Sei sulla pagina 1di 4

Computational Modeling in Quantitative Cancer Imaging

Thomas E. Yankeelov, Nkiruka C. Atuegwu, and John C. Gore


!

AbstractIn recent years there have been dramatic


increases in the range and quality of information available
from non-invasive imaging methods so that a number of
potentially valuable metrics are now available to quantitatively
assess tumor status. Several of these have been used in both
pre-clinical studies of animal models and clinical trials
involving patients. However, the optimal methods by which
these emerging imaging metrics are integrated and applied
have yet to be developed. Here we provide an example of the
kind of data available from quantitative imaging of cancer, and
then propose an approach for how these data can be combined
in order to offer a more comprehensive description of tumor
growth and treatment response.

I. INTRODUCTION

connotes the combination of a contrast


mechanism (either endogenous or exogenous) and a
device to map that contrast spatially. Cancer imaging
exploits contrast mechanisms that allow for tumors to be
identified and separated from surrounding healthy tissues.
Two phenomena associated with cancer that are attractive
targets for imaging are the increase in cellular proliferation
(and subsequent increase in cell density), and the increase in
vascularity when compared to healthy tissue.
A basic feature of cancer is that it is a set of diseases
characterized by unregulated cell growth and proliferation.
Thus, one of the principal goals for cancer imaging is being
able to measure the effects of increased cellularity.
Furthermore, since many anti-cancer drugs have as their
ultimate goal destruction of tumor cells, imaging methods
sensitive to changes to tissue cellularity are of great import.
As tumor cells proliferate, the mass will eventually reach 12 mm3 in volume, at which point it can no longer rely on the
passive diffusion of metabolites from host tissue blood
vessels in order to continue to grow, so new vasculature
must develop. This process of neovascularization, or
angiogenesis [1], is a signature of neoplasms and a second
important target for cancer imaging.
In this short contribution, we will focus on tissue
cellularity. Our goal is to present a new approach for how
MAGING

Manuscript received March 13, 2009. This work was supported in part
by the U.S. National Institutes of Health, Grant 1K25EB005936 and 5 R01
EB000214.
T. E. Yankeelov is with the Vanderbilt University Institute of Imaging
Science, Nashville, TN 37232 USA (615-322-8359; fax: 615-322-0734; email: thomas.yankeelov@vanderbilt.edu).
N. C. Atuegwu is with the Vanderbilt University Institute of Imaging
Science, Nashville, TN 37232 USA (615-322-8359; fax: 615-322-0734; email: nkiruka.atuegwu@vanderbilt.Edu).
J. C. Gore is with the Vanderbilt University Institute of Imaging Science,
Nashville, TN 37232 USA (615-322-8359; fax: 615-322-0734; e-mail:
john.gore@vanderbilt.Edu).

imaging data can be used to predict tumor growth and/or


response to treatment. We consider here only the simplest
model of tumor growth, namely avascular tumors, to
illustrate the approach. The incorporation of vascularity,
and other features, into more complex models of tumor
growth using imaging data is an exciting prospect and there
are several promising techniques that afford such
measurements such as dynamic contrast enhanced MRI [2].
Below we first review the state of the art in assessing
tumor growth, before providing a short review of imaging
data that are relevant to modeling tumor growth and
treatment response through changes in cellularity. We then
proceed to a brief description of how tumors can be modeled
and how imaging data can advance such methods. We
believe our proposal represents a new paradigm for
mathematical modeling of tumor growth.
II. ASSESSING THE RESPONSE OF TUMORS TO TREATMENT
The current standard-of-care radiological assessment of
tumor response to treatment is based on the Response
Evaluation Criteria in Solid Tumors (RECIST). RECIST
offers a simplified, practical method for extracting the
salient features of anatomical imaging data. Briefly,
RECIST divides treatment response into four categories [3]:
complete response (disappearance of lesions), partial
response (30% decrease in the sum of the longest diameter
of the lesions), progressive disease (20% increase in the sum
of the longest diameter of the target lesions), and stable
disease (changes that do not meet the above criteria). It is
well recognized that this approach needs to be significantly
improved for a number of reasons. For example, the metric
for positive response is based on one dimensional changes
(sum of longest diameters) which can be grossly misleading
in trying to describe a complex object changing in all three
spatial dimensions. Furthermore, this metric is based on
anatomical and morphological changes which are only
(temporally) downstream manifestations of underlying
pathophysiological changes.
Changes in cell proliferation and cellularity occur, by
definition, before changes in tumor size, so it is a reasonable
hypothesis that techniques designed to report on cellular
proliferation could offer an earlier indication of positive
response to therapy than what is currently offered by the
RECIST criteria. As discussed below, imaging methods
have emerged that can report on tumor cellularity, but the
most effective way to use this information has not yet been
developed.

III. IMAGING TECHNIQUES THAT REPORT ON CELLULARITY


Here we focus on approaches that are capable of
providing three dimensional imaging data on tissue
cellularity, have the ability to be quantitative, and can be
readily acquired in the clinical setting. To this end, we
consider magnetic resonance imaging (MRI) and positron
emission tomography (PET). We note that single photon
emission computed tomography (SPECT) can also provide a
very important metric of one aspect of cellularity (apoptosis,
or programmed cell death [4]), but we will not make use of
that data in this document. Instead we focus on diffusion
weighted MRI and 18F-fluorodeoxythymidine PET.
A. Diffusion weighted magnetic resonance imaging
The microscopic thermally-induced behavior of molecules
moving in a random pattern is referred to as self-diffusion or
Brownian motion. The rate of diffusion in cellular tissues is
lower in free solution and restricted by various barriers such
as semi-permeable membranes. It is then described by
means of an apparent diffusion coefficient (ADC), which
largely depends on the number and separation of barriers
that a diffusing water molecule encounters. Diffusion
weighted MRI (DW-MRI) methods have been developed to
map the ADC, and in well-controlled situations the
variations in ADC have been shown to correlate negatively
with tissue cellularity [5]. As applied to cancer, the common
interpretation of the ADC is that the more cells present
within a section of tissue, the greater the number of barriers
hindering free diffusion and thus the smaller the measured
ADC. After treating a tumor with (for example) a cytoxic
drug, there may be fewer cells and therefore fewer barriers
which would result in an increase in the measured ADC. It
has been shown that exposure of tumors to both
chemotherapy and radiotherapy leads to measurable
increases in water diffusion in cases of favorable treatment
response [6].
B. Fluorodeoxythymidine PET
The increased proliferative activity of tumor cells is a well
known identifying characteristic of cancer. A PET tracer
that reports on cellular proliferation is the thymidine
analogue fluorodeoxythymidine (FLT). Thymidine is a
native nucleoside which is taken up by cells via their surface
nucleoside transporters. Once inside the cell, thymidine is
phosphorylated by the enzyme thymidine kinase 1 (TK1)
into thymidine monophosphate. As both the quantity and
activity of TK1 is known to be upregulated during DNA
synthase phase, there is preferential uptake of thymidine in
tumors where cells are rapidly proliferating. FLT works in
an analogous fashion; FLT is taken into cells and once
phosphorylated is trapped. Thus, actively proliferating cells
will accumulate FLT and this forms the basis of FLT-PET
imaging of cell proliferation [7]. The common interpretation
is that tumor cells are rapidly proliferating and thus there is
an upregulation of TK1 resulting in greater retention of FLT
as compared to healthy tissue. After treating a tumor with
(for example) a drug designed to inhibit proliferation, the
amount of FLT retained within the tumor may decrease. It

has been shown that changes in FLT retention can be


predictive of treatment response [7].
C. Multi-modality Imaging
There has been much recent effort in developing methods
of co-aligning data taken by multiple imaging modalities.
For example, the Memorial Sloan-Kettering group
developed and assessed a method whereby small animals
could be reproducibly and serially imaged by multiple
modalities allowing for registration of the resulting imaging
data sets [8, 9]. Similar methods have been proposed by
others [10, 11]. By co-aligning data obtained from multiple
scanners (say, ADC by MRI and FLT retention by PET),
each voxel can be characterized by multiple measures. As
discussed below in section V, this is precisely the type of
data that we propose can be used to model tumor growth. It
is important to note that these image registration techniques
are not limited to the pre-clinical setting as recent efforts
have fused MRI and PET images in the case of human breast
cancer [12, 13] and hybrid MRI-PET scanners are under
development.
IV. PREVIOUS EFFORTS AT MODELING CANCER GROWTH
For decades biomathematicians have developed elegant
models of tumor growth to simulate and predict how tumors
can grow in particular environments [14, 15]. While these
models have taught us much about tumor growth patterns
and have generated hypotheses that could be tested in silico
and in vitro, their direct application to in vivo physiological
events has been extremely limited. Current mathematical
models of tumor growth often rely on knowledge of
molecular level data that are quite difficult or impossible to
measure in an intact living system (e.g., metrics of
chemotaxis, haptotaxis, or growth factor gradients). Thus,
the general limitation of these models is that they are almost
uniformly driven by parameters that can be measured only
by highly invasive methods (surgery and/or animal sacrifice)
or idealized (in vitro) systems. In addition to the obvious
difficulty of measuring these parameters in vivo and in three
dimensions throughout a tumor at even a single time point, it
is extremely difficult to measure such quantities with any
reasonable spatial or temporal resolution in the clinical
setting. In contrast, we propose to construct (or recast
existing) mathematical models of tumor growth that can be
parameterized and driven by data obtained from noninvasive
MRI, SPECT, and PET data. In this way, the proposed
approach is fundamentally different from existing
mathematical models of tumor growth as the models are
driven by parameters that can be obtained noninvasively in
3D and therefore can be measured repeatedly to update and
refine predictions. We now present a very simple model to
illustrate this approach.
V.

OPPORTUNITIES FOR MODELING IMAGING DATA

The logistic growth model (see, e.g., [16]) incorporates


exponential growth of tumor cells early in time, and then

asymptotically approaches the cellular carrying capacity.


The associated differential equation is given as Eq. (1):
dN
" N#
$ kN &1 % ' , with N (t $ 0) $ N 0 ,
(1)
dt
( * )
where N0 is the number of cells initially present, k is the
cells proliferative rate, and * is the carrying capacity of the
population. The solution is given as Eq. (2):
* N0
N (t ) $
.
(2)
N 0 + (* % N 0 ) e % kt
There are several ways in which Eq. (2) can be applied from
the imaging perspective and we explore two possible
approaches here.
The first method we consider is how to obtain maps of the
proliferation rate, k. We must first consider how the
parameters can be assigned from available imaging data.
First, the carrying capacity can be interpreted as the
maximum number of cells that can be contained within a
given volume. For imaging, the relevant volume scale is
that of a voxel and therefore the carrying capacity is merely
given by * = (voxel volume)/(cell volume). Since the voxel
volume and the individual cell volume (for a given cell
species) are relatively well-defined quantities, the carrying
capacity can be assigned. The next issue is to determine
how to measure cell number, N(t), or at least relative cell
number, Nrel(t). To do this, assume that
ADC(t) = ADCw ,N(t),
(3)
where ADCw is the ADC of free water, N(t) is the cell
number, and , is a proportionality constant.
This
relationship has been shown to hold in well-controlled
settings [5]. Then we can write the relative number of cells
at time t as Eq. (4):
Nrel(t) = N(t)/N0
= [ADC(t) ADCw]/[ADC(0) ADCw]. (4)
Thus, we have converted from measured ADC at time t to a
relative cell number at time t given by Nrel(t). (Note that if
Nrel(t) > 1 for a particular voxel, then that voxel showed
increase in cell number between the two time imaging
points. Note also that Nrel(t) can be obtained in 3D and at
multiple time points.) We can then recast Eq. (2) as:
N rel (t ) $

,
(5)
1 + (* % 1) e % kt
and every term in Eq. (5) is known except k.
Consider, for example, a rat brain tumor model where
multiple imaging sessions are planned in both treated and
control animals. The tumors are allowed to grow to (say)
250 mm3 and then ADC(t) is measured at t = 0 and at
subsequent times t > 0 to yield estimates of Nrel(t). Since *
is fixed, and Nrel(t) and Nrel(0) are measured, it is possible to
fit Eq. (5) to such data to extract k for each voxel thereby
yielding a proliferation rate map. One would hypothesize
(and could test) that rats from treated and untreated groups
would display different k distributions which could be used
to separate responders from non-responders.
A second possible method of applying Eq. (2) to imaging
data is to obtain estimates of k from FLT-PET data and use
that information to predict tumor growth and treatment

response. For example, Eq. (2) directly predicts the number


of cells provided the carrying capacity and proliferation rate
are known. It would be a straightforward experiment to
measure the proliferation rate as estimated from FLT-PET
data to estimate the number of cells available at a later time
point. The number of cells at that later time point could be
estimated from a number of ways, including standard
histological methods. In this way, the modeling can provide
hypotheses (i.e., cell number at a later time point given a
proliferation rate) that are testable.
VI. CONCLUSION
It is important to note that this simple model has been
parameterized by existing imaging techniques and therefore
can immediately be tested. Furthermore, it underscores the
point of having a model that makes predictions on tumor
growth given inputs obtained from imaging parameters
obtained in 3D and that these predictions can be directly
tested for individual animals and patients. Our longer term
goal is not only to synthesize existing methods, but also to
incorporate the information from new, more advanced
imaging techniques. In this way we will be constructing
models of tumor response to treatment that have built into
them parameters that are known to correlate with cancer
response and which can be measured by existing clinical
imaging scanners.
We have presented an approach whereby multi-modality
imaging data can be synthesized by driving a (simple)
mathematical model of tumor growth.
The example
provided here is quite simple and represents only an initial
step. Additionally, there are other ways by which large
imaging data sets can be assessed by computation methods.
These include bioinformatics [17, 18] and statistical [19]
based methods. We believe the concept of using imaging
data to drive mathematical models is a powerful and general
method in which to increase our understanding of the
response of tumors and to make useful predictions.
REFERENCES
[1]
[2]

[3]

[4]

[5]

J. Folkman, "Role of angiogenesis in tumor growth and metastasis,"


Semin Oncol, vol. 29, pp. 15-8, 2002.
T. E. Yankeelov and J. C. Gore, "Dynamic contrast enhanced
magnetic resonance imaging in oncology: theory, data acquisition,
analysis, and examples," Current Medical Imaging Reviews, vol. 3,
pp. 91-107, 2007.
P. Therasse, S. G. Arbuck, E. A. Eisenhauer, J. Wanders, R. S.
Kaplan, L. Rubinstein, J. Verweij, M. Van glabbeke, A.T. Van
Oosterom, M. C. Christian, and S. G. Gwyther, "New guidelines to
evaluate the response to treatment in solid tumors. European
organization for research and treatment of cancer, national cancer
institute of the United States, National Cancer Institute of Canada," J
Natl Cancer Inst, vol. 92, pp. 205-16, 2000.
H. H. Boersma, B. L. Kietselaer, L. M. Stolk, A. Bennaghmouch, L.
Hofstra, J. Narula, G. A. Heidendal, and C. P. Reutelingsperger,
"past, present, and future of annexin A5: from protein discovery to
clinical applications," J Nucl Med, vol. 46, pp. 2035-50, 2005.
A. W. Anderson, J. Xie, J. Pizzonia, R. A. Bronen, D. D. Spencer,
and J. C. Gore, "Effects of cell volume fraction changes on apparent
diffusion in human cells," Magn Reson Imaging, vol. 18, pp. 689-95,
2000.

[6]

[7]

[8]

[9]

[10]

[11]

[12]

[13]

[14]

[15]
[16]
[17]
[18]
[19]

K. C. Lee, B. A. Moffat, A. F. Schott, R. Layman, S. Ellingworth, R.


Juliar, A. P. Khan, M. Helvie, C. R. Meyer, T. L. Chenevert, A.
Rehemtulla, and B. D. Ross, "Prospective early response imaging
biomarker for neoadjuvant breast cancer chemotherapy," Clin Cancer
Res, vol. 13, pp. 443-50, 2007.
A. Salskov, V. S. Tammisetti, J. Grierson, and H. Vesselle, "FLT:
measuring tumor cell proliferation in vivo with positron emission
tomography and 3'-deoxy-3'-[18f]fluorothymidine," Semin Nucl
Med, vol. 37, pp. 429-39, 2007.
M. Zhang, M. Huang, C. Le, P. B. Zanzonico, F. Claus, K. S.
Kolbert, K. Martin, C. C. Ling, J. A. Koutcher, and J. L. Humm,
"Accuracy and reproducibility of tumor positioning during prolonged
and multi-modality animal imaging studies," Phys Med Biol, vol. 53,
pp. 5867-82, 2008.
J. L. Humm, D. Ballon, Y. C. Hu, S. Ruan, C. Chui, P. K. Tulipano,
A. Erdi, J. Koutcher, K. Zakian, M. Urano, P. Zanzonico, C. Mattis,
J. Dyke, Y. Chen, P. Harrington, J. A. O'donoghue, and C. C. Ling,
"A stereotactic method for the three-dimensional registration of
multi-modality biologic images in animals: NMR, PET, histology,
and autoradiography," Med Phys, vol. 30, pp. 2303-14, 2003.
M. L. Jan, K. S. Chuang, G. W. Chen, Y. C. Ni, S. Chen, C. H.
Chang, J. Wu, T. W. Lee, and Y. K. Fu, "A three-dimensional
registration method for automated fusion of micro pet-ct-spect
whole-body images," IEEE Trans Med Imaging, vol. 24, pp. 886-93,
2005.
W. Y. Guo, J. J. Lee, M. H. Lin, C. C. Yang, C. L. Chen, Y. H.
Huang, Y. S. Tyan, and T. H. Wu, "Merging molecular and
anatomical information: a feasibility study on rodents using
microPET and MRI," Nucl Med Commun, vol. 28, pp. 804-12, 2007.
L. Moy, F. Ponzo, M. E. Noz, G. Q. Maguire, Jr., A. D. Murphywalcott, A. E. Deans, M. T. Kitazono, L. Travascio, and E. L.
Kramer, "Improving specificity of breast MRI using prone PET and
fused mri and pet 3d volume datasets," J Nucl Med, vol. 48, pp. 52837, 2007.
L. Moy, M. E. Noz, G. Q. Maguire, Jr., F. Ponzo, A. E. Deans, A. D.
Murphy-Walcott, and E. L. Kramer, "Prone mammopet acquisition
improves the ability to fuse MRI and PET breast scans," Clin Nucl
Med, vol. 32, pp. 194-8, 2007.
S. Sanga, J. P. Sinek, H. B. Frieboes, M. Ferrari, J. P. Fruehauf, and
V. Cristini, "Mathematical modeling of cancer progression and
response to chemotherapy," Expert Rev Anticancer Ther, vol. 6, pp.
1361-76, 2006.
V. Quaranta, A. M. Weaver, P. T. Cummings, and A. R. Anderson,
"Mathematical modeling of cancer: the future of prognosis and
treatment," Clin Chim Acta, vol. 357, pp. 173-9, 2005.
H. M. Byrne, "Modelling avascular tumor growth," in Cancer
Modelling and Simulation, L. Preziosi, ed. London: Chapman and
Hall/CRC, 2003, pp. 45.
J. A. Cruz and D. S. Wishart, "Applications of machine learning in
cancer prediction and prognosis," Cancer Informatics, vol. 2, pp. 5978, 2006.
A. Vellido and P. J. G. Lisboa, "neural networks and other machine
learning methods in cancer research," Lecture Notes in Computer
Science, vol. 4507, pp. 964, 2007.
K. M. Mcmillan, B. P. Rogers, C. G. Koay, A. R. Laird, R. R. Price,
and M. E. Meyerand, "An objective method for combining multiparametric mri datasets to characterize malignant tumors," Med Phys,
vol. 34, pp. 1053-61, 2007.

Potrebbero piacerti anche