Upload
others
View
1
Download
0
Embed Size (px)
Citation preview
COMPUTATIONAL MODELS FOR LOCALIZED DRUG DELIVERY IN TUMORS
By
MAGDOOM MOHAMED KULAM NAJMUDEEN
A THESIS PRESENTED TO THE GRADUATE SCHOOLOF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT
OF THE REQUIREMENTS FOR THE DEGREE OFMASTER OF SCIENCE
UNIVERSITY OF FLORIDA
2011
c⃝ 2011 Magdoom Mohamed Kulam Najmudeen
2
To my mom, dad and family
3
ACKNOWLEDGMENTS
I take this opportunity to thank everyone who has helped me throughout this
journey. I would like to extend my gratitude to my advisor Dr. Malisa Sarntinoranont for
giving me an opportunity to do research with financial assistance, and helping me since
I joined UF. I thank all the members of my research group who have helped me with this
project in varying degrees. A special thanks goes to Dr. Gregory Pishko without whom
this project would not have been accomplished, I am truly indebted to him. I also want to
extend my thanks to my committee members : Dr. Brian Sorg and Dr.Tran-Son-Tay for
kindly accepting to be in my committee.
For my research, I would like to thank Dr. Dietmar Siemann, Dr. Lori Rice and Chris
Pampo for providing murine KHT sarcoma cells and tumor inoculation. I also thank
Dr. Thomas Mareci for providing MRI expertise and Garrett W. Astary for helping with
the DCE-MR experiments. I appreciate the help Dr. Gregory Pishko has offered in this
project, by providing me with tissue transport property maps and simulation results for
the non-voxelized model, along with segmented MR images. I also want to thank Dr.
Jung Hwan Kim for sharing his valuable experience on voxelized modeling.
More importantly, I thank my mom and dad for providing me with financial and moral
support. I also thank my friends and family members especially my uncles, cousins,
brothers and sisters for advising and helping me in difficult times. I would also like to
thank my friends at UF especially those from my undergraduate school in India for
providing me with valuable support when I came to this country for the first time. Finally,
I would like to thank all the professors I have interacted with for providing me with
beneficial knowledge, which has helped me become the person I am today.
4
TABLE OF CONTENTS
page
ACKNOWLEDGMENTS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
LIST OF TABLES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
LIST OF FIGURES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
ABSTRACT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
CHAPTER
1 INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131.2 Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.2.1 Specific Aim 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151.2.2 Specific Aim 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2 DEVELOPMENT OF VOXELIZED MODEL FOR SYSTEMIC DELIVERY INSOLID TUMORS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162.2 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.2.1 Estimation of spatial variation maps of vascular leakiness . . . . . 182.2.2 Mathematical Model . . . . . . . . . . . . . . . . . . . . . . . . . . 202.2.3 Computational Method . . . . . . . . . . . . . . . . . . . . . . . . . 222.2.4 Statistical Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.2.4.1 Root Mean Square Error . . . . . . . . . . . . . . . . . . 242.2.4.2 Pearson Product Moment Correlation Coefficient . . . . . 242.2.4.3 Error Histogram . . . . . . . . . . . . . . . . . . . . . . . 25
2.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 252.3.1 Sensitivity Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . 272.3.2 Validation Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3 APPLICATION OF VOXELIZED MODEL FOR CONVECTION-ENHANCEDDELIVERY IN A HIND LIMB TUMOR . . . . . . . . . . . . . . . . . . . . . . . 48
3.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 483.2 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.2.1 Mathematical Model . . . . . . . . . . . . . . . . . . . . . . . . . . 513.2.2 Computational Method . . . . . . . . . . . . . . . . . . . . . . . . . 53
3.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 553.3.1 Sensitivity Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
3.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
5
4 CONCLUSIONS AND FUTURE WORK . . . . . . . . . . . . . . . . . . . . . . 73
REFERENCES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
BIOGRAPHICAL SKETCH . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
6
LIST OF TABLES
Table page
2-1 Tissue and vascular parameters used for simulating distribution of Gd-DTPAfollowing bolus tail vein injection at the hind limb tumor in a mice. . . . . . . . . 33
2-2 Statistical parameters obtained while comparing voxelized and non-voxelizedmodel results for the baseline simulation in three animals. . . . . . . . . . . . . 34
2-3 Statistical parameters obtained while comparing voxelized and non-voxelizedmodel results for intermediate and fast arterial input function in animal I. . . . . 35
2-4 Comparison of tracer washout rates and root mean square error in tracerconcentration within the tumor volume between voxelized and non-voxelizedmodel results with experiment in three animals. . . . . . . . . . . . . . . . . . . 36
3-1 Tissue and vascular parameters used for simulating distribution of albuminfollowing CED at the hind limb tumor in a mice. . . . . . . . . . . . . . . . . . . 62
7
LIST OF FIGURES
Figure page
2-1 Normalized concentration of tracer in blood plasma approximated by differentAIFs used for sensitivity analysis . . . . . . . . . . . . . . . . . . . . . . . . . . 37
2-2 CFD compatible meshes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
2-3 Horizontal and vertical lines used for plotting the flow field and tracer transportin voxelized and non-voxelized models . . . . . . . . . . . . . . . . . . . . . . . 39
2-4 Contours of IFP predicted by voxelized and non-voxelized models, along withline plots along the horizontal and vertical bisectors in the mid-slice. . . . . . . 40
2-5 Contours of IFV predicted by voxelized and non-voxelized models along withline plots along the horizontal and vertical bisectors in the mid-slice. . . . . . . 41
2-6 Comparison of tracer concentration contours. Voxelized and non-voxelizedmodel compared with MR-derived tissue concentration at t = 5, 30, and 60min. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
2-7 Line plots comparing the predicted tracer concentration in the tissue by boththe models with experiment, along the horizontal and vertical bisectorsof mid-slice at t = 5, 30 and 60 min. . . . . . . . . . . . . . . . . . . . . . . . . 43
2-8 Error Histograms for flow and transport in baseline simulation for voxelizedmodel with respect to non-voxelized model. . . . . . . . . . . . . . . . . . . . . 44
2-9 Line plots comparing the IFP and IFV predicted by both the models for twodifferent AIF parameter sets (intermediate and fast) along the vertical bisectorof mid-slice. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
2-10 Line plots comparing the tracer concentration in the tissue predicted by boththe models for two different AIF parameter sets (intermediate and fast) alongthe vertical bisector of mid-slice at t = 5 and 20 min. . . . . . . . . . . . . . . . 46
2-11 Error Histograms for tracer concentration within the tumor for voxel andnon-voxel model results with respect to the experimental data at t = 5, 30 and60 min. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
3-1 Depiction of baseline CED simulation. . . . . . . . . . . . . . . . . . . . . . . . 63
3-2 Variation of scaled hydraulic conductivity with porosity for different values of m. 64
3-3 IFP and EFV contours at the tumor mid-slice for systemic and local infusion,along with a EFV cone plot colored by its magnitude for local infusion. . . . . . 65
8
3-4 Normalized tracer concentration contours at tumor mid-slice at t = 30, 60,and 120 min. Also included predicted evolution of distributed volume over timeshown by an iso-surface at the distribution volume threshold. . . . . . . . . . . 66
3-5 Variation of tissue distribution volumes with infusion volume for the whole legand tumor following CED of albumin (0.3 µL/min) at the center of the tumor . . 67
3-6 Comparison of IFP and EFV contours at the tumor mid-slice along with EFVcone plot colored by its magnitude, for infusions at m = 5 & 9. . . . . . . . . . . 68
3-7 Predicted evolution of distributed volume for infusions at the center of the tumorfor m = 5 and 9 at t = 30, 60, and 120 min. . . . . . . . . . . . . . . . . . . . . 69
3-8 Variation of tissue distribution volumes with infusion volume for the whole legand tumor following CED of albumin (0.3 µL/min) at the center of the tumorfor m = 0, 5 & 9. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
3-9 Comparison of normalized tracer concentration contours at tumor mid-slicefor infusions at the tumor-host interface and anterior end of the tumor at t =30, 60, and 120 min. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
3-10 Variation of tissue distribution volumes with infusion volume for the whole legand tumor following CED of albumin (0.3 µL/min) at the tumor-host tissueinterface and anterior end of the tumor with m = 0 . . . . . . . . . . . . . . . . 72
9
Abstract of Thesis Presented to the Graduate Schoolof the University of Florida in Partial Fulfillment of the
Requirements for the Degree of Master of Science
COMPUTATIONAL MODELS FOR LOCALIZED DRUG DELIVERY IN TUMORS
By
Magdoom Mohamed Kulam Najmudeen
August 2011
Chair: Malisa SarntinoranontMajor: Mechanical Engineering
Systemic drug delivery to malignant tumors involving macromolecular therapeutic
agents is challenging for many reasons. Amongst them is their chaotic microvasculature
which often leads to inadequate and uneven uptake in solid tumors. Tumors are known
to have highly tortuous, fenestrated, discontinuous vessels and large avascular areas.
Such an abnormal microvasculature is thought to cause heterogeneous extravasation of
drugs and elevated interstitial fluid pressures inside the tumor.
Localized drug delivery is increasingly being used to circumvent such obstacles and
convection-enhanced delivery (CED) which utilizes convection in addition to diffusion for
distributing macromolecules has emerged as a promising local drug delivery technique.
The focus of this thesis was to develop a three dimensional computational porous media
transport model for solid tumors based on voxelized modeling methodology, which
incorporates the actual tumor microvasculature from the data obtained through dynamic
contrast-enhanced magnetic resonance imaging (DCE-MRI). The model was used to
predict interstitial fluid flow and tracer transport in tumors.
First portion of the project was focused on the development and evaluation of
the voxelized model for tumor transport. The model was developed for predicting the
interstitial flow field and distribution of MR visible tracer (Gd-DTPA) in tumor following
bolus tail vein injection. The results of the voxelized model were compared with that
obtained from a previously developed CFD modeling approach using unstructured
10
meshes. Furthermore, simulated Gd-DTPA distribution within the tumor was compared
to MR-measured Gd-DTPA concentration data. The voxelized model was tested on three
tumors with its predictions compared against the non-voxelized model and experimental
results. Benefits of a voxel approach include less labor and less computational time.
Sensitivity of the model to changes in arterial input function (AIF) parameters was also
investigated. For comparison, statistical analysis and qualitative representation of both
model results were presented. The analysis indicated similarity in both the model results
with low root mean square error and high correlation coefficient. The voxelized model
captured features of the flow field and tracer distribution such as the high interstitial fluid
pressure (IFP) inside tumor and the heterogeneous distribution of tracer. Predictions of
tracer distribution by the voxelized approach resulted in low error when compared with
the MR-measured data over a 1 hr time course. The accuracy of the voxelized model
results with experiment and non-voxelized model predictions were maintained across
the tumors. The sensitivity of the model to changes in AIF parameters was found to be
similar to that of the previous model approach.
Secondly, the developed voxelized model was slightly modified for predicting the
interstitial flow field and distribution of albumin tracer following CED at the hind-limb
tumor in mice. The spatially varying transport properties were obtained via DCE-MRI
experiments following systemic delivery of MR visible tracer, as mentioned in the
previous paragraph. A point source was introduced in the governing equations to model
the local infusion. The model was able to capture the heterogeneous/asymmetric tracer
distribution and the linear variation of distribution volume with the infusion volume.
Sensitivity of the model to changes in hydraulic conductivity and catheter placement
were investigated. The albumin distribution was found to be sensitive to both the
parameters under study. Increasing the values of the hydraulic conductivity map
lowered the tumor IFP and raised the distribution volume within the whole leg. However
within the tumor, the distribution volume decreased with increasing value of hydraulic
11
conductivity, at later time points. The infusion at the tumor-host tissue interface resulted
in larger distribution volume compared to that at the center and anterior end of the
tumor, under baseline conditions. Within the tumor, the distribution volume was almost
identical for infusions at the interface and center of the tumor. This image-based model
thus serves as a potential tool for optimizing patient-specific cancer treatments and
exploring the effects of heterogeneous vasculature on tumor transport.
12
CHAPTER 1INTRODUCTION
1.1 Background
Tumor or neoplasm is an abnormal mass of tissue usually caused due to genetic
mutations. They can be classified into benign and malign tumors depending on their
ability to invade adjacent tissues. Malign tumors invade and destroy adjacent tissues
while benign tumors lacks the ability to metastasize. The onset of tumors are often
characterized by rapid formation of new blood vessels to supply nutrients to the
tumor cells. Angiogenesis in tumor tissues is different from that in the normal tissue,
tumor vasculature is irregular and often characterized by highly tortuous, fenestrated,
discontinuous vessels and large avascular areas [1, 25, 33, 40, 42]. Tumors are also
known to exhibit elevated interstitial fluid pressure (IFP), which is attributed to its
lack of lymphatics [60] along with the chaotic vasculature [7, 14]. There is significant
evidence for elevated IFP in tumors from the experiments performed by several
researchers [11, 13, 23, 48, 77]. These abnormalities form vascular and interstitial
barriers to the delivery of macromolecular therapeutic agents to tumors [6, 29].
Systemic drug delivery to tumors is often known to result in inadequate and uneven
uptake, thereby preventing the drug from reaching therapeutic concentrations at the
target site. The chaotic tumor microvasculature leads to heterogeneous extravasation
of drugs [20], thereby reducing its therapeutic efficiency. The high IFP increases the
drug transport away from the tumor into normal tissues and reduces the transcapillary
transport, causing undesirable side-effects and lower drug uptake in the tumor. Overall,
these characteristics of the tumor microenvironment hinder the systemic delivery of
therapeutic agents to tumor cells.
Localized drug delivery has emerged as a plausible alternative to systemic delivery
for transporting macromolecular therapeutic agents to the tumors [18, 22, 56, 72–74]. By
directly injecting into the tumor, this circumvents the previously mentioned vascular and
13
interstitial barriers and also reduces the side-effects associated with systemic exposure.
Amongst the available techniques, convection-enhanced delivery (CED) appears
promising because at a given time it can achieve larger distribution volumes than by
diffusion alone [2, 10]. In CED, an infusion pump delivers the drug at constant flow rate
or pressure thereby utilizing the bulk flow due to the infusion pressure difference, to
deliver and distribute macromolecules to larger volumes in the tissue. Since its advent,
CED has been widely used for in situ delivery of a wide range of substances including
nanoparticles [50], liposomes [44, 56], cytotoxins [57] and viruses [24, 62]. However
heterogeneous distribution remains as an obstacle for CED to tumors.
1.2 Objectives
The focus of this thesis was to develop a computational model for predicting drug
distributions following CED to tumors. Computational modeling has gained attention
partly because it could help in planning and optimizing patient-specific treatments.
Previous mathematical models of transport in tumors assume theoretical vasculature
and simpler geometries [7, 59, 64–66] neglecting the vascular heterogeneity. Given the
critical nature of the microvasculature in tumor drug delivery, our group developed a
framework accounting for the realistic tumor microvasculature using the data obtained
from dynamic contrast enhanced-magnetic resonance imaging (DCE-MRI) [53]. The
model accounting for the heterogeneous tumor microvasculature could potentially
help optimize patient-specific treatments with its realistic predictions, and understand
the biophysical IFP and interstitial fluid velocity (IFV) changes due to CED, which are
otherwise difficult to measure experimentally.
The model previously developed by our group for this purpose, involved complex
geometric re-construction which is time consuming and labor intensive. One of the
objectives of this project was to develop a simpler model for tumor transport based on
a voxelized modeling methodology. In this approach, tissue properties and anatomical
boundaries are assigned on a voxel-by-voxel basis using MRI data. These properties
14
are then incorporated into a porous media transport model to predict IFP, IFV, and
tracer transport, thereby allowing for quicker building of computational transport models
and rapid estimation of tracer distribution. This model avoids the complex geometric
reconstruction as the MR data is directly imported into the mesh.
1.2.1 Specific Aim 1
The first portion of the project was aimed at developing and studying the applicability
of the voxelized model for tumor transport. The voxelized model was developed for
predicting the distribution of systemically delivered MR visible tracer (Gd-DTPA) in
the hind limb of mice through bolus tail vein injection. The results of the model which
includes the predicted flow field and tracer transport were compared with those obtained
from a non-voxelized one [53]. A validation study for this approach was also conducted
by calculating the error between Gd-DTPA tissue concentrations within the tumor,
predicted using a voxelized model and those measured using MRI. Sensitivity of the
model to arterial input function (AIF) was also investigated. The model was tested with
three sets of animal data.
1.2.2 Specific Aim 2
The second portion of the project was focused on applying the developed voxelized
model for predicting the distribution of albumin tracer in the same tumor following CED
as opposed to systemic delivery. The governing flow and transport equations were
slightly modified to account for the point source and the voxelized methodology was
used to solve them. For sensitivity analysis, the effects of varying hydraulic conductivity
maps and catheter placement, on fluid flow and albumin transport were investigated.
Infusions were carried out separately at two different sites in the tumor namely at the
tumor-host tissue interface and anterior end of the tumor, in addition to the baseline
simulation at the center of the tumor. The model could serve as a potential tool
for optimizing patient-specific treatment and studying the effect of heterogeneous
vasculature on tumor transport.
15
CHAPTER 2DEVELOPMENT OF VOXELIZED MODEL FOR SYSTEMIC DELIVERY IN SOLID
TUMORS
2.1 Overview
Although enormous advancements have been made in the diagnosis and treatment
of cancers, targeted drug delivery to malignant tumors still remains a challenge.
Transport of macromolecular therapeutic agents in the tumor microvasculature plays
a vital role in the treatment of solid tumors [34, 35]. However, a major obstacle to
systemic transport in tumors is inadequate and uneven uptake, which is widely attributed
to the heterogeneous architecture of the tumor microvasculature [6]. Tumors are known
to contain highly tortuous, fenestrated, discontinuous vessels and large avascular
areas [1, 25, 33, 40, 42]. The resulting heterogeneous vasculature leads to irregular
perfusion [9, 32] which causes heterogeneous extravasation of therapeutic agents
across the blood vessel wall, depending on the pressure difference across the wall and
spatially varying vascular permeability [7, 32].
Another profound effect of abnormal vascular geometry, combined with a lack
of lymphatics [60] in tumors is thought to be the elevation of interstitial fluid pressure
(IFP) [7, 14]. Experiments performed by several researchers have revealed increased
IFP in tumors [11, 13, 23, 48, 77]. It has been also observed that IFP is uniform
throughout the center of the tumor and drops sharply at its periphery [11, 16]. However,
recent evidence also suggests a lesser uniform IFP inside the tumors [26]. A study
conducted by Hassid and his colleagues showed that the IFP inside ectopic human
non-small-cell lung cancer increased from the periphery inward, with a high plateau
inside the tumors. With the absence of pressure gradients in the center of tumor in
either case, convective transport of drugs is expected to be less than at the periphery
where pressure gradients exist, resulting in a hetergenous extravasation.
It is also expected that the interstitial fluid flow driven by the IFP gradient is affected.
Interstitial fluid velocity (IFV) within a human neuroblastoma was experimentally found
16
to increase from the center towards the periphery of the tumor [16]. From a modeling
study, elevated IFP is also thought to cause vascular constriction which may lead to
reduction in tumor blood flow [47], also the presence of a necrotic core was found to
have an adverse affect on the distribution of large, slow-diffusing molecules [8]. On the
whole, these characteristics of the tumor microenvironment hinder the systemic delivery
of therapeutic agents to tumor cells. Hence, quantification of extravasation and drug
distribution is paramount to developing successful treatment strategies.
Previous mathematical models of transport in tumors assume either uniformly
distributed or regular patterns of parallel and series blood vessels [7, 64–66] neglecting
the vascular heterogeneity. Jain and his colleagues modeled the effects of uniformly
distributed leaky blood vessels and minimally functioning lymphatics for the case
of a spherical solid tumor and showed how elevated IFP leads to heterogeneous
extravasation [36]. Pozrikidis developed a theoretical model to describe the blood
flow in which, tumor microvasculature was generated by branching capillaries using
deterministic and random parameters thus resulting in a capillary tree [54] . It should
be noted that tumor angiogenesis patterns in these previous blood vessel models are
theoretical and simulated based on rules to generate network structures.
Recently, computational fluid dynamics (CFD) approaches were used by our group
and others to study the extracellular transport in tumors [51–53, 63, 78]. In particular,
studies conducted by Pishko et.al. [53] accounted for realistic tumor vasculature by
using dynamic contrast enhanced-magnetic resonance imaging (DCE-MRI) data to
estimate the spatial variation of transport properties (rate transfer constant between
plasma and extracellular space, K trans and porosity, ϕ), which were mapped into a
unstructured mesh of a CFD model that solves for IFP, IFV and tracer transport. The
results of these studies are encouraging; however, the time-intensive labor involved in
the approach motivated us to develop a simpler model for tumor transport based on
a voxelized modeling methodology. Earlier, this methodology has been used by our
17
group to model interstitial transport in the rat spinal cord and brain during tissue infusion
[38, 39]. In this approach, tissue properties and anatomical boundaries are assigned on
a voxel-by-voxel basis using MRI data. These properties are then incorporated into a
porous media transport model to predict IFP, IFV, and tracer transport, thereby allowing
for quicker building of computational transport models and rapid estimation of tracer
distribution. This voxel method circumvents the laborious geometric reconstruction
involved in its non-voxelized counterpart by directly importing MRI data.
In this study, a voxelized model for systemic transport in tumors was developed
and its results were compared with those obtained from a non-voxelized one [53]. A
validation study for this approach was conducted by calculating the error between
Gd-DTPA tissue concentrations predicted using a voxelized model and those measured
using MRI. The model was applied to three tumors and its predictions were compared
as described previously. Sensitivity of the model to arterial input function (AIF) was also
investigated. The shape of the AIF determines the time variation of the concentration of
MR visible tracer in blood plasma. The choice of AIF is critical in the pharmacokinetic
modeling of tissue tranport properties [28]. A faster AIF signifies higher wash-out rate of
the tracer and vice-versa.
2.2 Methods
2.2.1 Estimation of spatial variation maps of vascular leakiness
DCE-MRI was used to obtain vascular leakiness maps. The lower hind limb of an
anesthetized mouse (C3H), inoculated with murine sarcoma cells (KHT) was used in
the MR experiment. Serial DCE-MR images, consisting of a T1-weighted spin-echo
sequence were acquired before and after contrast agent (tracer) administration. The
same MRI data as presented in Pishko et.al, [53] was used. The data consisted of 9
slices with a matrix of 192 × 96 voxels per slice. The size of each voxel was 0.104 ×
0.104× 1 mm3.
18
DCE-MRI measures the tissue uptake of a MR visible tracer, which in this case
is gadolinium-diethylene-triamine penta-acetic acid (Gd-DTPA, MW ∼ 590 Da), after
a systemic bolus tail vein injection. The tracer concentration in tissue and the method
to calculate K trans and ϕ were identical to that presented in Pishko et.al, [53]. The
tracer deposition in tissue was measured by signal enhancement which is defined
as the ratio of the signal intensities after and before injection of the tracer. This is
then mapped to the actual tracer concentration in the tissue (Ct) by assuming a linear
relationship between Ct and relaxation times (T1 & T2), and substituting it into the
standard spin-echo equation [49, 55, 69]. After algebraic manipulations, the following
expression for Ct was obtained with an added assumption that transverse-relaxation
contribution to signal is unity,
CMRI,t =1
R1
[1
TRln
S(0)
S(0)− S(CMRI,t).(1− e−TR/T10)− 1
T10
](2–1)
where CMRI,t is the tissue concentration of Gd-DTPA determined by MRI, R1 is the
longitudinal relaxivity of the tracer in water, TR is the time for recovery, S(CMRI,t) and
S(0) are the signal intensities at tracer concentrations CMRI,t and zero respectively and
T10 is the T1 relaxtion time without tracer.
Vascular leakiness characterised by K trans and ϕ were estimated using a
two-compartment kinetic model [67]. This model describes the exchange of tracer
between the plasma and tissue compartments in each voxel. The two compartment
model can be described by,
dCtdt= K transCp −
K trans
ϕCt (2–2)
where Ct and Cp are the concentrations of Gd-DTPA in tissue and blood plasma
respectively.
19
The tracer concentration in the blood plasma, Cp following a bolus injection can be
described by an AIF of biexponential decay:
Cp(t) = d[a1e
−m1t + a2e−m2t
](2–3)
where a1,m1 refers to the amplitude and rate constant of the fast equilibrium between
plasma and extracellular space respectively, a2,m2 refers to the amplitude and rate
constant of the slow component of kidney clearance respectively and d is the dose
of the bolus injection. The baseline washout parameters used were as follows: a1 =
3.99 kg/L, m1 = 0.114 min−1, a2 = 4.78 kg/L, and m2 = 0.0111 min
−1 [67, 76].
In order to study the effects of AIF parameters in the model, two different sets of AIF
parameters were also used, which are as follows : a1 = 9.2 kg/L, m1 = 0.23 min−1, a2 =
4.2 kg/L, and m2 = 0.05 min−1 described the fast AIF [27]; a1 = 13 kg/L, m1 =
0.30 min−1, a2 = 16 kg/L, and m2 = 0.026 min−1 described the intermediate AIF [4]. A
qualitative representation of the different AIFs are provided in Figure 2-1. In this figure,
Cp was normalized such that the initial concentration is the same for all the AIFs.
Knowing the Cp values from Equation (2–3), Equation (2–2) was then solved
analytically to find an expression for Ct(t) which was then fitted with the experimental
values (CMRI,t) at early time points (∼ 20 min) to obtain the K trans and ϕ maps. These
maps were incorporated into the porous media transport model to predict the tracer
distribution at later time points.
2.2.2 Mathematical Model
The tissue continuum was modeled as a porous media with continuity [53] and
momentum (Darcy’s law) equations given by,
∇.v = Ktrans
K transJVV
− Lp,lySLV(p − p
L) (2–4)
v = −K∇p (2–5)
20
where v is the IFV, K trans is the average value of K trans over tumor and host tissue voxels,
Lp,ly is the lymphatic vessel permeability, SL/V is the lymphatic vessel surface area
per unit volume which was set to zero in tumor tissue, p is the IFP, pL
is pressure in the
lymphatic vessels which was set to zero and K is the tissue hydraulic conductivity. JV /V
is the filtration rate of plasma per unit volume of tissue into the interstitial space which is
given by Starling’s law as follows [68],
JVV= Lp
S
V(pv − p − σ
T(πv − πi)) (2–6)
where Lp is the hydraulic conductivity of the microvascular wall, S/V is the blood
vessel surface area per unit volume, pv is the vascular fluid pressure, σT
is the osmotic
reflection coefficient for plasma proteins, πv ,πi are the osmotic pressures of the plasma
and interstitial fluid, respectively.
The first term on the right side of the continuity equation (Equation (2–4)) represents
the fluid flux across the microvascular wall per unit volume of the tissue. The second
term accounts for the lymphatic drainage from interstitial space per unit volume of tissue.
Transport of interstitial Gd-DTPA was solved using the convection and diffusion
equation for porous media [53],
∂Ct∂t+
vϕ.∇Ct −D∇2Ct = K trans
(Cp −
Ctϕ
)− Lp,ly
SLV(p − p
L)Ctϕ
(2–7)
where D is the diffusion coefficient for Gd-DTPA. The following assumptions are made
in the above equation : the diffusion coefficient is isotropic and uniform and that the
dispersion coefficient is much smaller than D and there are no binding interactions
between the molecules. The terms on the left side of the above equation refers to the
transient, convection and diffusion fluxes respectively. The first term on the right hand
side of the equation denotes the transvascular solute exchange and the second term
21
denotes the tracer outflux due to the lymphatics. The values of the parameters in the
above equations are listed in Table 2-1.
The MR image consists of voxels which niether belong to tumor or host tissue, i.e.
exterior voxels which correspond to surrounding air. In these voxels, the source terms in
continuity and transport equations (Equations (2–4) and (2–7)) and diffusivity was set to
zero.
2.2.3 Computational Method
The continuity, momentum and tracer transport equations were solved using
the CFD software package, FLUENT (version 6.3, Fluent, Lebanon, NH). For the 3D
computational tissue model, a rectangular volume (20 × 10 × 9 mm3) enclosing the
tumor was created and meshed with quadrilateral elements (voxels) of size equal to
the MRI resolution (0.104 × 0.104 × 1 mm3) using the meshing software (GAMBIT,
Fluent, Lebanon, NH), with one-to-one mapping between the CFD mesh and MR data.
In the non-voxelized model [53], the geometry was meshed using an unstructured grid
with approximately 2.7, 2.5 and 2.3 million tetrahedral elements for animals I, II and III
respectively (Figures 2-2A and 2-2B).
Governing equations were discretized with a control-volume based technique
using FLUENT as done with the non-voxel approach. Within FLUENT, an user defined
function was used to assign K trans and ϕ for each voxel in the mesh. For continuity
and tracer transport equations, a user defined flux macro was used to account for
the source terms. Standard pressure interpolation scheme was used to solve for
pressure and first order upwind method was used to solve for velocity and the transport
equations. The SIMPLE (Semi-Implicit Method for Pressure-Linked Equations [3])
pressure-velocity coupling method was chosen and convergence criterion was set to
1E-5. Initial conditions for tracer transport assumed no initial tracer in the tissue, Ct = 0.
A zero fluid pressure condition, p = 0, was applied along the cut ends and the remaining
outer boundaries of the geometry were assigned as wall.
22
There is one difference in the modeling strategy between voxelized and non-voxelized
models, the impermeability condition along the skin boundary in the voxelized model
was achieved by assigning hydraulic conductivity two orders of magnitude lower than
the normal tissue in the exterior voxels, while the non-voxelized model implements it
by directly assigning them with a wall boundary condition with zero normal flux. The
assignment of low hydraulic conductivity in the exterior voxels creates a material that is
resistant to fluid motion. For the chosen value of hydraulic conductivity at the exterior
voxels the mean velocity at the skin boundary was calculated to be close to zero (0.001
µm/s).
The effect of changing the bi-exponential arterial input function (AIF) parameters on
the solutions was studied to understand the sensitivity of the voxelized model compared
to its counterpart. The sensitivity analysis was performed only for animal I. Apart
from the baseline value, flow and transport for two different sets of AIF (Figure 2-1)
parameters was simulated. For the analysis, tracer concentration was simulated for
t ≤ 20 min and the data was compared at discrete time points, t = 5, 10 and 20 min.
2.2.4 Statistical Analysis
Quantitative methods compared IFP, IFV, and tracer concentration in tissue
predicted by both the models. Such an evaluation required a one-to-one mapping
between both the meshes (unstructured and cartesian) which was mathematically
cumbersome to derive, hence a set of elements common to both the meshes were
identified based on the location of their cell centers and used for the analysis. Assuming
the variations in dependent variables across different non-voxel elements within a
given voxel to be small, the criteria for matching was that the non-voxel element should
lie within the voxel compared with. The above criteria resulted in approximately 97%
match for animal I, 98 % for animal II and 98 % for animal III. After finding the matching
elements in both the meshes, the values of IFP, IFV, and tracer concentration in tissue,
in these elements were used for the analysis.
23
Additionally, a quantitative comparison was conducted between tracer concentration
in tumor tissue predicted by both models and MRI-obtained experimental data at later
time points. Gd-DTPA concentrations from experimental data and voxelized model were
mapped to points within the tumor boundary of the non-voxelized model to compare
distribution of tracer at a given time point. Various statistical measures were used to
ascertain the similarity between model predictions. These include root mean square
error, correlation coefficient and error histogram.
2.2.4.1 Root Mean Square (RMS) Error, ε
The error in the magnitude of dependent variables were measured using the root
mean square error which was defined as the square root of the average of the squares
of the error. The RMS error for IFP and IFV were computed as shown below. For IFV, in
addition to the magnitude, the RMS error of the angle between the two velocity vectors
were also calculated,
εx =
√√√√√ N∑j=1
(x jvox − x jnvox
)2N
(2–8)
Where x was replaced with IFP, IFV magnitude, and tracer concentration, N is
the total number of matching elements, vox refers to voxel value and nvox refers to
non-voxel value.
2.2.4.2 Pearson Product Moment Correlation Coefficient (PMCC), r
Correlation coefficient was used to measure the statistical relationships between
both the results. PMCC is a measure of linear dependence between two variables. It
assumes that the relationship between both the variables can be best described by a
linear function and it is defined as the ratio of covariance of the two variables and the
product of their standard deviations. The value of the coefficient ranges from -1 to 1.
A positive sign indicates that the variables increase and decrease together. A large
magnitude (close to 1) implies that there is a strong linear relationship between both the
variables. This is can summarized as follows,
24
r =
−1 indicates perfect negative correlation
0 indicates no correlation
1 indicates perfect positive correlation
2.2.4.3 Error Histogram
Error histograms were generated to provide a graphical representation of frequency
distribution of errors in the dependent variables, which in this case was the absolute
value of the difference between the computed values of voxelized and non-voxelized
model. A suitable range for the error was chosen and divided into equal sized intervals
(or bins). The number of occurrences of the error was then calculated for each bin and
represented as a bar plot.
2.3 Results
Tumor flow fields and tracer transport obtained using both computational approaches
were compared using statistical analysis for all the three animal data sets, and
qualitative presentation of the dependent variables (IFP, IFV and tracer concentration
in tissue) for animal I using contour plots at the mid-slice supplemented with line
plots along the horizontal and vertical bisectors at the mid-slice (Figure 2-3). For a
detailed description of predicted fluid flow, tracer transport and sensitivity analysis in the
non-voxelized tumor model, the reader is referred to Pishko et.al.,[53].
The IFP contour and line plots for the tumor predicted by the voxelized and
non-voxelized model are shown in Figures 2-4A to 2-4D. The voxelized model predicted
elevated IFP inside the tumor, pressure reached peak value (0.73 - 1.62 kPa in animal
I, 0.32 - 0.71 kPa in animal II and 0.39 - 0.87 kPa in animal III) at the tumor core and
rapidly decreased at the tumor boundary. As expected within the tumor, predicted
pressure gradients were lowest close to the tumor center (∼ 14.2, 38.5 and 11.2 Pa/mm
in animal I, II and III respectively) and highest (∼ 1136.5, 579.6 and 763.6 Pa/mm in
animal I, II and III respectively) near its periphery. The pressure pattern was captured
by the voxelized model. However, the line plots clearly indicated a difference in the
25
predicted pressures between both the models. The magnitude of peak IFP predicted by
the voxelized model was found to be 15 % higher than that of its counterpart in animal I,
29 % in animal II and 18 % in animal III.
Despite changes in predicted IFP between both the models, the IFV contours
and line plots (Figures 2-5A to 2-5D) indicated that the distributions predicted are
qualitatively similar with highest velocities (∼ 0.75, 0.40 and 0.42 µm/s in animal I, II and
III respectively) occurring along the tumor boundary near the cut ends. The computed
IFV values were found to be lower inside the tumor (∼ 0.03, 0.03 and 0.01 µm/s in
animal I, II and III respectively). The low velocity regions were also observed far away
from the tumor boundary.
Interstitial distribution of Gd-DTPA tracer was simulated at various times (t = 5,
30 and 60 min) after infusion. The predicted tracer distribution of both the models and
the actual experimental data, was heterogeneous with high concentration regions (∼
0.4, 0.18 and 0.19 mM in animal I, II and III respectively at t = 5 min) outside the tumor
(Figure 2-6). It can also be observed that lowest tracer concentration (∼ 0.03, 0.05 and
0.03 mM in animal I, II and III respectively at t = 5 min) occurs within the tumor. The
line plots (Figure 2-7) shows that the tracer extravasation appears to be less affected
by the differences in the flow field predicted by both the models. Conforming with the
statistical findings, the accuracy of voxelized model’s prediction with respect to its
non-voxel counterpart was maintained for all the times simulated. As time proceeds,
tracer concentration was reduced and the distribution became more uniform.
The statistical parameters comparing both the model predictions for all the three
animals are listed in Table 2-2. The statistics of the model results appeared similar
across the animals. The Pearson coefficient for IFP was high (r > 0.7) indicating
similar patterns in both the model predictions. The value of its RMS error reflected the
difference in the peak pressures predicted by both the models. The low RMS error
in IFV and the high correlation coefficient (r > 0.7) showed a reasonable degree of
26
similarity between both the model predictions. The RMS error in tracer concentration
was maximum at initial time points and decreases thereafter with time. However,
correlation coefficients did not change much with time. Error histograms for flow and
tracer transport (Figure 2-8) followed an exponential distribution with peak around zero.
2.3.1 Sensitivity Analysis
Results of sensitivity analysis are presented in terms of line plots along the vertical
bisector of the mid-slice for flow field and tracer concentration. Similar to the baseline
results, the IFP pattern predicted by the voxelized model matched with that of the
non-voxel model for all AIFs, although there are differences in the predicted magnitude
(Figures 2-9A and 2-9B). The predicted pressures for intermediate AIF were found to
be closely matching for both the models. The IFV predicted by non-voxelized model
matched well with that of its counterpart (Figures 2-9C and 2-9D). Concentrations of
Gd-DTPA predicted by the voxelized model roughly followed the non-voxelized one
(Figure 2-10). The accuracy of the predicted concentration did not seem to change with
AIFs and time.
Statistics of the sensitivity analysis are provided in Table 2-3. The correlation
coefficients for IFP across the AIF’s were almost identical although the RMS errors were
different reflecting the differences in predicted pressures. Highest and lowest pressures
were observed for the baseline and intermediate AIF respectively. It was observed that
PMCCs and RMS errors in IFV were similar for the intermediate and fast AIFs. Tracer
concentration statistics also exhibited a similar behaviour with almost identical RMS
error values and PMCCs across the AIFs. With increasing time, the RMS error decayed
for all the cases although PMCCs remain similar.
2.3.2 Validation Study
Qualitatively, a similar pattern of Gd-DTPA distribution and washout was observed
for the voxelized model, non-voxelized model, and experimental data over the course of
1 hr (Figure 2-6). High concentration regions were observed outside the tumor and at
27
the edge of the tumor just within the tumor boundary. Washout rate was compared by
calculating volume-averaged Gd-DTPA concentration within the tumor for various time
points and fitting the data into a mono-exponential function (Table 2-4). The voxelized
and non-voxelized model both compared well with the experimental data. RMS error
was calculated for both models throughout the entire tumor volume as well as error
frequency histograms (Figure 2-11) to illustrate the comparison of the models with the
experimental data in space and time. Both the voxelized and non-voxelized models
showed low RMS error and high error frequency close to zero. However, for animal III
there was slightly higher RMS error in the voxelized and non-voxelized model predictions
with the experiment, eventhough the washout rate was very close with the experimental
data.
2.4 Discussion
A voxelized modeling approach was used to study the transport of Gd-DTPA
following systemic injection in tumors. Benefits of this methodology include easier
and more rapid building of computational porous media transport models compared
to traditional CFD approaches which involves complex geometric reconstruction.
Thus the voxel model is less labor intensive and potentially simpler to implement.
Spatially-varying tissue transport properties and realistic anatomical tissue geometries
were incorporated into a three-dimensional, image-based computational model. The
porous media simulation predicted interstitial fluid pressure, interstitial fluid velocity, and
tracer transport through the tissue interstitium. These results were compared with that
obtained using a non-voxel approach [53]. The sensitivity of the voxelized model for
different AIFs was investigated and compared with the non-voxel model. The voxelized
and non-voxelized model’s predictions of tracer distribution within the tumor were
compared to MRI-determined tracer distribution and the voxelized model was further
evaluated with additional animal data.
28
The voxelized model predicted elevated IFP conforming with the experimental
observations [11, 13, 23, 48, 77] and previous modeling results [7, 8, 53, 78]. However,
it can be found that the IFP predicted by the voxelized model was higher than that
predicted by the non-voxelized model. The value of RMS error reflects this difference as
it can be interpreted as the standard deviation between two variables. This discrepancy
can be explained by the differences in the tumor volume in both the models. The tumor
volume approximated by the voxelized model was found to be higher than that of the
non-voxel model (39 % higher in animal I, 83 % in animal II and 36 % in animal III).
This is due to the differences in the meshing strategy of the models. The non-voxelized
model used variable sized elements (unstructured mesh) which likely approximates
the tumor volume slightly better than its voxel counterpart which relies only on fixed
size elements (cuboids). This effect was particularly more pronounced in animal II
which has the smallest tumor volume of all. Since IFP values are found to be correlated
with the tumor volume, with higher IFP for large tumors [23], the voxelized model with
higher tumor volume is expected to have IFP higher than the non-voxelized one. The
lower differences observed in the predicted IFP by both the models for animals II and III
compared to animal I, could be attributed to their actual differences in the tumor volume
approximated by both the models. The additional tumor volume for animals II and III in
the voxelized model was an order of magnitude lower than that in animal I, thus resulting
in smaller change. These differences in the predicted IFP by both the models does not
have much effect on the predicted extracellular flow which is driven by the IFP gradient
which was similar in both the cases. It should be noted that the correlation coefficient
can also be interpreted as the degree of similarity between the slopes of two variables,
in other words a similarity index for the gradients of the variables. From Table 2-2, it is
clear that they are high for all the three animals, thereby indicating the high degree of
similarity in the IFP gradient computed by both the models thus supporting this previous
29
argument. The exponential error distribution for IFP with high frequency near zero error
implies the decaying nature of the number of voxels with higher errors.
The IFV predicted by the voxelized model reflects previous experimental finding [16].
In the experimental study conducted by DiResta and his colleagues, IFV in human
neuroblastoma was found to increase from the center towards the periphery of the
tumor. This is due to the high pressure gradient at the tumor boundary which increases
the IFV. On the other hand a more uniform pressure distribution in the tumor core leads
to low velocities in those regions. Statistical parameters obtained for IFV indicate a
higher degree of similarity between the predictions of both the models. As mentioned
previously, the IFV driven by the IFP gradient is less affected by the changes in the
predicted pressure. This is also reflected in their error histograms, it can be observed
that the error decays more rapidly than that of IFP, thus indicating the high accuracy of
the voxelized model in predicting IFV.
Distribution of Gd-DTPA was heterogeneous due to spatially varying deposition and
limited interstitial transport by diffusion and convection. The low tracer concentration
inside the tumor is consistent with the reduced fluid filtration and high IFP. As the
concentration is advected through the velocity field, its correlation coefficient is similar to
that obtained for IFV. Error histograms also reflect this behaviour, a strong peak around
zero clearly shows the reliability of voxelized model in predicting the tracer concentration
despite some changes in the predicted flow field. It should be noted that, although there
are differences between the results obtained through both the models, the voxelized
model faithfully captures the tracer extravasation which is essential for any drug delivery
model.
In sensitivity analysis, the effects of varying AIF parameters were also investigated.
It has been found that the flow and transport are sensitive to these parameters [53].
Changes in the flow and transport can be mainly attributed to the differences in
the K trans and ϕ maps. The sensitivity of voxelized model compared with that of its
30
counterpart seems to be the same for different AIFs as indicated by similar correlation
coefficients and RMS errors. This analysis also demonstrates the applicability of the
model in a diverse set of conditions. Overall, the sensitivity of voxelized model was
similar to its non-voxelized counterpart.
The validation study revealed that the Gd-DTPA distribution results obtained
via non-voxelized and voxelized models were consistent with the experimental
observations. This became more clear when the voxelized model predictions were
similar to non-voxelized predictions and experiment, across the tumors. The slightly
high RMS error in animal III could be due to errors in AIF parameters which varies
across the tumors and to which the results are sensitive. However, the washout rate
was accurately predicted by the voxelized model. This outcome lends credence to
the usage of voxelized porous media tumor models for predictions of low molecular
weight tracers and drug distribution. However, matching the modeling results with the
actual experimental values is difficult due to the presence of a large number of model
parameters which need to be determined experimentally. Also the differences in the
grid sizes between the non-voxelized model (approximately 2.3-2.7 million elements)
and voxelized model (approximately 165,000 elements which is just 6-7% of that in the
non-voxel mesh) may also account for discrepancies between them. The low resolution
of the voxelized model is due to limitations of MRI resolution as its data are directly
mapped into the model. Non-voxelized models on the other hand are more flexible in
this aspect as they do not directly map the MR data, thus allowing for variable resolution.
The non-voxel model [53] was also used for extensive sensitivity analysis requiring
it to capture steep pressure gradients at the tumor boundary, hence the mesh size
was increased for attaining convergence in FLUENT. The current voxelized model was
aimed at gaining an overall understanding of the fluid flow and transport in tumors, and
providing a reliable alternative to the non-voxelized approach.
31
High correlation coefficients between the voxelized and non-voxelized model results
indicates that both the results are in agreement with each other. However, there is some
disparity in the results especially the IFP which can be attributed to one of the basic
differences between both the approaches, mesh structure. The voxelized model uses
an uniform and rectangular mesh while the non-voxelized model uses an unstructured
mesh. The type of mesh used can affect the solution in two ways : (1) tumor/host
tissue volume approximation, (2) resolution. Using the cartesian mesh, the voxelized
model approximates the tumor and host tissue volumes with rectangular elements
thereby neglecting curvature at tissue boundaries while the non-voxelized model with
its variable size elements can account for this. This results in slight differences in tumor
and host tissue volumes which in turn affects the solution as the tumor shape and size
are important factors determining the interstitial fluid flow [17, 58]. The mesh density
also affects the solution as it determines the discretization of the domain with better
resolution in finer meshes and vice-versa. In this aspect, the voxelized model has much
lesser mesh density compared with its non-voxelized couterpart resulting in a lower
resolution as mentioned earlier. It should however be noted that the usage of very fine
meshes is computationally intensive and time consuming. Despite these differences,
voxelized model was still able to capture key features in the flow and transport thus
making it a attractive alternate candidate for tumor modeling.
32
Table 2-1. Tissue and vascular parameters used for simulating distribution of Gd-DTPAfollowing bolus tail vein injection at the hind limb tumor in a mice.
Variable Description Value ReferencesLp(m/Pa.s) Vessel permeability 2× 10−11t ; 3× 10−12n; [53]S/V (m−1) Microvascular surface area per unit volume 20000t ; 7000n [7]Lp,lySL/V (m
−1) Lymphatic filtration coefficient 1× 10−7 [53]K(m2/Pa.s) Hydraulic conductivity 1.9× 10−12t ; 3.8× 10−13n [53]
7.7× 10−15epv(Pa) Microvascular pressure 2300 [53]πi(Pa) Osmotic pressure in interstitial space 3230t ; 1330n [53]πv(Pa) Osmotic pressure in microvasculature 2670 [53]σT(Pa) Average osmotic reflection coefficient for plasma 0.82t ; 0.91n [53]
D(m2/s) Diffusion coefficient of Gd-DTPA 1× 10−9 [53]
t - tumor, n - normal tissue, e - exterior.
33
Table 2-2. Statistical parameters obtained while comparing voxelized and non-voxelizedmodel results for the baseline simulation in three animals.
Variable Quantity Animal I (Baseline) Animal II Animal IIIε r ε r ε r
IFP Magnitude 167.09 Pa 0.95 44.45 Pa 0.97 38.72 Pa 0.97IFV Magnitude 0.07 µm/s 0.81 0.01 µm/s 0.89 0.02 µm/s 0.92
Direction 23.10◦ 0.72 25.64◦ 0.58 16.44◦ 0.74Ct At t = 5 min 0.10 mM 0.79 0.10 mM 0.71 0.08 mM 0.74
At t = 30 min 0.05 mM 0.79 0.06 mM 0.75 0.04 mM 0.77At t = 60 min 0.04 mM 0.78 0.03 mM 0.77 0.03 mM 0.77
34
Table 2-3. Statistical parameters obtained while comparing voxelized and non-voxelizedmodel results for intermediate and fast arterial input function in animal I.
Variable Quantity Intermediate AIF Fast AIFε r ε r
IFP Magnitude 47.48 Pa 0.99 119.88 Pa 0.99IFV Magnitude 0.12 µm/s 0.78 0.14 µm/s 0.78
Direction 18.22◦ 0.70 18.45◦ 0.70Ct At t = 5 min 0.10 mM 0.72 0.07 mM 0.68
At t = 10 min 0.07 mM 0.75 0.07 mM 0.72At t = 20 min 0.06 mM 0.75 0.03 mM 0.74
35
Table 2-4. Comparison of tracer washout rates and root mean square error in tracerconcentration within the tumor volume between voxelized and non-voxelizedmodel results with experiment in three animals.
Washout rate ofvolume-averaged
Gd-DTPA concentrationwithin tumor volume
(min−1)
RMS error for concentration withintumor volume (mM)
Animal Case t = 5 min t = 30 min t = 60 minI Experimental -0.031 – – –
Voxelized Model -0.022 0.072 0.039 0.031Non-voxelized Model -0.020 0.120 0.064 0.048
II Experimental -0.020 – – –Voxelized Model -0.021 0.070 0.060 0.036Non-voxelized Model -0.022 0.089 0.062 0.040
III Experimental -0.025 – – –Voxelized Model -0.022 0.295 0.258 0.034Non-voxelized Model -0.022 0.297 0.259 0.037
36
0 5 10 15 200
0.2
0.4
0.6
0.8
1
Time (minutes)
Nor
mal
ized
pla
sma
conc
entr
atio
n
BaselineIntermediateFast
Figure 2-1. Normalized concentration of tracer in blood plasma approximated bydifferent AIFs used for sensitivity analysis
37
0
5
10
15
20
0
2
4
6
8
10
0
2
4
6
8
10
A B
Figure 2-2. CFD compatible meshes. (A) Schematic of voxelized (cartesian) mesh (B)Unstructured mesh of reconstructed hind limb. Includes tumor (light green),skin (green), cut ends (yellow), and representation of mid-slice (dark blue).
38
0 5 10 15 200
2
4
6
8
10
A B
Figure 2-3. Horizontal and vertical lines used for plotting the flow field and tracertransport in voxelized (A) and non-voxelized (B) models
39
A B
0 5 10 15 200
200
400
600
800
1000
1200
1400
1600
Distance (mm)
IFP
(P
a)
Non VoxelVoxel
C
0 2 4 6 8 100
200
400
600
800
1000
1200
1400
1600
Distance (mm)
IFP
(P
a)
Non VoxelVoxel
D
Figure 2-4. Contours of IFP predicted by (A) voxelized model (B) non-voxelized model.Tumor and skin boundaries are overlaid on the contours. Also included, lineplots comparing the predicted IFP (C & D) by both the models along thehorizontal and vertical bisectors in the mid-slice respectively. The tumor andskin boundaries are represented by dashed and dash-dot lines respectively.
40
A B
0 5 10 15 200
0.1
0.2
0.3
0.4
0.5
0.6
Distance (mm)
IFV
(µm
/s)
Non VoxelVoxel
C
0 2 4 6 8 100
0.1
0.2
0.3
0.4
0.5
0.6
Distance (mm)
IFV
(µm
/s)
Non VoxelVoxel
D
Figure 2-5. Contours of IFV predicted by (A) voxelized model (B) non-voxelized model.Tumor and skin boundaries are overlaid on the contours. Also included, lineplots comparing the predicted IFV (C & D) by both the models along thehorizontal and vertical bisectors in the mid-slice respectively. The tumor andskin boundaries are represented by dashed and dash-dot lines respectively.
41
Exp
erim
enta
lt = 5 min t = 30 min t = 60 min
Voxe
lized
Non
-vox
eliz
ed
Figure 2-6. Comparison of tracer concentration contours. Voxelized and non-voxelizedmodel compared with MR-derived tissue concentration at t = 5, 30, and 60min. Tumor and skin boundaries are overlaid on the contours.
42
0 5 10 15 200
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
Distance (mm)
Ct (
mM
)
Non VoxelVoxelExperimental
A
0 2 4 6 8 100
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
Distance (mm)
Ct (
mM
)
Non VoxelVoxelExperimental
B
0 5 10 15 200
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
Distance (mm)
Ct (
mM
)
Non VoxelVoxelExperimental
C
0 2 4 6 8 100
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
Distance (mm)
Ct (
mM
)
Non VoxelVoxelExperimental
D
0 5 10 15 200
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
Distance (mm)
Ct (
mM
)
Non VoxelVoxelExperimental
E
0 2 4 6 8 100
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
Distance (mm)
Ct (
mM
)
Non VoxelVoxelExperimental
F
Figure 2-7. Line plots comparing the predicted tracer concentration in the tissue by boththe models with experiment, along the horizontal and vertical bisectors ofmid-slice at t = 5 (A & B), 30 (C & D), 60 (E & F) min respectively. The tumorand skin boundaries are represented by dashed and dash-dot linesrespectively.
43
0 200 400 600 800 10000
1000
2000
3000
4000
5000
6000
Pressure (Pa)
Fre
quen
cy
A
0 0.2 0.4 0.6 0.80
5000
10000
15000
20000
25000
Velocity magnitude (µm/s)
Fre
quen
cy
B
0 50 100 1500
2000
4000
6000
8000
10000
12000
Velocity direction (degrees)
Fre
quen
cy
C
0 0.2 0.4 0.6 0.8 10
5000
10000
15000
20000
Concentration (mM)
Fre
quen
cy
D
0 0.1 0.2 0.3 0.4 0.5 0.60
2000
4000
6000
8000
10000
12000
14000
Concentration (mM)
Fre
quen
cy
E
0 0.1 0.2 0.3 0.4 0.5 0.60
5000
10000
15000
Concentration (mM)
Fre
quen
cy
F
Figure 2-8. Error Histograms for flow (A, B & C) and transport at t = 5 (D), 30 (E) and 60(F) mins in baseline simulation for voxelized model with respect tonon-voxelized model
44
0 2 4 6 8 10350
400
450
500
550
600
650
700
Distance (mm)
IFP
(P
a)
Non VoxelVoxel
A
0 2 4 6 8 10350
400
450
500
550
600
650
700
Distance (mm)IF
P (
Pa)
Non VoxelVoxel
B
0 2 4 6 8 100
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
Distance (mm)
IFV
(µm
/s)
Non VoxelVoxel
C
0 2 4 6 8 100
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
Distance (mm)
IFV
(µm
/s)
Non VoxelVoxel
D
Figure 2-9. Line plots comparing the IFP (A & B) and IFV (C & D) predicted by both themodels for two different AIF parameter sets (intermediate and fast) along thevertical bisector of mid-slice respectively. The tumor and skin boundaries arerepresented by dashed and dash-dot lines respectively.
45
0 2 4 6 8 100
0.1
0.2
0.3
0.4
0.5
Distance (mm)
Ct (
mM
)
Non VoxelVoxel
A
0 2 4 6 8 100
0.1
0.2
0.3
0.4
0.5
Distance (mm)C
t (m
M)
Non VoxelVoxel
B
0 2 4 6 8 100
0.1
0.2
0.3
0.4
0.5
Distance (mm)
Ct (
mM
)
Non VoxelVoxel
C
0 2 4 6 8 100
0.1
0.2
0.3
0.4
0.5
Distance (mm)
Ct (
mM
)
Non VoxelVoxel
D
Figure 2-10. Line plots comparing the tracer concentration in the tissue predicted byboth the models for two different AIF parameter sets (intermediate and fast)along the vertical bisector of mid-slice at t = 5 (A & B) and 20 (C & D) minrespectively. The tumor and skin boundaries are represented by dashedand dash-dot lines respectively.
46
0 0.1 0.2 0.3 0.4 0.5 0.60
50
100
150
200
Concentration (mM)
Fre
quen
cy
Voxe
lized
t = 5 min
0 0.1 0.2 0.3 0.4 0.5 0.60
50
100
150
200
Concentration (mM)
Fre
quen
cy
t = 30 min
0 0.1 0.2 0.3 0.4 0.5 0.60
50
100
150
200
Concentration (mM)
Fre
quen
cy
t = 60 min
0 0.1 0.2 0.3 0.4 0.5 0.60
100
200
300
400
500
600
Concentration (mM)
Fre
quen
cy
Non
-vox
eliz
ed
0 0.1 0.2 0.3 0.4 0.5 0.60
100
200
300
400
500
600
Concentration (mM)
Fre
quen
cy
0 0.1 0.2 0.3 0.4 0.5 0.60
100
200
300
400
500
600
Concentration (mM)
Fre
quen
cy
Figure 2-11. Error Histograms for tracer concentration within the tumor for voxel andnon-voxel model results with respect to the experimental data at t = 5, 30and 60 min.
47
CHAPTER 3APPLICATION OF VOXELIZED MODEL FOR CONVECTION-ENHANCED DELIVERY
IN A HIND LIMB TUMOR
3.1 Overview
Cancer treatments based on systemic delivery of therapeutic agents are often
hindered due to the poor and uneven uptake of the drugs within the tumor. The unique
characteristics of the tumor microenvironment are known to be an important factor
affecting the efficacy of the anti-cancer treatments such as chemotherapy. Tumors are
known to exhibit elevated interstitial fluid pressure (IFP) [11, 13, 23, 48, 77] and irregular
microvasculature [19, 31, 33] which leads to inadequate uptake and heterogeneous
extravasation of drugs [20] respectively, consequently lowering their therapeutic index.
In the recent years, localized drug delivery has emerged as a plausible alternative
to systemic delivery for transporting macromolecular therapeutic agents to the
tumors [18, 22, 56, 72–74]. By directly injecting into the tumor, this circumvents
the previously mentioned vascular and interstitial barriers and also reduces the
side-effects associated with systemic exposure. Amongst the available techniques,
convection-enhanced delivery (CED) appears promising because at a given time it can
achieve larger distribution volumes than by diffusion alone [2, 10]. In CED, an infusion
pump delivers the drug at constant flow rate or pressure thereby utilizing the bulk flow
due to the infusion pressure difference, to deliver and distribute macromolecules to
larger volumes in the tissue.
Since its advent, CED has been used for in situ delivery of a wide range of
substances including nanoparticles [50], liposomes [44, 56], cytotoxins [57] and
viruses [24, 62]. Experimental studies on CED of liposomes into brain tumor (glioma)
in rats were encouraging, it was found that the technique effectively distributed the
liposomes in the tumor and the surrounding normal tissue [56]. A broad anisotropic
distribution was reported to have resulted from CED of cytotoxins into human gliomas [57].
Such an asymmetric distribution was also reported by Boucher and his colleagues in
48
their study involving intratumoral infusion of Evans blue-albumin in saline into sarcoma
HSTS 26T [12]. It should however be noted that spherically symmetric distributions for
colon adenocarcinoma LS174T were also reported in their study.
Computational modeling of CED has gained attention recently partly because it
could help in planning and optimizing patient-specific treatments. Earlier theoretical
models were focused on predicting drug distributions following CED in mediums like
agarose gels/ brain tissue [15, 38, 39, 43]. For tumors, Smith and Humphrey developed
a theoretical model in which infusions in a spherical tumor with a necrotic core was
simulated [59]. A main objective of their study was to analyze the effect of transvascular
fluid exchange on the flow field during the infusion. They found that the flow field
was very sensitive to the ratio of vascular conductivity and hydraulic conductivity,
and infusion close to the tumor was retarded by the outward flow. Weinberg and his
colleagues developed a finite element model to predict the distribution of doxirubicin
following intratumoral delivery [75]. However, the convective effects in the model were
replaced with a elimination coefficient instead of the actual interstitial fluid velocity (IFV).
It should be noted that these models utilized theoretical tumor microvasculature and
simplified tumor geometries.
Patient-specific computational porous media models, incorporating realistic
geometries and spatially varying transport properties obtained through MRI, for
predicting drug distributions have been developed by our group [38, 39, 45, 51–53, 78].
For tumors in particular, our group developed a framework which accounts for the actual
tumor microvasculature by using DCE-MRI data to estimate the spatial variation of
transport properties (rate transfer constant between plasma and extracellular space,
K trans and porosity, ϕ) which were included in a porous media model to solve for flow
and transport using computational fluid dynamics (CFD) techniques [45, 53, 78]. In
this study, this method was used to predict the distribution of albumin in a murine
sarcoma following CED as opposed to systemic delivery in the aforementioned papers.
49
In particular, CFD simulations were carried out based on a voxelized modeling approach
described in the previous chapter, where it was shown that the predicted flow field and
transport using this approach was similar to that of a more traditional approach based
on unstructured meshes. Earlier, this methodology has also been used by our group
to model CED in rat spinal cord and brain tissue [38, 39]. In this approach, anisotropic
tissue properties and anatomical boundaries are assigned on a voxel-by-voxel basis
using MRI data. These properties are then incorporated into a porous media transport
model to predict IFP, IFV and tracer concentrations. These voxelized models allow for
quicker building of computational transport models and rapid estimation of concentration
profiles.
In this study, a DCE-MRI based voxelized model was developed for predicting
albumin tracer distribution following CED in the lower limb of a mouse (C3H) inoculated
with murine sarcoma cells (KHT). The model accounting for the heterogeneous tumor
microvasculature could potentially help optimize patient-specific treatments with its
realistic predictions, and understand the biophysical IFP and IFV changes due to CED,
which are otherwise difficult to measure experimentally. A sensitivity analysis was
performed to study the effects of varying hydraulic conductivity maps and catheter
placement on fluid flow and albumin tracer transport. This was done to understand the
sensitivity of the model and relate them to key factors contributing to CED. The choice
of varying hydraulic conductivity is because of its direct influence on the tumor IFP and
convective flow field in intratumoral infusions. The higher values of hydraulic conductivity
were thought to reduce IFP thereby increasing the filtration of fluids and extravasation of
macromolecules [12]. The effect of catheter placement was known to be very important
in CED [2]. Studies involving infusions in different sites in the brain have revealed the
presence of a optimal site for achieving maximum distribution volume at the targetted
area [39, 43]. In the current study, infusions were carried out separately at two different
sites in the tumor namely at the tumor-host tissue interface and anterior end of the
50
tumor, in addition to the tumor center. The reason for choosing a site at the tumor-host
tissue interface was because of the presence of higher convective effects in that region
due to the sudden decrease in IFP which could result in higher IFV. The choice of an
infusion site at the anterior end and center of the tumor was to study the distribution at
various positions inside the tumor.
3.2 Methods
3.2.1 Mathematical Model
The study was divided into two parts : First the spatially-varying transport
properties of the KHT murine sarcoma were found through DCE-MRI following bolus
tail vein injection of MR visible tracer gadolinium-diethylene-triamine penta-acetic acid
(Gd-DTPA, MW ∼ 590 Da). The methods for obtaining DCE-MRI derived data such as
Gd-DTPA concentration in tissue, rate transfer constant (K trans) maps, and porosity (ϕ)
maps are identical to those in Pishko et.al., [53]. The second part involves incorporating
the above calculated variable transport properties into the computational porous media
model for flow and transport by CED.
The tissue continuum was modeled as a porous media and the governing equations
were solved at each voxel after assigning their respective K trans and ϕ values. The
continuity equation is given by,
∇.v = Q
VinfAt the infusion voxel (3–1)
=K trans
K transJVV
− Lp,lySLV(p − p
L) At all other voxels in tumor and host tissue (3–2)
where v is the IFV, Q is the infusion flow rate of albumin, Vinf is the volume of the
infused voxel, K trans is the average value of K trans in host and tumor tissue voxels, Lp,ly is
lymphatic vessel permeability, SL/V is the lymphatic vessel surface area per unit volume
which was set to zero in tumor tissue, p is the IFP and pL
is pressure in the lymphatic
vessels which was set to zero. JV /V is the filtration rate of plasma per unit volume of
51
tissue into the interstitial space which is given by Starling’s law as follows [68],
JVV= Lp
S
V(pv − p − σ
T(πv − πi)) (3–3)
here Lp is the hydraulic conductivity of the microvascular wall, S/V is the blood vessel
surface area per unit volume, pv is the vascular fluid pressure, σT
is the osmotic
reflection coefficient for plasma proteins, πv ,πi are the osmotic pressures of the plasma
and interstitial fluid, respectively.
The first term on the right side of the continuity equation for voxels that are not
infused with albumin (Equation (3–2)) represents the transvascular fluid flux across
the microvascular wall per unit volume of the tissue, scaled by the normalized K trans to
account for the heterogeneity in the model. The second term accounts for the lymphatic
drainage from interstitial space per unit volume of tissue.
For a porous medium, the momentum equation is given by Darcy’s law,
v = −K∇p (3–4)
where K is the hydraulic conductivity which is likely to be heterogeneous in tumors
and can vary with the local changes in porosity of the tissues [30, 41, 61, 70]. In
particular Lai and Mow [41], proposed an exponential variation of hydraulic conductivity
with deformation which in turn was related to porosity. By using a similar relation, the
exponential term was normalized with its mean value to ensure that the mean hydraulic
conductivity calculated over the tumor/host tissue voxels equals their baseline values.
The resulting expression is given as follows,
K =
K 0t em(ϕ+0.1)Nt
Nt∑i=1
em(ϕi+0.1)For tumor
K 0h em(ϕ+0.1)Nh
Nh∑i=1
em(ϕi+0.1)For host
(3–5)
52
where Nt ,Nh are the number of tumor and host tissue voxels respectively, K 0t ,K 0h are
the baseline hydraulic conductivities of tumor and host tissues respectively and m is an
empirical exponent.
Albumin (MW ∼ 66, 776 Da) is a non-binding and non-reacting macromolecule
which is widely used as tracer in CED studies. Assuming no tissue sources and sinks
since large molecular weight albumin is not expected to go back into the capillaries,
transport in the tissue was given by the convection and diffusion equation,
∂Ct∂t+
vϕ.∇Ct −Deff∇2Ct = 0 (3–6)
where Ct is the concentration of tracer in the tissue, Deff is the effective diffusivity
of albumin in the porous medium given by the following empirical relation based on
diffusion in porous media [21],
Deff = Dfreeϕn (3–7)
where Dfree is the self-diffusion coefficient of albumin in water and n is an empirical
exponent set to 4. The concentration in the equation was normalized using the following
relation,
C =CtC(t,i)ϕi
(3–8)
where C(t,i) and ϕi are the infusate concentration and porosity of the infused voxel
respectively. The values of the parameters in the governing equations are listed in
Table 3-1.
The MR image also consisted of voxels present outside the mouse which belong
neither to tumor or host tissue, i.e. exterior voxels. In these voxels, the whole source
term for the continuity equation and the diffusivity were set to zero.
3.2.2 Computational Method
The continuity, momentum and albumin transport equations were solved using the
CFD software package, FLUENT (version 12.0.16, ANSYS, Inc., Canonsburg, PA). For
53
the 3D computational tissue model, a rectangular volume (20 × 10 × 9 mm3) enclosing
the tumor was created and meshed with quadrilateral elements (voxels) of size equal
to the MRI resolution (0.104 × 0.104 × 1 mm3) using the meshing software (GAMBIT,
Fluent, Lebanon, NH) with one-to-one mapping between the CFD mesh and MR data.
The amount of tumor and host-tissue contained in the resulting volume were calculated
to be 51.49 and 741.16 mm3 respectively with the exterior voxels occupying the rest.
Governing equations were discretized with a control-volume based technique using
FLUENT. Darcy’s law was substituted for the conservation of momentum equation.
Within FLUENT, a user-defined function was used to assign K trans, porosity, hydraulic
conductivity and diffusivity for each voxel in the mesh. For the continuity equation, a
user-defined flux macro was used to account for the source terms. A standard pressure
interpolation scheme was used to solve for pressure and a second-order upwind
method was used to solve for the flow equations. The SIMPLEC (Semi-Implicit Method
for Pressure-Linked Equations Consistent [71]) pressure-velocity coupling method
was chosen. The transport equation was set-up using the user defined scalar (UDS)
equation in FLUENT and solved using first order upwind method. The convergence
criterion for all the three equations was set to 0.001.
Infusion simulations were carried out upto t = 2 hrs and the interstitial distribution
of albumin was simulated at intermittent time points, t = 5, 30, 60 and 120 mins.
Initial conditions for tracer transport assumed no tracer in the tissue, C = 0 except
at the infusion site which is one voxel (0.104 × 0.104 × 1 mm3), where it was set to a
normalized value of 1 at all the times through an user-defined function which was fed in
during the transport simulation. The distribution volume was calculated as the volume
occupied by voxels having an albumin concentration greater than 1 % of the infusion
concentration [10]. A zero fluid pressure condition, p = 0, was applied along the cut
ends and the remaining outer boundaries of the geometry were assigned as wall. The
impermeability condition along the skin boundary was achieved by assigning hydraulic
54
conductivity two orders of magnitude lower than the normal tissue, in the exterior voxels.
The assignment of low hydraulic conductivity in the exterior voxels creates a material
that is resistant to fluid motion. For the chosen value of hydraulic conductivity at the
exterior voxels the mean velocity at the skin boundary was calculated to be close to zero
(0.001 µm/s).
The infusion at the center of the tumor with locally constant hydraulic conductivity
(m = 0) was taken as the baseline case (Figure 3-1). For comparison, the flow field was
also simulated for the systemic delivery of albumin by neglecting the infusate source
term (Equation (3–1)), in the continuity equation. The effect of changing the hydraulic
conductivity was achieved by varying the empirical exponent (m) in the expression for
hydraulic conductivity. Apart from the baseline value (m = 0), flow and transport for two
different values of m = 5 and 9 (Figure 3-2) were also simulated. The effect of catheter
placement on the distribution was also studied through infusions at the tumor-host
tissue interface and anterior end of the tumor with m = 0, in addition to the baseline
simulation at the tumor center. The vessel permeability and diffusivity was not included
in the sensitivity analysis based on the results of our previous study on transport in
tumors [53], where these parameters were found to be insensitive to tracer transport.
Moreover, diffusion being a slow process, changes in diffusivity is not expected to affect
the tracer distribution in the small time window (2 hrs) under study. The changes in
flow rate is also not expected to affect the transport as the model does not have any
mechanism for back flow and other associated effects.
3.3 Results
The baseline results along with the sensitivity analysis for the model are provided.
The predicted IFP for systemic and local infusion are represented by contour plots
at the mid-slice of the tumor as shown in Figures 3-3A and 3-3B. The local infusion
at 0.3 µL/min increased the pressure at the infusion site by approximately 1.27 kPa.
The voxelized model predicted elevated IFP inside the tumor than the host tissue. The
55
contour plots reveal a local increase in IFP at the infusion site which masked the high
pressure inside the tumor compared to the host tissue. At the tumor mid-slice, the
magnitude of the pressure gradient was maximum at the infusion site (∼ 4.85 kPa/mm)
although significant values were also observed along the tumor-host tissue interface (∼
0.41 - 1.23 kPa/mm).
The extracellular fluid velocity (EFV,v
ϕ) for systemic and local infusion, is shown
by a contour along the tumor mid-slice (Figures 3-3C and 3-3D). This is further
supplemented by a cone plot depicting the velocity vectors colored by its magnitude
for the whole leg with local infusion (Figure 3-3E). Higher velocity regions were observed
near the infusion site for local infusion. At the tumor mid-slice for local infusion, peak
velocities were observed at the point of infusion (∼ 36 µm/s) followed by significant
velocities at the tumor-host tissue interface (∼ 0.25 - 6.15 µm/s). There was also
side-ways flow of the fluid along the skin boundary closer to the tumor.
The contours of the normalized albumin concentration at various time points, at
the tumor mid-slice are shown in Figures 3-4A to 3-4C. The predicted distribution of
albumin over time was asymmetric reflecting the anisotropic flow field. The effect of the
skin boundary condition near the tumor on the distribution pattern was evident at later
time points with a gradual outward flux of albumin along the skin boundary closer to the
tumor.
An iso-surface at the distribution volume threshold (0.01) for times t = 30, 60 and
120 mins shown in Figures 3-4D to 3-4F, depicts the evolution of the concentration
profile with time. The iso-surfaces confirms the asymmetric nature of the distribution
and the side-wise flux of albumin along the skin boundary near the tumor. After two
hours of infusion at 0.3 µL/min, albumin was distributed to approximately 58 % of the
tumor volume. The variation of distribution volume (Vd ) with infusion volume (Vi ) within
the whole leg and tumor in particular, is shown in Figure 3-5. The results data indicate
that the distribution volume varies linearly with the infusion volume for the whole leg.
56
However the variation was slightly non-linear within the tumor. The ratio Vd/Vi obtained
through linear fit was found to be 2.9 for the whole leg and 0.71 for the tumor.
3.3.1 Sensitivity Analysis
Similar to the baseline results, the model predicted higher IFP for m = 5 and 9
although the peak pressure values were different (Figure 3-6). The simulation results
indicated an 48 and 75 % reduction in the peak IFP from its baseline value for m = 5
and 9 respectively in the tumor mid slice. Increasing the value of m lowered the peak
IFP inside the tumor and the convection velocity became more heterogeneous with
increasing m. The increase in m appeared to reinforce fluid pathways with higher
porosities. The velocity vector plot reveals the increase in flow in the coronal plane at
m = 5 compared to the baseline value. This phenomenon became more visible at
m = 9 where there was a large outflow from the tumor. The fluid leakage across the skin
boundary closer to the tumor was present at both values of m.
The predicted evolution of the distribution volume over time for different values of
m is shown in Figure 3-7. The convective effects were apparent on the shapes of the
distribution volume, at m = 5 the distribution pattern tends to get more skewed into
the tumor than the baseline value. However as time proceeds, the albumin tracer tends
to go away from the tumor. A similar pattern was observed at m = 9 for initial time
points but the distribution got more heterogeneous and outward from the tumor as time
progressed. The distribution volume in the whole leg varied linearly with infusion volume
for m = 5 and 9 with slopes equal to 3.8 and 4.7 respectively (Figure 3-8A). However the
variation within the tumor, tends to become non-linear at later time points (Figure 3-8B).
At later time points, increasing the m decreased the distribution volume within the tumor.
This effect became more apparent for larger values of m. For m = 5, two hours of
infusion at 0.3 µL/min resulted in covering approximately 55 % of the tumor volume.
Whereas for m = 9, approximately 43 % of the tumor volume was covered by the tracer.
57
The effect of catheter placement on albumin distribution is shown as contours at the
tumor mid-slice in Figure 3-9. An asymmetric distribution was observed for infusions at
both the locations, tumor-host tissue interface and anterior end of the tumor. Infusion
at the interface tends to distribute albumin more along the dorsal side whereas at the
anterior end it was more skewed towards the anterior side of the leg. For the whole leg,
the results data indicated a linear variation of Vd with Vi , with higher distribution volume
for infusion at the interface than at the anterior end of the tumor (Figure 3-10A). Within
the tumor, infusion at the interface resulted in covering approximately 58 % of the tumor
while infusion at the anterior end resulted in approximately 18 % (Figure 3-10B).
3.4 Discussion
A computational model for predicting distribution of a macromolecular protein tracer
following CED in the hind limb tumor of a mice using voxelized modeling approach was
developed. This approach accounted for realistic tumor microvasculature and geometry,
and allowed for more easier and rapid building of computational porous media transport
model compared to traditional approaches utilizing unstructured meshes involving
complex geometric reconstruction. This makes the model less labor intensive and
easier to implement. Spatially-varying tissue transport properties based on the actual
heterogeneous tumor microvasculature, tissue structure and natural anatomical tissue
geometries were incorporated into a three-dimensional, image-based computational
porous media model. The model solves for interstitial fluid pressure, interstitial fluid
velocity, and albumin concentration through the tissue interstitium, following CED. The
sensitivity of the model for different hydraulic conductivity maps and catheter placements
were investigated.
The predicted IFP reflected the previous experimental findings which suggested
elevated pressures inside the tumor [11, 13, 23, 48, 77]. However, the infusion
induced a local pressure gradient thereby exhibiting the advantage convection gives
in distributing albumin to larger tissue volumes following CED. Except at the infusion
58
site, the pressure was uniform inside the tumor and dropped steeply in its periphery in
agreement with the previous findings [29]. Outside the tumor, the boundary condition
played a critical role in determining IFP. The close proximity of the tumor to one portion
of the impermeable skin resulted in a pressure gradient near that portion of the skin,
approximately four times higher than at the skin farther from the tumor.
The convection velocity field predicted by the model reflected the computed IFP as
the flow is driven by the pressure gradient. The high pressure gradient at the infusion
site and tumor-host tissue interface, resulted in higher velocities in those regions. In
addition, the presence of skin closer to the tumor was predicted to affect the flow field
causing side-ways flow of the fluid along the skin.
The linear variation of distribution volume with the infusion volume captured by
the model is in accordance with the previous experimental finding by Saito and his
colleagues [56]. They reported such a trend for CED of liposomes in rat gliomas. The
value of the ratio was known to depend on many factors but not limited to infusate
properties, extracellular matrix (ECM) among others, and a wide range of values
from 1 to 8.7 has been reported in the literature [2]. Furthermore, the distribution of
albumin was asymmetric and heterogeneous conforming with the previous experimental
findings [12, 46, 57]. Such a distribution is the result of the flow field which advects the
albumin in pathways of least resistance (higher porosity). The distribution pattern was
closely interlinked with the flow field with high concentration at the infusion site and
gradual leakage of albumin along the skin boundary closer to the tumor. At the end of
two hours, CED was able to cover approximately 58 % of the tumor volume thereby
exhibiting the effectiveness of the method in delivering macromolecular drugs.
In this study, we investigated the possibility of reducing the tumor IFP by increasing
the sensitivity of tissue hydraulic conductivity to tissue porosity. This was also done to
increase the heterogeneous transport. Mathematically this was implemented by varying
the empirical parameter m in the expression for hydraulic conductivity. Increasing the
59
hydraulic conductivity has been previously thought to reduce IFP and thus increase
extravasation of macromolecules [12]. The results of the sensitivity analysis at m = 9
indicated that the peak tumor IFP got reduced by approximately four times and the
resulting distribution volume increased by approximately 60 % from the baseline after
two hours of infusion. However this effect was not reflected in the computed distribution
volume within the tumor, which gradually reduced with time for higher values of m.
This is because of the very high reduction in the resulting tumor IFP, which directs
the interstitial fluid and albumin tracer away from the tumor. This demonstrates the
importance of measuring the parameter m for a given tumor, for achieving accurate
tracer distribution within the tumor.
The sensitivity analysis was extended to study tracer distribution at different
catheter positions, in an attempt to find an optimal placement which could maximize
distribution volume in the target site. The infusions were carried out at two other sites
in addition to the baseline position: tumor-host tissue interface and anterior end of
the tumor. For the given set of baseline parameters, we found that the infusion at the
tumor-host tissue interface produced the maximum distribution volume for the whole
leg. This is due to the presence of larger convective effects around the periphery as
opposing to just one at the infusion site. It should however be noted that the distribution
volume within the tumor was almost identical to the baseline value. The increased
convective effect apparently did not have significant effect on the distribution volume
within the tumor. The outward flow of albumin from the tumor for infusions at the
anterior end of the tumor is due to proximity of site to the cut ends of the tumor where
zero pressure boundary condition was specified. Similar pattern can be expected for
infusions at the posterior end of the tumor. By increasing the flow rate or infusion time
and/or testing additional infusion sites, the model can be used to find an optimal catheter
placement for a given tumor to achieve total coverage. In this way, the model can help in
surgical planning by providing effective treatment strategies on a case-by-case basis.
60
To our knowledge, this is the first image-based tumor model that incorporates the
actual tumor microvasculature and predicts heterogeneous/asymmetric drug distribution
following CED. Our model serves as a potential tool for conducting and optimizing
patient-specific treatments. Altering the extracellular matrix (ECM) of the tumor is being
explored by researchers as a possible technique to acheive better distribution [29].
Several compounds such as VEGF inhibitors, hyaluronidase, mannitol among others
were used to disrupt the heterogeneous tumor microvasculature and normalize it,
thereby improving drug delivery and efficacy [29, 37]. This model could be used to
study these effects once the transport properties (K trans and ϕ) of the resulting ECM
were found using the methods described in [53]. Although the results discussed in
this study were restricted to the hind limb tumor under study, it should be noted that
the applicability of voxelized model to a wide range of tumors is possible. The relative
ease in implementing the model and its reasonable predictions, makes it a promising
candidate for predicting drug distributions following CED in tumors.
61
Table 3-1. Tissue and vascular parameters used for simulating distribution of albuminfollowing convection-enhanced delivery at the hind limb tumor in a mice.
Variable Description Value ReferencesLp (m/Pa.s) Vessel permeability 2× 10−11t ; 3× 10−12n; [53]S/V (m−1) Microvascular surface area per unit volume 20000t ; 7000n [7]Lp,lySL/V (m−1) Lymphatic filtration coefficient 1× 10−7 [53]K 0 (m2/Pa.s) Baseline hydraulic conductivity 1.9× 10−12t ; 3.8× 10−13n [53]
7.7× 10−15epv (Pa) Microvascular pressure 2300 [53]πi (Pa) Osmotic pressure in interstitial space 3230t ; 1330n [53]πv (Pa) Osmotic pressure in microvasculature 2670 [53]σT
(Pa) Average osmotic reflection coefficient for plasma 0.82t ; 0.91n [53]Dfree (m2/s) Self diffusion coefficient of albumin 5.8× 10−11 [5]Q (µL/min) Infusion flow rate 0.3 [39]
t - tumor, n - normal tissue, e - exterior.
62
Figure 3-1. Depiction of baseline CED simulation. The size of the infusion needle wasexaggerated for clarity. Includes skin (green), tumor (blue), tumor mid-slice(red) and infusion needle (magenta)
63
0 0.1 0.2 0.3 0.4 0.50
1
2
3
4
5
6
7
8
9
10
φ
K/K
0
m=0m=5m=9
Figure 3-2. Variation of scaled hydraulic conductivity with porosity for different values ofm.
64
0 5 10 15 200
2
4
6
8
10
A
0 5 10 15 200 Pa
3000 Pa
B
0 5 10 15 200
2
4
6
8
10
C D
E
Figure 3-3. Interstitial fluid pressure (IFP) and extracellular fluid velocity (EFV) withsystemic (A & C) and local (B & D) infusion described by its contours at thetumor mid slice. Tumor and skin boundaries are overlaid on the contours andthe infusion site is shown by a plus sign. In the bottom, a EFV cone plot (E)colored by its magnitude for local infusion. Includes point source (blacksphere), tumor (blue) and skin (green)
65
0 5 10 15 200
2
4
6
8
10
A t = 30 min0 5 10 15 20
B t = 1 hr
0 5 10 15 200
0.3
C t = 2 hr
D t = 30 min E t = 1 hr F t = 2 hr
Figure 3-4. On the top, normalized tracer concentration contours at tumor mid-slice at t= 30, 60, and 120 min. Tumor and skin boundaries are overlaid on thecontours and the infusion site is shown by a plus sign. On the bottom,predicted evolution of distributed volume over time shown by an iso-surfaceat the distribution volume threshold. Includes tumor (blue), skin (green) anddistributed volume (red)
66
0 10 20 30 400
20
40
60
80
100
120
Vd,leg
= 2.9Vi+0.4
Vd,tum
= 0.71Vi+6.2
Infusion volume in µL
Dis
trib
utio
n vo
lum
e in
mm
3
Whole legTumor
Figure 3-5. Variation of tissue distribution volumes with infusion volume for the whole legand tumor following CED of albumin (0.3 µL/min) at the center of the tumor.Includes equation for the linear fit on the data.
67
m = 5 m = 9
0 5 10 15 200
2
4
6
8
10
A
0 5 10 15 200 Pa
1000 Pa
B
0 5 10 15 200
2
4
6
8
10
C D
E F
Figure 3-6. Comparison of interstitial fluid pressure (IFP, A & B), extracellular fluidvelocity (EFV, C & D) contours at the tumor mid-slice for infusions at m = 5 &9 respectively. Tumor and skin boundaries are overlaid on the contours andthe infusion site is shown by a plus sign. The EFV cone plots (E & F) coloredby its magnitude for m = 5 and 9 is also shown. Includes point source (blacksphere), tumor (blue) and skin (green)
68
t = 30 min t = 60 min t = 120 min
m=
5m
=9
Figure 3-7. Predicted evolution of distributed volume for infusions at the center of thetumor for m = 5 and 9 at t = 30, 60, and 120 min. Includes tumor (blue), skin(green) and distributed volume (red)
69
0 10 20 30 400
20
40
60
80
100
120
140
160
180
Vd,m=0
= 2.9Vi+0.4
Vd,m=5
= 3.8Vi−2.7
Vd,m=9
= 4.7Vi−5.1
Infusion volume in µL
Dis
trib
utio
n vo
lum
e in
mm
3
m = 0m = 5m = 9
A
0 10 20 30 400
5
10
15
20
25
30
Vd,m=0
= 0.71Vi+6.2
Vd,m=5
= 0.66Vi+7.1
Vd,m=9
= 0.47Vi+7.8
Infusion volume in µL
Dis
trib
utio
n vo
lum
e in
mm
3
m = 0m = 5m = 9
B
Figure 3-8. Variation of tissue distribution volumes with infusion volume for the whole leg(A) and tumor (B) following CED of albumin (0.3 µL/min) at the center of thetumor for m = 0, 5 & 9. Includes equation for the linear fit on the data.
70
t = 30 min t = 60 min t = 120 min
0 5 10 15 200
2
4
6
8
10
Inte
rface
0 5 10 15 20
0 5 10 15 200
0.3
0 5 10 15 200
2
4
6
8
10
Ant
erio
r
0 5 10 15 20
0 5 10 15 200
0.3
Figure 3-9. Comparison of normalized tracer concentration contours at tumor mid-slicefor infusions at the tumor-host interface and anterior end of the tumor at t =30, 60, and 120 min. Tumor and skin boundaries are overlaid on thecontours and the infusion site is shown by a plus sign in a circle.
71
0 10 20 30 400
20
40
60
80
100
120
Vd,center
= 2.9Vi+0.4
Vd,interface
= 3Vi−0.88
Vd,anterior end
= 2.1Vi+4.4
Infusion volume in µL
Dis
trib
utio
n vo
lum
e in
mm
3
CenterInterfaceAnterior end
A
0 10 20 30 400
5
10
15
20
25
30
Vd,center
= 0.71Vi+6.2
Vd,interface
= 0.72Vi+6.1
Vd,anterior end
= 0.18Vi+3.8
Infusion volume in µL
Dis
trib
utio
n vo
lum
e in
mm
3
CenterInterfaceAnterior end
B
Figure 3-10. Variation of tissue distribution volumes with infusion volume for the wholeleg (A) and tumor (B) following CED of albumin (0.3 µL/min) at thetumor-host tissue interface and anterior end of the tumor with m = 0.Includes equation for the linear fit on the data.
72
CHAPTER 4CONCLUSIONS AND FUTURE WORK
A computational model for predicting the distribution of macromolecular drug
following convection-enhanced delivery (CED) in the hind limb tumor of a mice using
voxelized modeling approach was developed. The approach accounted for realistic
tumor microvasculature and geometry, and allowed for more easier and rapid building
of computational porous media transport model compared to traditional approaches
utilizing unstructured meshes involving complex geometric reconstruction. This
makes the model less labor intensive and easier to implement. Spatially-varying
tissue transport properties based on the actual heterogeneous tumor microvasculature,
tissue structure and natural anatomical tissue geometries were incorporated into a
three-dimensional, image-based computational porous media model. The model solves
for interstitial fluid pressure, interstitial fluid velocity, and drug concentration through the
tissue interstitium, following injection. This framework was previously evaluated with
experiment and predictions from a traditional CFD approach for systemic delivery of MR
visible tracer.
The first portion of this thesis demonstrated the applicability of the voxelized model
for predicting tumor transport. This was done by comparing its results with that obtained
from a previously developed CFD modeling approach using unstructured meshes for
systemic delivery of the tracer, using statistical methods and qualitative presentation.
The resulting analysis indicated similarity in both the model results with low root mean
square error and high correlation coefficient. The voxelized model also captured typical
features of the flow field and tracer distribution in the tumor interstitium such as the high
interstitial fluid pressure (IFP) inside tumor and the heterogeneous distribution of tracer.
The obtained tracer distribution within the tumor was also similar to MR-measured
tracer concentration data. Furthermore, the accuracy of the voxelized model results
with experiment and non-voxelized model predictions were maintained across the three
73
tumors under study. Sensitivity of the voxelized and non-voxelized model to changing
arterial input function (AIF) parameters was also found to be similar.
In the second portion of the thesis, this model was slightly modified to predict
the tracer distribution following CED. The model was able to capture the asymmetric
tracer distribution and the linear variation of distribution volume with the infusion
volume. Sensitivity of the model to changes in hydraulic conductivity and catheter
placement were investigated. The tracer distribution was found to be sensitive to
both the parameters under study. Increasing the values of the hydraulic conductivity
map lowered the tumor IFP and raised the distribution volume within the whole leg.
However within the tumor, the distribution volume decreased with increasing value of
the empirical parameter (m) used to increase the hydraulic conductivity, at later time
points. The infusion at the tumor-host tissue interface resulted in larger distribution
volume compared to that at the center and anterior end of the tumor, under baseline
conditions. Within the tumor, the distribution volume was almost identical for infusions at
the interface and center of the tumor.
The accuracy of the model’s CED predictions could be improved with the following
modifications which would be the subject of future work. Firstly, transvascular solute
exchange can be accounted in the transport equation for smaller molecular weight
compounds. It can be expected to affect the tracer distribution because of the leakiness
of the tumors. Secondly, binding, metabolism and uptake of the macromolecular drug
can be accounted for using reaction and binding kinetics. Finally, the model can be
validated with experiments across a wider range of tumors. In addition to the above
extensions, the voxelized model solution also needs to be tested for grid and time step
independency, to ensure that discretization and truncation errors are small.
74
REFERENCES
[1] Ahlstrom, H., Christofferson, R., Lorelius, L. E., 1988. Vascularization of thecontinuous human colonic cancer cell line LS 174 T deposited subcutaneously innude rats. APMIS 96 (7-12), 701–710.
[2] Allard, E., Passirani, C., Benoit, J. P., 2009. Convection-enhanced delivery ofnanocarriers for the treatment of brain tumors. Biomaterials 30 (12), 2302–2318.
[3] Anderson, D. A., Tannehill, J. C., Pletcher, R. H., 1984. Computational fluidmechanics and heat transfer. Hemisphere Publishing, New York, NY, pp. 671–674.
[4] Aref, M., Brechbiel, M., Wiener, E. C., 2002. Identifying tumor vascular permeabilityheterogeneity with magnetic resonance imaging contrast agents. InvestigativeRadiology 37 (4), 178–192.
[5] Arunyawongsakorn, U., Johnson, C. S., Gabriel, D. A., 1985. Tracer diffusioncoefficients of proteins by means of holographic relaxation spectroscopy:Application to bovine serum albumin. Analytical Biochemistry 146 (1), 265 –270.
[6] Baish, J. W., Gazit, Y., Berk, D. A., Nozue, M., Baxter, L. T., Jain, R. K., 1996.Role of tumor vascular architecture in nutrient and drug delivery: an invasionpercolation-based network model. Microvascular research 51 (3), 327–346.
[7] Baxter, L. T., Jain, R. K., 1989. Transport of fluid and macromolecules in tumors. I.role of interstitial pressure and convection. Microvascular research 37 (1), 77–104.
[8] Baxter, L. T., Jain, R. K., 1990. Transport of fluid and macromolecules in tumors.II. role of heterogeneous perfusion and lymphatics. Microvascular research 40 (2),246–263.
[9] Bjørnaes, I., Rofstad, E. K., 2001. Microvascular permeability to macromolecules inhuman melanoma xenografts assessed by contrast-enhanced MRI–intertumor andintratumor heterogeneity. Magnetic resonance imaging 19 (5), 723–730.
[10] Bobo, R. H., Laske, D. W., Akbasak, A., Morrison, P. F., Dedrick, R. L., Oldfield,E. H., 1994. Convection-enhanced delivery of macromolecules in the brain.Proceedings of the National Academy of Sciences 91 (6), 2076–2080.
[11] Boucher, Y., Baxter, L., Jain, R., 1990. Interstitial pressure gradients intissue-isolated and subcutaneous tumors: implications for therapy. Cancer research50 (15), 4478–4484.
[12] Boucher, Y., Brekken, C., Netti, P. A., Baxter, L. T., Jain, R. K., 1998. Intratumoralinfusion of fluid: estimation of hydraulic conductivity and implications for the deliveryof therapeutic agents. British Journal of Cancer 78 (11), 1442–1448.
75
[13] Boucher, Y., Kirkwood, J. M., Opacic, D., Desantis, M., Jain, R. K., 1991. Interstitialhypertension in superficial metastatic melanomas in humans. Cancer research51 (24), 6691–6694.
[14] Butler, T. P., Grantham, F. H., Gullino, P. M., 1975. Bulk transfer of fluid in theinterstitial compartment of mammary tumors. Cancer research 35 (11), 3084–3088.
[15] Chen, Z. J., Broaddus, W. C., Viswanathan, R. R., Raghavan, R., Gillies, G. T.,2002. Intraparenchymal drug delivery via positive-pressure infusion: experimentaland modeling studies of poroelasticity in brain phantom gels. IEEE Trans BiomedEng 49 (2), 85–96.
[16] DiResta, G. R., Lee, J., Larson, S. M., Arbit, E., 1993. Characterization ofneuroblastoma xenograft in rat flank. I. Growth, interstitial fluid pressure, andinterstitial fluid velocity distribution profiles. Microvascular research 46 (2), 158–177.
[17] El-Kareh, A. W., Secomb, T. W., 1995. Effect of increasing vascular hydraulicconductivity on delivery of macromolecular drugs to tumor cells. Internationaljournal of radiation oncology, biology, physics 32 (5), 1419–1423.
[18] Engebraaten, O., Hjortland, G. O., Juell, S., Hirschberg, H., Fodstad, O.,2002. Intratumoral immunotoxin treatment of human malignant brain tumors inimmunodeficient animals. International Journal of Cancer 97 (6), 846–852.
[19] Fu, Y., Nagy, J. A., Dvorak, A. M., Dvorak, H. F., 2008. Tumor Blood Vessels.Antiangiogenic Agents in Cancer Therapy, 205–224.
[20] Fukumura, D., Duda, D. G., Munn, L. L., Jain, R. K., 2010. Tumor microvasculatureand microenvironment: novel insights through intravital imaging in pre-clinicalmodels. Microcirculation 17 (3), 206–25.
[21] Grathwohl, P., 1998. Diffusion in natural porous media: Contaminant transport,sorption/desorption and dissolution kinetics. Kluwer Academic Publishers, Norwell,MA, p. 31.
[22] Guerin, C., Olivi, A., Weingart, J. D., Lawson, H. C., Brem, H., 2004. Recentadvances in brain tumor therapy: Local intracerebral drug delivery by polymers.Investigational New Drugs 22 (1), 27–37.
[23] Gutmann, R., Leunig, M., Feyh, J., Goetz, A. E., Messmer, K., Kastenbauer, E.,Jain, R. K., 1992. Interstitial hypertension in head and neck tumors in patients:correlation with tumor size. Cancer research 52 (7), 1993–1995.
[24] Hadaczek, P., Yamashita, Y., Mirek, H., Tamas, L., Bohn, M. C., Noble, C., Park,J. W., Bankiewicz, K., 2006. The ”perivascular pump” driven by arterial pulsation isa powerful mechanism for the distribution of therapeutic molecules within the brain.Molecular Therapy 14 (1), 69–78.
76
[25] Hamberg, L. M., Kristjansen, P. E. G., Hunter, G. J., Wolf, G. L., Jain, R. K., 1994.Spatial heterogeneity in tumor perfusion measured with functional computedtomography at 0.05 µl resolution. Cancer research 54 (23), 6032–6036.
[26] Hassid, Y., Furman-Haran, E., Margalit, R., Eilam, R., Degani, H., 2006.Noninvasive magnetic resonance imaging of transport and interstitial fluid pressurein ectopic human lung tumors. Cancer research 66 (8), 4159–4166.
[27] Heilmann, M., Walczak, C., Vautier, J., Dimicoli, J. L., Thomas, C. D., Lupu, M.,Mispelter, J., Volk, A., 2007. Simultaneous dynamic T1 and T2* measurementfor AIF assessment combined with DCE-MRI in a mouse tumor model. MagneticResonance Materials in Physics, Biology and Medicine 20 (4), 193–203.
[28] Heisen, M., Fan, X., Buurman, J., van Riel, N. A. W., Karczmar, G. S., terHaar Romeny, B. M., 2010. The use of a reference tissue arterial input functionwith low-temporal-resolution DCE-MRI data. Physics in Medicine and Biology 55,4871–4883.
[29] Heldin, C.-H., Rubin, K., Pietras, K., Ostman, A., 2004. High interstitial fluidpressure - an obstacle in cancer therapy. Nature Reviews Cancer 4 (10), 806–813.
[30] Holmes, M. H., Mow, V. C., 1990. The nonlinear characteristics of soft gels andhydrated connective tissues in ultrafiltration. Journal of biomechanics 23 (11),1145–1156.
[31] Huang, G., Chen, L., 2008. Review: Tumor Vasculature and MicroenvironmentNormalization: A Possible Mechanism of Antiangiogenesis Therapy. Cancerbiotherapy & radiopharmaceuticals 23 (5), 661–668.
[32] Jain, R. K., 1987. Transport of molecules across tumor vasculature. Cancer andMetastasis Reviews 6 (4), 559–593.
[33] Jain, R. K., 1988. Determinants of tumor blood flow: a review. Cancer research48 (10), 2641–2658.
[34] Jain, R. K., 1994. Barriers to drug delivery in solid tumors. Scientific American 271,58–65.
[35] Jain, R. K., 1994. Transport phenomena in tumors. Advances in chemicalengineering 19, 130–200.
[36] Jain, R. K., Baxter, L. T., 1988. Mechanisms of heterogeneous distribution ofmonoclonal antibodies and other macromolecules in tumors: significance ofelevated interstitial pressure. Cancer research 48 (24 Part 1), 7022–7032.
[37] Jain, R. K., di Tomaso, E., Duda, D. G., Loeffler, J. S., Sorensen, A. G., Batchelor,T. T., 2007. Angiogenesis in brain tumours. Nature Reviews Neuroscience 8 (8),610–622.
77
[38] Kim, J. H., Astary, G. W., Chen, X., Mareci, T. H., Sarntinoranont, M., 2009.Voxelized model of interstitial transport in the rat spinal cord following direct infusioninto white matter. Journal of biomechanical engineering 131, 071007.
[39] Kim, J. H., Mareci, T. H., Sarntinoranont, M., 2010. A voxelized model of directinfusion into the corpus callosum and hippocampus of the rat brain: modeldevelopment and parameter analysis. Medical and Biological Engineering andComputing 48 (3), 203–214.
[40] Konerding, M. A., Steinberg, F., Budach, V., 1989. The vascular system ofxenotransplanted tumors–scanning electron and light microscopic studies.Scanning microscopy 3 (1), 327–335.
[41] Lai, W. M., Mow, V. C., 1980. Drag-induced compression of articular cartilageduring a permeation experiment. Biorheology 17 (1-2), 111–123.
[42] Less, J. R., Skalak, T. C., Sevick, E. M., Jain, R. K., 1991. Microvasculararchitecture in a mammary carcinoma: branching patterns and vessel dimensions.Cancer research 51 (1), 265–273.
[43] Linninger, A. A., Somayaji, M. R., Mekarski, M., Zhang, L., 2008. Prediction ofconvection-enhanced drug delivery to the human brain. Journal of theoreticalbiology 250 (1), 125–138.
[44] MacKay, J. A., Deen, D. F., Jr, F. C. S., 2005. Distribution in brain of liposomes afterconvection enhanced delivery; modulation by particle charge, particle diameter, andpresence of steric coating. Brain research 1035 (2), 139–153.
[45] Magdoom-Mohamed, K. N., Pishko, G. L., Kim, J. H., Sarntinoranont, M., 2011.Evaluation of a voxelized model for tumor transport based on DCE-MRI, underpreparation.
[46] McGuire, S., Yuan, F., 2001. Quantitative analysis of intratumoral infusion of colormolecules. American Journal of Physiology-Heart and Circulatory Physiology281 (2), H715–H721.
[47] Milosevic, M. F., Fyles, A. W., Hill, R. P., 1999. The relationship between elevatedinterstitial fluid pressure and blood flow in tumors: a bioengineering analysis.International Journal of Radiation Oncology Biology Physics 43 (5), 1111–1123.
[48] Nathanson, S. D., Nelson, L., 1994. Interstitial fluid pressure in breast cancer,benign breast conditions, and breast parenchyma. Annals of Surgical Oncology1 (4), 333–338.
[49] Okuhata, Y., 1999. Delivery of diagnostic agents for magnetic resonance imaging.Advanced drug delivery reviews 37 (1-3), 121–137.
78
[50] Perlstein, B., Ram, Z., Daniels, D., Ocherashvilli, A., Roth, Y., Margel, S., Mardor,Y., 2008. Convection-enhanced delivery of maghemite nanoparticles: increasedefficacy and MRI monitoring. Neuro-oncology 10 (2), 153–161.
[51] Pishko, G., 2011. Magnetic resonance imaging-based computational models ofsolid tumors. PhD Dissertation, University of Florida.
[52] Pishko, G. L., Astary, G. W., Mareci, T. H., Sarntinoranont, M., June 2008. Highresolution DCE-MRI vascular characterization of murine sarcoma and human renalcell carcinoma for computational modeling. In: ASME Summer BioengineeringConference. Marco Island, Florida.
[53] Pishko, G. L., Astary, G. W., Mareci, T. H., Sarntinoranont, M., 2011. Sensitivityanalysis of an image-based solid tumor computational model with heterogeneousvasculature and porosity. Annals of Biomedical Engineering, 1–14.
[54] Pozrikidis, C., 2010. Numerical simulation of blood and interstitial flow through asolid tumor. Journal of mathematical biology 60 (1), 75–94.
[55] Roberts, T. P. L., 1997. Physiologic measurements by contrast-enhanced MRimaging: Expectations and limitations. Journal of Magnetic Resonance Imaging7 (1), 82–90.
[56] Saito, R., Bringas, J. R., McKnight, T. R., Wendland, M. F., Mamot, C.,Drummond, D. C., Kirpotin, D. B., Park, J. W., Berger, M. S., Bankiewicz,K. S., 2004. Distribution of liposomes into brain and rat brain tumor models byConvection-Enhanced delivery monitored with magnetic resonance imaging.Cancer Research 64 (7), 2572–2579.
[57] Sampson, J. H., Brady, M. L., Petry, N. A., Croteau, D., Friedman, A. H.,Friedman, H. S., Wong, T., Bigner, D. D., Pastan, I., Puri, R. K., Pedain, C., 2007.Intracerebral infusate distribution by convection-enhanced delivery in humans withmalignant gliomas: descriptive effects of target anatomy and catheter positioning.Neurosurgery 60 (2), 89–99.
[58] Sevick, E. M., Jain, R. K., 1989. Geometric resistance to blood flow in solid tumorsperfused ex vivo: effects of tumor size and perfusion pressure. Cancer research49 (13), 3506–3512.
[59] Smith, J. H., Humphrey, J. A. C., 2007. Interstitial transport and transvascularfluid exchange during infusion into brain and tumor tissue. Microvasc. Res. 73 (1),58–73.
[60] Swartz, M. A., 2001. The physiology of the lymphatic system. Advanced drugdelivery reviews 50 (1-2), 3–20.
[61] Swartz, M. A., Fleury, M. E., 2007. Interstitial flow and its effects in soft tissues.Annu. Rev. Biomed. Eng. 9, 229–256.
79
[62] Szerlip, N. J., Walbridge, S., Yang, L., Morrison, P. F., Degen, J. W., Jarrell, S. T.,Kouri, J., Kerr, P. B., Kotin, R., Oldfield, E. H., Lonser, R. R., 2007. Real-timeimaging of convection-enhanced delivery of viruses and virus-sized particles.Journal of Neurosurgery: Pediatrics 107 (3), 560–567.
[63] Tan, W. H. K., Wang, F., Lee, T., Wang, C. H., 2003. Computer simulation of thedelivery of etanidazole to brain tumor from PLGA wafers: comparison betweenlinear and double burst release systems. Biotechnology and bioengineering 82 (3),278–288.
[64] Tannock, I. F., 1968. The relation between cell proliferation and the vascular systemin a transplanted mouse mammary tumour. British Journal of Cancer 22 (2),258–273.
[65] Tannock, I. F., 1972. Oxygen diffusion and the distribution of cellular radiosensitivityin tumours. British Journal of Radiology 45 (535), 515–524.
[66] Thomlinson, R. H., Gray, L. H., 1955. The histological structure of some humanlung cancers and the possible implications for radiotherapy. British Journal ofCancer 9 (4), 539–549.
[67] Tofts, P. S., Kermode, A. G., 1991. Measurement of the blood-brain barrierpermeability and leakage space using dynamic MR imaging. 1. Fundamentalconcepts. Magnetic Resonance in Medicine 17 (2), 357–367.
[68] Truskey, G. A., Yuan, F., Katz, D. F., 2004. Transport phenomena in biologicalsystems. Pearson Prentice Hall, Upper Saddle River, NJ, p. 473.
[69] Tweedle, M. F., Wedeking, P., Telser, J., Sotak, C. H., Chang, C. A., Kumar, K.,Wan, X., Eaton, S. M., 1991. Dependence of MR signal intensity on Gd tissueconcentration over a broad dose range. Magnetic resonance in medicine 22 (2),191–194.
[70] Urciuolo, F., Imparato, G., Netti, P. A., 2008. Effect of dynamic loading on solutetransport in soft gels implication for drug delivery. AIChE Journal 54 (3), 824–834.
[71] van Doormaal, J. P., Raithby, G. D., 1984. Enhancement of the simple method forpredicting incompressible fluid flows. Num. Heat Transfer 7, 147–163.
[72] Walter, K. A., Tamargo, R. J., Olivi, A., Burger, P. C., Brem, H., 1995. Intratumoralchemotherapy. Neurosurgery 37 (6), 1129–1145.
[73] Wang, Y., Hu, J. K., Krol, A., Li, Y., Li, C., Yuan, F., 2003. Systemic disseminationof viral vectors during intratumoral injection. Molecular Cancer Therapeutics 2 (11),1233–1242.
[74] Wang, Y., Wang, H., Li, C., Yuan, F., Feb. 2006. Effects of rate, volume, and doseof intratumoral infusion on virus dissemination in local gene delivery. MolecularCancer Therapeutics 5 (2), 362–366.
80
[75] Weinberg, B. D., Patel, R. B., Exner, A. A., Saidel, G. M., Gao, J., 2007. Modelingdoxorubicin transport to improve intratumoral drug delivery to RF ablated tumors. J.Control. Release 124 (1–2), 11–19.
[76] Weinmann, H. J., Laniado, M., M utzel, W., 1984. Pharmacokinetics ofGd-DTPA/dimeglumine after intravenous injection into healthy volunteers.Physiological chemistry and physics and medical NMR 16 (2), 167–172.
[77] Young, J. S., Llumsden, C. E., Stalker, A. L., 1950. The significance of the tissuepressure of normal testicular and of neoplastic (brown-pearce carcinoma) tissue inthe rabbit. The Journal of Pathology and Bacteriology 62 (3), 313–333.
[78] Zhao, J., Salmon, H., Sarntinoranont, M., 2007. Effect of heterogeneousvasculature on interstitial transport within a solid tumor. Microvascular research73 (3), 224–236.
81
BIOGRAPHICAL SKETCH
Magdoom Mohamed received his Bachelor of Technology degree in Mechanical
Engineering from National Institute of Technology (NIT), Tiruchirappalli, India in 2008.
Between 2008-09, he worked at Indian Institute of Technology (IIT), Madras as project
assistant in the Department of Bio-technology. In Spring 2010, he was admitted to the
graduate program in the Department of Mechanical and Aerospace Engineering at the
University of Florida. In Summer 2011, he received his MS in Mechanical Engineering
from the University of Florida.
82