gaehtgens. Red cell distribution at microvascular bifurcations.
Microvasc. Res., 38:81–101, 1989.
[164] A.R. Pries. Microcirculation abnormalities: assessment techniques. Medicographia, 25:231–236, 2003.
175
Bibliography
[165] A.R. Pries, B. Reglin, and T.W. Secomb. Structural adaptation of vascular networks role of the pressure
response. Hypertension, 38(6):1476–1479, 2001.
[166] A.R. Pries and T.W. Secomb. Microvascular blood viscosity in vivo and the endothelial surface layer. Am. J.
Physiol. Heart Circ. Physiol., 289(6):H2657–H2664, 2005.
[167] A.R. Pries, T.W. Secomb, P. Gaehtgens, and J.F. Gross. Blood flow in microvascular networks. experiments
and simulation. Circulation research, 67(4):826–834, 1990.
[168] A.R. Pries, T.W. Secomb, T. Gessner, M.B. Sperandio, J.F. Gross, and P. Gaehtgens. Resistance to blood flow
in microvessels in vivo. Circ. Res., 75(5):904–915, 1994.
[169] M. Quigley and S. Cohen. A new pressure attenuation index to evaluate retinal circulation: a link to protective
factors in diabetic retinopathy. Arch. Ophthalmol.-Chic., 117(1):84–89, 1999.
[170] A. Rachev. A model of arterial adaptation to alterations in blood flow. J. Elast., 61(1-3):83–111, 2000.
[171] A. Rachev, N. Stergiopulos, and J.-J. Meister. Theoretical study of dynamics of arterial wall remodeling in
response to changes in blood pressure. J. Biomech., 29(5):635–642, 1996.
[172] L. Ramm, S. Jentsch, S. Peters, R. Augsten, and M. Hammer. Investigation of blood flow regulation and oxy-
gen saturation of the retinal vessels in primary open-angle glaucoma. Graefes Arch. Clin. Exp. Ophthalmol.,
252(11):1803–1810, 2014.
[173] B. Ranguelov, D. Goranova, V. Tonchev, and R. Yakimova. Diffusion limited aggregation with modified local
rules.
[174] V.P. Ranta, E. Mannermaa, K. Lummepuro, A. Subrizi, A. Laukkanen, M. Antopolsky, L. Murtomäki,
M. Hornof, and A. Urtti. Barrier analysis of periocular drug delivery to the posterior segment. J. Control
Release, 148(1):42–48, 2010.
[175] V.P. Ranta and A. Urtti. Transscleral drug delivery to the posterior eye: prospects of pharmacokinetic model-
ing. Adv. Drug. Deliv. Rev., 58(11):1164–1181, 2006.
[176] J.A.G. Rhodin. Ultrastructure of mammalian venous capillaries, venules, and small collecting veins. J.
Ultrastruct. Res., 25(5):452–500, 1968.
[177] C. E. Riva, J. E. Grunwald, S. H. Sinclair, and B. L. Perring. Blood velocity and volumetric flow rate in
human retinal vessels. Invest. Ophthalmol. Vis. Sci., 26(8):1124–1132, 1985.
[178] C.E. Riva, J.E. Grunwald, and S.H. Sinclair. Laser doppler velocimetry study of the effect of pure oxygen
breathing on retinal blood flow. Invest. Ophthalmol. Vis. Sci., 24:47–51, 1983.
[179] C.E. Riva, C.J. Pournaras, and M. Tsacopoulos. Regulation of local oxygen tension and blood flow in the
inner retina during hyperoxia. J. Appl. Physiol., 61:592–598, 1986.
[180] M.W. Roos. Theoretical estimation of retinal oxygenation during retinal detachment. Comput. Biol. Med.,
37(6):890–896., 2007.
[181] W. M. Roos. Theoretical estimation of retinal oxygenation during retinal artery occlusion. Physiol. Meas.,
25(6):1523–1532, 2004.
[182] R. Sacco, L. Carichino, C. de Falco, M. Verri, F. Agostini, and T. Gradinger. A multiscale thermo-fluid
computational model for a two-phase cooling system. Comput. Methods Appl. Mech. Eng., 282:239–268,
2014.
[183] M. Satilmis, S. Orguel, B. Doubler, and J. Flammer. Rate of progression of glaucoma correlates with retrob-
ulbar circulation and intraocular pressure. Am. J. Ophthalmol., 135(5):664–669, 2003.
[184] G.W. Schmid-Schönbein. Biomechanics of microcirculatory blood perfusion. Annu. Rev. Biomed. Eng.,
1(1):73–102, 1999.
[185] G.W. et al. Schmid-Schönbein. Microvascular mechanics: hemodynamics of systemic and pulmonary mi-
crocirculation. In Jen-Shih Lee and Thomas C Skalak, editors, Microvascular mechanics: hemodynamics of
systemic and pulmonary microcirculation. Springer Science & Business Media, 2012.
[186] S Schröder, M Brab, GW Schmid-Schönbein, M Reim, and H Schmid-Schönbein. Microvascular network
topology of the human retinal vessels. Fortschritte der Ophthalmologie: Zeitschrift der Deutschen Ophthal-
mologischen Gesellschaft, 87(1):52–58, 1989.
[187] D. Schweitzer, M. Hammer, J. Kraft, E. Thamm, E. Konigsdorffer, and J. Strobel. In vivo measurement of
the oxygen saturation of retinal vessels in healthy volunteers. IEEE Trans Biomed Eng, 46(12):1454–1465,
1999.
176
Bibliography
[188] D. Shanthi, V. Ananthaswamy, and L. Rajendran. Analysis of non-linear reaction-diffusion processes with
michaelis-menten kinetics by a new homotopy perturbation method. Natural Science, 5(9):1034 – 1046,
2013.
[189] M. Sharan and A. Popel. A compartmental model for oxygen transport in brain microcirculation in the
presence of blood substitutes. J. Theor. Biol., 216:479–500, 2002.
[190] M. Simionescu, N. Simionescu, and G.E. Palade. Segmental differentiations of cell junctions in the vascular
endothelium. the microvasculature. J. Cell Biol., 67(3):863–885, 1975.
[191] I. Simonini and A. Pandolfi. Customized finite element modelling of the human cornea. PloS One,
10(6):e0130426, 2015.
[192] D.M Snodderly, R.S. Weinhaus, and J.C. Choi. Neural-vascular relationships in central retina of macaque
monkeys (macaca fascicularis). J. Neurosci., 12(4):1169–1193, 1992.
[193] W. Song, Q. Wei, W. Liu, T. Liu, J. Yi, N. Sheibani, A.A. Fawzi, R.A. Linsenmeier, S. Jiao, and H.F. Zhang.
A combined method to quantify the retinal metabolic rate of oxygen using photoacoustic ophthalmoscopy
and optical coherence tomography. Sci. Rep., 4, 2014.
[194] M.S. Stay, J. Xu, T.W. Randolph, and V.H. Barocas. Computer simulation of convective and diffusive trans-
port of controlled-release drugs in the vitreous humor. Pharm. Res., 20(1):96–102, 2003.
[195] E. Stefansson, H.G. Wagner, and M. Seida. Retinal blood flow and its autoregulation measured by intraocular
hydrogen clearance. Exp. Eye Res., 47:669–678, 1988.
[196] Ş. Ţ and S. Giovanzana. Image analysis of the normal human retinal vasculature using fractal geometry. HVM
Bioflux, 4(1):14–18, 2012.
[197] I. Tadjbakhsh and F. Odeh. Equilibrium states of elastic rings. J. Math. Anal. Appl., 18(1):59–74, 1967.
[198] T. Takahashi. Microcirculation in fractal branching networks. Springer, 2014.
[199] T. Takahashi, T. Nagaoka, H. Panagida, T. Saitoh, A. Kamiya, T. Hein, L. Kuo, and A. Yoshida. A mathe-
matical model for the distribution of hemodynamic parameters in the human retinal microvascular network.
J. Biorheol., 23(77–86):2999–3013, 2009.
[200] D. Thassu and G.J. Chader. Ocular drug delivery systems: barriers and application of nanoparticulate
systems. CRC Press, ISBN 9781439848005, 2012.
[201] SP Timoshenko and JN Goodier. Theory of Elasticity (3rd edit.). McGraw-Hill, New York, 1970.
[202] K. Tojo. A pharmacokinetic model for ocular drug delivery. Chem. Pharm. Bull., 52(11):1290–1294, 2004.
[203] N. M. Tsoukias, D. Goldman, A. Vadapalli, R. N. Pittman, and A. S. Popel. A computational model of oxygen
delivery by hemoglobin-based oxygen carriers in three-dimensional microvascular networks. J. Theor. Biol.,
248(4):657–674, 2007.
[204] S. Tsuboi and J.E. Pederson. Effect of plasma osmolality and intraocular pressure on fluid movement across
the blood-retinal barrier. Invest. Ophthalmol. Vis. Sci., 29(11):1747–1749, 1988.
[205] R.F. Tuma, W.N. Duran, and K. Ley, editors. Microcirculation. Elsevier.
[206] M. Ursino and C.A. Lodi. Interaction among autoregulation, co2 reactivity, and intracranial pressure: a
mathematical model. Am. J. Physiol. Heart Circ. Physiol., 274(5):H1715–H1728, 1998.
[207] V. Vullo. Circular Cylinders and Pressure Vessels, volume 3. Springer, 2014.
[208] N.D. Wangsa-Wirawan and R.A. Linsenmeier. Retinal oxygen: fundamental and clinical aspects. Arch.
Ophthalmol-Chic., 121(4):547–557, 2003.
[209] G.N. Watson. A treatise on the theory of Bessel functions. Cambridge university press, 1995.
[210] B. Weinberger, D.L. Laskin, D.E. Heck, and J.D. Laskin. Oxygen toxicity in premature infants. Toxicol. Appl.
Pharm., 181(1):60–67, 2002.
[211] R.N. Weinreb and A. Harris. Ocular blood flow in glaucoma. World Glaucoma Association Consensus Series.
Kugler Publications, 2009.
[212] R. M Werkmeister, N. Dragostinoff, S. Palkovits, R. Told, A. Boltz, R.A. Leitgeb, M. Gröschl, G. Garhöfer,
and L. Schmetterer. Measurement of absolute blood flow velocity and blood flow in the human retina by
dual-beam bidirectional Doppler Fourier-domain optical coherence tomography. Invest. Ophth. Vis. Sci.,
53(10):6062–6071, 2012.
[213] F. M. White. Viscous fluid flow. McGraw-Hill, 1974.
177
Bibliography
[214] T. A. Witten and L. M. Sander. Diffusion-limited aggregation, a kinetic critical phenomena. Phys Rev. Lett.,
47(19):1400–1403, 1981.
[215] T.Y. Wong and P. Mitchell. The eye in hypertension. Lancet, 369(9559):425–435, 2007.
[216] J. Yang, J.W. Clark, R.M. Bryan, and C.S. Robertson. Mathematical modeling of the nitric oxide/cgmp
pathway in the vascular smooth muscle cell. Am. J. Physiol. Heart Circ. Physiol., 289(2):H886–H897, 2005.
[217] G. F. Ye, T. W. Moore, D. G. Buerk, and D. Jaron. A compartmental model for oxygen-carbon dioxide
coupled transport in the microcirculation. Ann. Biomed. Eng., 22(5):464–479, 1994.
[218] G. F. Ye, T. W. Moore, and D. Jaron. Contributions of oxygen dissociation and convection to the behavior of
a compartamental oxygen transport model. Microvasc. Res., 46(1):1–18, 1993.
[219] R.Z. Zhang, A.A. Gashev, D.C. Zawieja, and M.J. Davis. Length-tension relationships of small arteries, veins,
and lymphatics from the rat mesenteric microcirculation. Am. J. Physiol. Heart Circ. Physiol., 292(4):H1943–
H1952, 2007.
[220] P. Zunino. Mathematical and numerical modelling of mass transfer in the vascular system. Phd dissertation,
EPFL, 2002.
178
Short Curriculum Vitae
Francesca MALGAROLI was born in 1988 in Borgomanero (Italy). After attending Liceo Scientifico "Galileo
Galilei", she obtained a Master Degree with mention in Mathematics in April 2013 at Università degli Studi di Mi-
lano, Italy. During her studies, she pursued a curriculum in Applied Mathematics and Numerical Analysis. In July
2013 she won a PhD position in Engineering Mathematics at Politecnico di Milano, Italy.
Her research interests, already the object of her degree thesis, deal with the mathematical modelling and numer-
ical simulation of microcirculatory networks, with particular focus on eye retina blood vessels. She is (co)-author of
the following scientific papers:
• P. Causin, G. Guidoboni, F. Malgaroli, R. Sacco, A. Harris. “Impact of blood flow on ocular pathologies:
can mathematical and numerical modeling help preventing blindness?", Progress in Industrial Mathematics
at ECMI 2014, Mathematics in Industry, Springer, 2015.
• Causin, P., Malgaroli, F., Zorzan, F, “A mechano-physiological Model of Metabolic Autoregulation in Eye
Retinal Microcirculation", proceedings of the 19th International Conference on Mechanics in Medicine and
Biology , 220-224,ISBN:9788890167515,2015.
• Causin, P., Guidoboni, G., Malgaroli, F., Sacco, R., Harris, A. “Blood flow mechanics and oxygen transport
and delivery in the retinal microcirculation: multiscale mathematical modeling and numerical simulation”,
Biomech Model Mechan, 1-18, 2015.
• Causin, P., Malgaroli, F. “A mathematical and computational model of blood flow regulation in microvessels:
application to the eye retina circulation”, J Mech Med Biol, 15 (2), 2015
• Causin, P., Malgaroli, F. “Blood flow repartition in distensible microvascular networks: Implication of inter-
stitial and outflow pressure conditions”, J Coupled Syst Multiscale Dyn, 4(1), 2016
• Causin, P., Malgaroli, F. “Mathematical assessment of drug build-up in the posterior eye following transscle-
ral delivery", Journal of Mathematics in Industry 6.1 (2016): 9.
• Causin, P., Malgaroli, F. “Mathematical modeling of local perfusion in large distensible microvascular net-
works", arXiv preprint arXiv:1610.02292, submitted to Comput Methods Appl Mech Eng, 2016
179