Dependence of DCE-MRI biomarker values on analysis algorithm
Dependence of DCE-MRI biomarker values on analysis algorithm
Chaan S. Ng 0 1 2 3
Wei Wei 0 1 2 3
James A. Bankson 0 1 2 3
Murali K. Ravoori 0 1 2 3
Lin Han 0 1 2 3
David W. Brammer 0 1 2 3
Sherry Klumpp 0 1 2 3
John C. Waterton 0 1 2 3
Edward F. Jackson 0 1 2 3
0 1 Department of Radiology, University of Texas M.D. Anderson Cancer Center , Houston , Texas, United States of America, 2 Department of Biostatistics, University of Texas M.D. Anderson Cancer Center , Houston , Texas, United States of America, 3 Department of Biostatistics Imaging Physics, University of Texas M.D. Anderson Cancer Center , Houston , Texas, United States of America, 4 Department of Biostatistics Veterinary Medicine and Surgery, University of Texas M.D. Anderson Cancer Center , Houston , Texas, United States of America , 5 Personalised Healthcare and Biomarkers, AstraZeneca, Alderley Park, Cheshire , United Kingdom , 6 Department of Medical Physics, University of Wisconsin , Madison, WI , United States of America
1 Funding: This work was supported by MDACC - AstraZeneca Research Collaboration Alliance; Cancer Center Support Grant (P30-CA016672); John S. Dunn Sr. Distinguished Chair in Diagnostic Imaging. AstraZeneca provided support in the form of salary for author JCW
2 Editor: Zhuoli Zhang, Northwestern University Feinberg School of Medicine, UNITED STATES
3 Current address: Biomedical Imaging Institute, University of Manchester, Manchester Academic Health Sciences Centre , Manchester , United Kingdom
Competing Interests: CS Ng and EF Jackson
received research funding from AstraZeneca. CS Ng
receives research funding from GE Healthcare. JC
Dynamic contrast-enhanced MRI (DCE-MRI) biomarkers have proven utility in tumors in
evaluating microvascular perfusion and permeability, but it is unclear whether
measurements made in different centers are comparable due to methodological differences.
To evaluate how commonly utilized analytical methods for DCE-MRI biomarkers affect both
the absolute parameter values and repeatability.
Materials and Methods
DCE-MRI was performed on three consecutive days in twelve rats bearing C6 xenografts.
Endothelial transfer constant (Ktrans), extracellular extravascular space volume fraction (ve),
and contrast agent reflux rate constant (kep) measures were computed using: 2-parameter
(“Tofts” or “standard Kety”) vs. 3-parameter (“General Kinetic” or “extended Kety”)
compartmental models (including blood plasma volume fraction (vp) with 3-parameter models);
individual- vs. population-based vascular input functions (VIFs); and pixel-by-pixel vs. whole
tumor-ROI. Variability was evaluated by within-subject coefficient of variation (wCV) and
variance components analyses.
DCE-MRI absolute parameter values and wCVs varied widely by analytical method. Absolute
parameter values ranged, as follows, median Ktrans, 0.09–0.18 min-1; kep, 0.51–0.92 min-1;
ve, 0.17–0.23; and vp, 0.02–0.04. wCVs also varied widely by analytical method, as follows:
Waterton was employed by AstraZeneca. This does
not alter the authors' adherence to PLOS ONE
policies on sharing data and materials.
mean Ktrans, 32.9–61.9%; kep, 11.6–41.9%; ve, 16.1–54.9%; and vp, 53.9–77.2%. Ktrans and
kep values were lower with 3- than 2-parameter modeling (p<0.0001); kep and vp were lower
with pixel- than whole-ROI analyses (p<0.0006). wCVs were significantly smaller for ve, and
larger for kep, with individual- than population-based VIFs.
DCE-MRI parameter values and repeatability can vary widely by analytical methodology.
Absolute values of DCE-MRI biomarkers are unlikely to be comparable between different
studies unless analyses are carefully standardized.
Imaging biomarkers can assess tumor perfusion and permeability, and are useful in assessing
response to therapy . In particular, dynamic contrast-enhanced magnetic resonance imaging
(DCE-MRI) provides biomarkers of tissue perfusion with proven utility in oncologic imaging,
including the assessment of treatment responses and development of anti-cancer therapies [2–
4]. However, these biomarkers are little-used outside the single-center setting, probably
because different implementations of the imaging acquisition and analysis have not been
shown to provide comparable biomarker values.
The DCE-MRI technique has been, and can be, used in clinical and pre-clinical settings, the
latter in particular where novel therapeutic agents are under investigation [5–9]. In both
settings, quantitative evaluations of the changes in derived tissue perfusion biomarkers have often
been the main objectives. While any one study will use the same algorithm and analytical
implementation for all subjects pre- and post-therapy, there is little consistency between
studies. Although biomarker values are quoted in absolute units (e.g. ktrans /min-1), it is unclear to
what extent absolute values reported from different studies are comparable. In this study we
evaluated three important analysis options: the choice of model, the method of derivation of
the input function, and the algorithm for aggregating pixel-wise data to derive whole-tumor
The technique of DCE-MRI depends on acquiring dynamic MRI data and applying an
appropriate physiological model to that data. A variety of tracer kinetic models have been
developed for these purposes; two commonly utilized models are variably termed the Tofts and
Kermode, “standard” Kety, or 2-parameter model [10–12], and the generalized kinetic,
“extended” Kety, or 3-parameter model . Application of these models allows derivation of
specific MRI perfusion parameters, such as the endothelial transfer constant (Ktrans), the
contrast agent reflux rate constant (kep), the extracellular extravascular space volume fraction (ve),
and the blood plasma volume fraction (vp).
Model-based derivations of DCE-MRI parameters require a vascular input function (VIF).
Obtaining reliable VIF data has been, and is, challenging, particularly in pre-clinical settings
where even the central vessels, e.g., aorta and inferior vena cava, are extremely small. Imaging
artifacts and the high cardiac rate of small animals add to the challenges. The unreliable nature
of some VIFs from individual subjects can potentially confound the overall estimates of
perfusion parameter values. In these situations, model or population-based VIFs have been
Tissue perfusion parameters for a region of interest (ROI) can be derived on a “whole
tumor” or “pixel-by-pixel” basis. Pixel-level data in principle offers a more detailed evaluation
and allows for intratumoral assessment of the heterogeneity of each measured parameter .
It is, however, prone to the potential challenges of additional computation time and
signal-tonoise ratio limitations.
In this study, we computed DCE-MRI parameter values utilizing all combinations of the
above methods on DCE-MRI images obtained on three successive days in each of twelve rat
xenografts. Absolute parameter values and repeatability were compared. An understanding of
repeatability provides data for assessing study results and for study design (namely,
determining sample sizes).
Our objectives were to compare the absolute values and test-retest repeatability of
DCE-MRI parameters analyzed by two tracer kinetic models (2-parameter vs. 3-parameter),
two different VIF input strategies (individual- vs. population-based), and two tissue ROI
approaches (whole tumor vs. pixel-by-pixel) in a rat tumor model.
Materials and Methods
The study was approved by the Institutional Animal Care and Use Committee (IACUC) of
The University of Texas M.D. Anderson Cancer Center (Protocol Number 09-06-12041). C6
rat glioma cells were obtained from the American Type Culture Collection (Manassas, VA,
USA). Five thousand C6 cells were injected subcutaneously into male Crl: NIH-Foxn1rnu T-cell
deficient, athymic nude rats (Charles River, Wilmington, MA). Rats weighted approximately
220 grams at the time of the experiment. Cells were injected in the flank region, at the
approximate axial level of the inferior aspect of the kidneys and distal aorta. Tumor measurements
were undertaken using calipers, and tumors allowed to grow until they reached a nominal size
of approximately 1 cm diameter. Each rat then underwent DCE-MRI on three consecutive
days. Animals were scanned in batches of 5 to 6 animals per cohort.
Animals were placed in an MRI-compatible cradle in which a 5-cm hole had been cut into
which the subcutaneous tumor could be located. Hair from around the tumor was shaved, and
the region of the tumor was placed in a “bath” of ultrasound gel to minimize air/tumor
susceptibility effects in the MRI imaging studies. A temperature controlled pad was placed
underneath the animals, and the animals were gently immobilized with tape. Animals were
anesthetized with 1–2% isoflurane in a 1 l/min O2 flow, and imaging was undertaken in free
respiration throughout. The imaging volume was targeted on the central portion of the tumor.
An estimate of tumor volume (cm3) was obtained from the formula for a spheroid, i.e.,
(π X Y2)/6000, where X and Y (in mm) were orthogonal tumor diameter measurements.
Data from a total of 12 sets of three consecutive days of scanning were obtained: 10 rats
underwent three consecutive DCE-MRI studies during a single week period; one animal
underwent the three consecutive DCE-MRI studies in two separate weeks. There was a technical
scanning failure on one MRI scan visit in one rat. The median size of tumors was 0.67 cm3
(range 0.09–1.53 cm3). At the end of the study, the rats were euthanized humanely by
inhalation of carbon dioxide.
MRI studies were undertaken using a 7.0 Tesla / 30 cm bore dedicated animal MRI scanner
(Bruker BioSpin, Billerica, MA). The MR scanning protocol consisted of the acquisition of
sagittal and axial T2-weighted images, axial T1-weighted images, axial DCE-MRI images, and
post-Gd axial T1-weighted images. For the DCE-MRI acquisition, a 3D fast spoiled gradient
echo sequence was used with TE = 1.7ms, TR = 10ms, 15° excitation pulse, 16-mm slab
thickness (yielding eight 2-mm slices), 128 x 80 matrix, and 60mm x 50mm field of view. To reduce
artifacts in the VIF from inflow effects, a spoiled hermite magnetization preparation pulse was
applied to excite an 8-cm slab located 2 mm caudal to the DCE-MRI slice package . The
temporal resolution was 6.4s, with total scan time of 320s (50 x 6.4s). Contrast agent was
administered after 10 baseline scans were acquired.
The MRI contrast agent was delivered through a tail vein as follows: 0.2 mmol/kg dose of
gadopentetate dimeglumine (Magnevist, Bayer Healthcare Pharmaceuticals, Wayne, NJ), via
an MR-compatible injection system (Harvard Apparatus, PHD 2000 Programmable, Plymouth
Meeting, PA). For a 200-gram rat, for example, 200 μl of contrast media at 1:5 dilution of
Magnevist:saline was administered over a period of 10 seconds. This was followed by a saline flush
of the same volume and injection rate.
The acquired DCE-MRI data were analyzed using the Kinmod module (version 3.0) within
the CineTool (version 8.2.1) environment (GE Healthcare, Waukesha, WI), utilizing two tracer
kinetic models (2-parameter vs. 3-parameter), two VIF input strategies (population-based vs.
individual-based), and two ROI approaches (whole tumor cross-sectional ROI vs.
pixel-bypixel within the ROI).
Tracer kinetic models: two- vs. three- parameter
Analyses were undertaken using two standard two-compartment tracer kinetic models: a)
2-parameter (“Tofts” or “standard Kety”) and b) 3-parameter (“General Kinetic” or “extended
a) 2-parameter model:.
where Ct(t) is the tracer concentration in tissue, Cp(t) is the tracer concentration in arterial
blood plasma, Ktrans is the volume endothelial transfer constant between blood plasma and
extravascular extracellular space (in min-1), and ve is the extravascular extracellular tissue
volume fraction (dimensionless).
This model is variably named the Tofts and Kermode, Larsson, or “standard (flow-limited)”
Kety model [10–12].
b) 3-parameter model:.
where the additional term, vp, is the blood plasma volume fraction (dimensionless).
This model is otherwise named the General Kinetic Model (GKM), or “extended” Kety
For both models, the contrast agent reflux rate constant, kep (in min-1) = Ktrans / ve.
Vascular input function: individual vs. population
We evaluated two commonly utilized vascular input functions (VIF): a) individually measured,
and b) population-based.
a) Individually measured VIF. For each animal and time point, the VIF was obtained
using a mask ROI defined by the study radiologist (CSN, more than 15 years’ experience)
containing the inferior vena cava in an imaging section near the center of the DCE-MRI scan
volume. From the mask ROI, the VIF was determined using an automated VIF identification
algorithm within the Kinmod module (Figs 1A and 2A). The VIF represents vascular
gadopentetate concentration measured in units of signal intensity change from baseline as a function of
time, i.e., ΔSI(t) = SI(t)–SI(baseline), where SI(baseline) was obtained by averaging the signal
Fig 1. Example of signal intensity profiles. A) Vascular input function (VIF) from a representative individual
animal (blue), and population average (red). Green symbols: tumor whole ROI data. Y-axis in units of signal
intensity change from baseline, ΔSI. B) Average signal intensity change (VIF) from all animals and scan
visits, including the standard error, for each data point (blue); and the fitted population-average VIF curve (red
line). Y-axis in units of signal intensity change from baseline, ΔSI.
intensity from frames 5–10, after achieving steady state but prior to contrast agent
b) Population average VIF. The population-based VIF was derived from the individual
VIFs (discussed above) obtained from all 12 rats of the study cohort and all available scan visits,
i.e., 3 visits per rat. The individual VIF curves for all animals were averaged and the resulting
data fitted to a biexponential function resulting in a population VIF given by
where t is in seconds and Cp(t) is in units of signal intensity change, ΔSI(t) (Fig 1B). The short
and long component amplitude and clearance rates thus determined were in good agreement
with previously published data [23,24].
Tumor ROI approaches: pixel-by-pixel vs. whole
ROIs were drawn by a single observer (CSN) using an electronic cursor and mouse for the
three central slices of tumor for each rat. DCE-MRI parameters were computed using all
combinations of the two tracer kinetic models and the two VIF methods above, using the same set
of ROIs for each tumor. DCE-MRI parameter values were obtained for each tumor ROI based
on: a) whole cross section ROI data, and b) pixel-by-pixel data within the ROI. In both cases,
the values of Ct(t) represent the tissue gadopentetate concentration measured in units of signal
intensity change from baseline as a function of time, as for the VIF curves. The same ROIs for
any given rat were used in each analysis.
a) “Pixel-by-pixel ROI” analyses. The intensity values of individual pixels within the
ROIs were used in computations. Pixels which demonstrated non-significant changes in signal
intensity following contrast agent injection, specifically less than a 50% increase, were not
included in further statistical analyses. In addition, positive-definite limits were placed on all fit
parameters and any pixels for which fit parameters were outside of such limits were flagged as
having fit parameters that were “not a number” (NaN).
b) “Whole ROI” analyses. The average intensity value of the whole tumor ROI was used
The average of the DCE-MRI parameter values derived from each of the three imaging slices
were used in subsequent analyses. Illustrative examples of DCE-MRI parametric maps are
presented in Fig 2.
Summary statistics of DCE-MRI parameters (Ktrans, kep, ve, and where applicable, vp) were
provided in the form of median and range. All data were transformed to the logarithmic scale, due
to right-skewness, prior to statistical analyses.
A linear mixed model was used to assess if each DCE-MRI parameter changed significantly
over three days. The linear mixed model took into account the correlation between
measurements of the same rat. No statistically significant trend was detected for any DCE-MRI
parameter in the span of three days (results not shown).
Comparisons between DCE-MRI parameters by analytical method were also based on linear
mixed models. Interactions between analytical methods and time were not significant (results
not shown), therefore, analytical methods and time were fit as main effects only. All the
Fig 2. Illustrative parametric maps using 3-parameter physiological model, individual VIF and pixel based analysis. Same animal as Fig 1. A) Source
trans-axial DCE-MRI image [blue outline is VIF input mask; red outline is tumor ROI mask]; B) Ktrans, C) kep, D) ve, and E) vp parameteric maps.
pairwise comparisons between analytical methods were estimated and p-values were adjusted
using the Bonferroni method to control the overall type I error rate at 5%.
A variance components analysis was used to estimate the inter- and intra-rat variances. The
intra-rat coefficient of variation (wCV) was calculated using the Bland-Altman method  as
follows: logarithms of the data were taken and within-rat standard deviations estimated, the
results were back-transformed to the raw scale (anti-log), and a value of one was subtracted
from them to obtain the wCVs.
All statistical analyses were two-sided and p-values of 0.05 or less were considered
statistically significant. Statistical analysis was carried out using SAS version 9 (SAS Institute, Cary,
NC). Plotting was performed using Spotfire S+ 8.2 (TIBCO Software Inc., Somerville, MA).
Summary statistics for the DCE-MRI parameters, by analytic method, are presented in Table 1.
Overall, DCE-MRI parameter values varied widely across the various analytical methods, i.e.
the tracer kinetic model, VIF input, and method of ROI analysis, with the range in median
DCE-MRI parameter values as follows: Ktrans, 0.09–0.18 min-1; kep, 0.51–0.92 min-1; ve, 0.17–
0.23; and vp, 0.02–0.04 (Table 1).
The wCV values also varied widely by analytical method, with mean wCVs ranging as
follows: Ktrans, 32.9–61.9%; kep, 11.6–41.9%; ve, 16.1–54.9%; and vp, 53.9–77.2% (Table 1).
Effect of tracer kinetic model: 2-parameter vs. 3-parameter
Ktrans and kep were significantly and consistently lower when using the 3-parameter model
compared to the 2-parameter model (p<0.0001); ve values were not significantly different
(Tables 1 and 2; Fig 3 (green vs. red datasets)).
The effects on repeatability of the DCE-MRI parameters (Ktrans, kep and ve) were
comparable, with no significant differences in wCVs across the two tracer kinetic models, as
demonstrated by overlapping wCV confidence bounds for corresponding parameters (Table 1).
Effect of VIF: individual- vs. population-based
There were no significant differences in DCE-MRI parameter values (Ktrans, kep, ve, and vp)
when comparing the utilization of individual-based and population-based VIFs in analyses.
Utilization of individual-based VIFs was associated with significantly smaller wCVs for ve
than with the population-based VIF, and conversely for kep (Fig 3 (1st and 2nd vs. 3rd and 4th
columns); Table 1, showing non-overlapping confidence bounds for corresponding wCVs).
There were no significant differences for Ktrans and vp when comparing individual- vs.
population-based VIF analyses.
Effect of ROI: pixel-by-pixel vs. whole tumor
kep and vp values were significantly lower for pixel-by-pixel compared to whole ROI analyses
(p<0.0006), and not significantly different for Ktrans and ve values (Tables 1 and 2; Fig 3 (1st
and 3rd vs. 2nd and 4th columns)).
There were no significant differences in variability (wCV) between pixel and whole ROI
analyses across all the DCE-MRI parameters (Ktrans, kep, ve and vp).
Between- and within-animal variation
The results of our variance component analyses are presented in Table 3. This shows generally
larger intra-rat than inter-rat variances across all DCE-MRI parameters.
lIeRO eanwM % .501 .158* ††.940 .358 .411* ††.251 .359 .239 .433** †.619 .434 .149** †.817 .772
raam raaPm trasnK kep ve trasnK kep ve vp trasnK kep ve trasnK kep ve vp
Table 2. Summary of pairwise comparisons between DCE-MRI parameters by analytical method. P-values based on linear mixed model on the
logarithmic scale. The linear mixed model took into account correlation between measurements from the same rat. A Bonferroni adjustment was used to control
the overall type I error rate, which with 13 pairwise comparisons set the cutoff point for declaring statistical significance as 0.05/13 = 0.0038, which are
The parameters evaluated here are widely regarded by drug developers and others as
“biomarkers” within the definition introduced by Atkinson et al. . However, to qualify as
biomarkers, they must be “objectively quantified” . If different studies derive different values for the
same biomarker because of subtle differences in analysis, the “objective quantification” is
deficient . This work was undertaken to assess the impact on parameter values and
repeatability of commonly utilized analytical methods in the DCE-MRI arena. Specifically, we explored
the effects of two commonly utilized tracer kinetic models (2-parameter vs. 3-parameter), two
VIF options (population- vs. individual-based), and two ROI analytical approaches (whole
tumor ROI vs. pixel-by-pixel tumor ROI). Each, in principle, has its theoretical advantages and
Our results suggest that DCE-MRI parameter values vary widely depending on the
analytical methods utilized, in some cases almost two-fold (e.g., Ktrans, kep, vp). Overall wCV values
also varied widely by analytical methods, with wCVs ranging as follows: Ktrans 32.9–61.9%; kep,
11.6–41.9%; ve, 16.1–54.9%; and vp, 53.9–77.2%.
In terms of the tracer kinetic models utilized, the 3-parameter model might in principle be
expected to provide a more complete reflection of the underlying tracer kinetics compared to
the 2-parameter model since it does not neglect the intravascular tracer contribution (i.e., the
vp term) as does the 2-parameter model. Our results suggested that the 3-parameter model
yielded significantly lower Ktrans and kep values than the corresponding 2-parameter model,
and no significant differences in ve values. Since the 2-parameter model neglects the
intravascular signal, the bias (i.e. artefactual elevation of Ktrans and kep in the 2-parameter model) is not
unexpected. Absolute values of Ktrans and kep from different studies which use these different
models cannot be assumed comparable, even though the same biomarker name and units are
reported. There were no significant differences in wCVs for all parameters, which suggests that
repeatability was not substantially affected by the physiological model applied for these
With regards to the choice of VIF inputs for the model-based analyses, utilization of
individual-based VIFs might be expected to yield more reliable results than using a
populationbased VIF, since vascular tracer profiles can vary quite widely due to variations in IV contrast
delivery, cardiac output, renal function, etc. It is quite possible, however, that the theoretical
Fig 3. Scatter plots of 3 day time points, of horizontal row a) Ktrans, b) kep, c) ve, d) vp, by 2- (red lines) vs. 3-parameter (green lines) models; with
separate plots for pixel-by-pixel vs. whole tumor analyses, and by individual- vs. population-based VIFs. Y-axes for Ktrans and kep in min-1: ve and vp,
unitless. Note: vp can only be derived with the 3-parameter model. (Note: one missing data point for one rat)
advantage of utilizing individual-based VIFs might be out-weighed by the substantial technical
challenges in obtaining reliable VIFs from DCE-MRI studies. Our results indicated that there
were no significant differences in DCE-MRI parameter values obtained between
individualand population-based VIF analyses. However, wCVs when using individual VIFs were
significantly smaller for ve than when using the population VIF, and conversely for kep. In
circumstances in which individually measured VIFs might be unreliable, the utilization of population
VIFs may be necessary [19,28]. Our results do not suggest that the use of a population VIF will
bias the absolute DCE-MRI biomarker values.
Since each of our DCE-MRI biomarkers is an intensive scalar variable, and since the voxel
volumes are identical, the whole-tumor value for each biomarker should in principle be
identical to the mean of the individual voxel biomarker values. However, for real-world (noisy) data,
that identity may be lost in the propagation of errors, leading to variability and/or bias.
Regarding pixel-by-pixel vs. whole tumor based ROI evaluations, the former, in principle, is able to
display and better reflect the heterogeneous nature of tumors. However, it is computationally
more demanding and more vulnerable to signal-to-noise ratio constraints. Furthermore,
deriving a simple statistic that adequately summarizes the resultant individual pixel-based
parameter distributions is challenging. Adopting the simple approach of using median values to
summarize the distributions, our results indicated that pixel-by-pixel based evaluation of ROIs
yielded significantly lower kep and vp values than whole-tumor based ROI evaluations. As such,
it may be important in some circumstances to be able to capture detailed spatial information
about tumor heterogeneity.
The tumor time-intensity profile is clearly affected by the tracer input profile. The latter in
turn is affected by tracer input delivery (i.e., the intravenous injection) and physiological
parameters (e.g., cardiac output and renal/excretion function), all of which can vary between
studies. Acquisition of a reliable VIF presents substantial challenges. Difficulties include
motion- and flow-related artifacts. The difficulties of acquiring reliable VIFs are compounded
in small animal studies by the very small cross-sectional area of the major vessels. Indeed,
intensity variations (from noise and artifacts) were evident in our VIF time-intensity plots (Fig
1A). We were able to mitigate the flow related artifacts in our acquisition protocol by
application of a saturation band between the site of injection and the imaging volume, but inevitably
with some loss of temporal resolution. We took precautions to control for the delivery of
intravenous contrast medium. We used a pump injector, with fixed gadolinium and saline flush
volumes and flow rates, a fixed site of injection (the tail vein), and a constant length of tubing
between the injector and tail vein.
There have been some conflicting reports as to the effect of using individual- compared to
population-based VIFs: Rijpkema and co-authors  has reported that individual arterial
input functions (AIFs), compared to population-based AIFs, improved repeatability of kep.
Parker and co-authors  reported that variation in Ktrans, ve, and vp values were smaller
when using a population-based AIF compared to an individual-based AIF in a study of tumors
in human patients. Their differing conclusions may be partly due to the relative differences in
the consistency of the individual VIFs obtained in their studies. Also a variety of models have
been proposed to derive population VIFs, and these two studies employed different
approaches. The extent to which such models might influence the conclusions is beyond the
scope of this work.
The differing views related to VIF estimations in the studies above in humans are paralleled
in the pre-clinical arena. The small blood volume and rapid vascular dynamics inherent to
small animals necessitate very rapid sampling schemes in order to accurately capture the peak
of intravascular enhancement, corresponding to the maximum concentration of contrast agent
after injection, and acquisition strategies that are tuned for rapid AIF sampling generally
compromise the spatial resolution and coverage of tumor. Studies utilizing acquisitions that are
optimized for AIF measurement with very rapid sampling may provide reduced variability
using individual measurements [23,30,31]. In the absence of AIF estimates with high temporal
resolution, or in the presence of high noise, repeatability may be improved by use of a
parameterized population average . It has also been shown that measurements derived from
individual and averaged AIFs correlate strongly when a strictly controlled contrast administration
protocol is used . In this work, we employed a 3D acquisition protocol that is biased
towards anatomic coverage with relatively slow temporal sampling of the AIF. Our study
identified no statistically significant differences in parameter values when using the individual or
averaged VIFs, but did find some differences in repeatability (wCV) with some specific
Previous studies with small animals have reported intra-animal wCVs for Ktrans and ve of
18% and 7%, respectively, in a mouse model using PC3 prostate tumors, a 2-parameter model
and a pooled (population) VIF . A previous study of ours which compared DCE-MRI and
DCE-CT in the same tumor model as in the current study reported wCVs for Ktrans, kep and ve
of 23%, 16% and 20% respectively . This study employed a different intravenous injection
technique, VIF model and acquisition protocol compared to the current study. The first study
above examined repeatability over just two, unlike our three, scan visits in our previous and
Our variance components analysis allowed us to assess the relative contributions of
variations between animals (inter-rat) and between individual scan days (intra-rat), to the overall
variation. We found that for all DCE-MRI parameters, the intra-rat contributions were
comparable to or larger than the inter-rat contributions to the overall variation in this C6 model.
Tumor vasculature is intrinsically chaotic and unstable, and it is not surprising that its
day-today variations may exceed relatively small between-rat variations, given the uniform and
controlled tumor implantation techniques and animals utilized.
We acknowledge and recognize limitations in our study. Our study was undertaken with
only one tumor model. The physiological compartment models used in our study are widely
used. However, we utilized a single software implementation of these models in this study. It is
likely that algorithmic differences within software implementations may give rise to differences
in the parameter values.
In our repeatability study, it would have been desirable to obtain scan-rescan measurements
in close succession. Unfortunately, any persistent gadolinium would potentially alter signal
intensities and saturation, thereby confounding the measurements. It was considered that this
difficulty would be mitigated by imposing a one day interval between scans. Unlike other
studies which have restricted their evaluations to just two sequential evaluations, we undertook
serial imaging evaluations on three consecutive days. Tumors inevitably changed in size and
potentially in their perfusion properties over the 48 hours of our serial imaging. Although we
observed increases in tumor size with time, we did not observe any systematic changes in
perfusion parameters on formal statistical testing (data not presented). Even if possible,
undertaking repeat scans in short succession in one scanner visit would have removed important
variables in DCE-MRI experiments, which include handling of animals, anesthesia and
physical re-positioning/localization. The variabilities obtained, therefore, more closely mimic
longitudinal experiments in which there are typically intervening therapeutic interventions.
We limited our tumor ROI evaluations to a single observer; an examination of inter- and
intra-observer variability was beyond the scope of this work. It was also beyond the scope of
the current study to explore the impact of our various analytical options with different animal
models and DCE-MRI acquisition protocols.
Our results indicate that DCE-MRI parameter values and repeatability can vary widely
depending on the specific analytical methods utilized. Comparisons across studies especially of
absolute parameter values should be interpreted with caution. Given the wide ranges in
variability of parameters, it would be prudent to incorporate scan-rescan repeatability in studies.
Efforts should be made to acquire reliable VIF data for analyses. An understanding of
repeatability of DCE-MRI measurements provides insight into what observed changes can be
considered significant, and can also assist in design of future studies and sample-size calculations.
We thank Sandeep N. Gupta, PhD, GE Healthcare, for use of the DCE-MRI analytical software
package, Kinmod, under a research agreement, and the staff of the Small Animal Imaging
Facility, MD Anderson Cancer Center.
Conceived and designed the experiments: CSN WW JAB EFJ. Performed the experiments:
CSN JAB MKR LH DWB SK. Analyzed the data: CSN WW EFJ. Contributed
reagents/materials/analysis tools: JAB MKR LH EFJ. Wrote the paper: CSN WW JAB JCW EFJ.
1. O'Connor JP , Jackson A , Parker GJ , Roberts C , Jayson GC ( 2012 ) Dynamic contrast enhanced MRI in clinical trials of antivascular therapies . Nat Rev Clin Oncol 9 : 167 - 177 . doi: 10.1038/nrclinonc. 2012 .2 PMID: 22330689
2. Laking GR , West C , Buckley DL , Matthews J , Price PM ( 2006 ) Imaging vascular physiology to monitor cancer treatment . Crit Rev Oncol Hematol 58 : 95 - 113 . PMID: 16387510
3. O'Connor JP , Jackson A , Parker GJ , Jayson GC ( 2007 ) DCE-MRI biomarkers in the clinical evaluation of antiangiogenic and vascular disrupting agents . Br J Cancer 96 : 189 - 195 . PMID: 17211479
4. Leach MO , Morgan B , Tofts PS , Buckley DL , Huang W , Horsefield MA , et al. ( 2012 ) Imaging vascular function for early stage clinical trials using dynamic contrast-enhanced magnetic resonance imaging . Eur Radiol 22 : 1451 - 1464 . doi: 10.1007/s00330- 012 - 2446 -x PMID: 22562143
5. Morgan B , Thomas AL , Drevs J , Hennig J , Buchert M , Jivan A , et al. ( 2003 ) Dynamic contrastenhanced magnetic resonance imaging as a biomarker for the pharmacological response of PTK787/ ZK 222584, an inhibitor of the vascular endothelial growth factor receptor tyrosine kinases, in patients with advanced colorectal cancer and liver metastases: results from two phase I studies . J Clin Oncol 21 : 3955 - 3964 . PMID: 14517187
6. Stevenson JP , Rosen M , Sun W , Gallagher M , Haller DG , Vaughn D , et al. ( 2003 ) Phase I trial of the antivascular agent combretastatin A4 phosphate on a 5-day schedule to patients with cancer: magnetic resonance imaging evidence for altered tumor blood flow . J Clin Oncol 21 : 4428 - 4438 . PMID: 14645433
7. Evelhoch JL , LoRusso PM , He Z , DelProposto Z , Polin L , Corbett TH , et al. ( 2004 ) Magnetic resonance imaging measurements of the response of murine and human tumors to the vascular-targeting agent ZD6126 . Clin Cancer Res 10 : 3650 - 3657 . PMID: 15173071
8. Willett CG , Boucher Y , di Tomaso E , Duda DG , Munn LL , Tong RT , et al. ( 2004 ) Direct evidence that the VEGF-specific antibody bevacizumab has antivascular effects in human rectal cancer . Nat Med 10 : 145 - 147 . PMID: 14745444
9. Liu G , Rugo HS , Wilding G , McShane TM , Evelhoch JL , Ng C , et al. ( 2005 ) Dynamic contrast-enhanced magnetic resonance imaging as a pharmacodynamic measure of response after acute dosing of AG013736, an oral angiogenesis inhibitor, in patients with advanced solid tumors: results from a phase I study . J Clin Oncol 23 : 5464 - 5473 . PMID: 16027440
10. Tofts PS , Kermode AG ( 1991 ) Measurement of the blood-brain barrier permeability and leakage space using dynamic MR imaging . 1. Fundamental concepts . Magn Reson Med 17 : 357 - 367 . PMID: 2062210
11. Larsson HB , Stubgaard M , Frederiksen JL , Jensen M , Henriksen O , Paulson OB , et al. ( 1990 ) Quantitation of blood-brain barrier defect by magnetic resonance imaging and gadolinium-DTPA in patients with multiple sclerosis and brain tumors . Magn Reson Med 16 : 117 - 131 . PMID: 2255233
12. Kety SS ( 1951 ) The theory and applications of the exchange of inert gas at the lungs and tissues . Pharmacol Rev 3 : 1 - 41 . PMID: 14833874
13. Tofts PS ( 1997 ) Modeling tracer kinetics in dynamic Gd-DTPA MR imaging . J Magn Reson Imaging 7 : 91 - 101 . PMID: 9039598
14. Weinmann HJ , Laniado M , Mutzel W ( 1984 ) Pharmacokinetics of GdDTPA/dimeglumine after intravenous injection into healthy volunteers . Physiol Chem Phys Med NMR 16 : 167 - 172 . PMID: 6505043
15. Galbraith SM , Lodge MA , Taylor NJ , Rustin GJ , Bentzen S , Stirling JJ , et al. ( 2002 ) Reproducibility of dynamic contrast-enhanced MRI in human muscle and tumours: comparison of quantitative and semiquantitative analysis . NMR Biomed 15 : 132 - 142 . PMID: 11870909
16. Lankester KJ , Taylor NJ , Stirling JJ , Boxall J, D'Arcy JA , Leach MO , et al. ( 2005 ) Effects of platinum/ taxane based chemotherapy on acute perfusion in human pelvic tumours measured by dynamic MRI . Br J Cancer 93 : 979 - 985 . PMID: 16234826
17. Roberts C , Issa B , Stone A , Jackson A , Waterton JC , Parker GJ ( 2006 ) Comparative study into the robustness of compartmental modeling and model-free analysis in DCE-MRI studies . J Magn Reson Imaging 23 : 554 - 563 . PMID: 16506143
18. Lankester KJ , Taylor JN , Stirling JJ , Boxall J, d'Arcy JA , Collins DJ , et al. ( 2007 ) Dynamic MRI for imaging tumor microvasculature: comparison of susceptibility and relaxivity techniques in pelvic tumors . J Magn Reson Imaging 25 : 796 - 805 . PMID: 17347990
19. McGrath DM , Bradley DP , Tessier JL , Lacey T , Taylor CJ , Parker GJ ( 2009 ) Comparison of modelbased arterial input functions for dynamic contrast-enhanced MRI in tumor bearing rats . Magn Reson Med 61 : 1173 - 1184 . doi: 10.1002/mrm.21959 PMID: 19253360
20. Loveless ME , Halliday J , Liess C , Xu L , Dortch RD , Whisenant J , et al. ( 2012 ) A quantitative comparison of the influence of individual versus population-derived vascular input functions on dynamic contrast enhanced-MRI in small animals . Magn Reson Med 67 : 226 - 236 . doi: 10.1002/mrm.22988 PMID: 21688316
21. Ragan DK , Bankson JA . Suppression of vascular enhancement artifacts through the use of a multiband, selectively spoiled radiofrequency excitation pulse . J Magn Reson Imaging 33 : 1256 - 1261 , 2011 . doi: 10.1002/jmri.22523 PMID: 21509887
22. Ashton EA , Kwok E , Evelhoch J. Conversion from signal intensity to Gd concentration may be unnecessary for perfusion assessment of tumors using DCE-MRI . Proceedings of the 15th Annual Meeting of ISMRM , Berlin, 2007 . (abstract 2813)
23. Bradley DP , Tessier JL , Checkley D , Kuribayashi H , Waterton JC , Kendrew J , et al. ( 2008 ) Effects of AZD2171 and vandetanib (ZD6474, Zactima) on haemodynamic variables in an SW620 human colon tumour model: an investigation using dynamic contrast-enhanced MRI and the rapid clearance blood pool contrast agent, P792 (gadomelitol) . NMR Biomed 21 : 42 - 45 . PMID: 17458919
24. Bradley DP , Tessier JJ , Lacey T , Scott M , Jurgensmeier JM , Odedra R , et al. ( 2009 ) Examining the acute effects of cediranib (RECENTIN, AZD2171) treatment in tumor models: a dynamic contrastenhanced MRI study using gadopentate . Magn Reson Imaging 27 : 377 - 384 . doi: 10.1016/j.mri. 2008 . 07.021 PMID: 18814988
25. Bland M ( 2003 ) Clinical measurement . In: Bland M, editor. An introduction to medical statistics . Oxford: Oxford University Press. pp. 268 - 275 .
26. Atkinson AJ , Colburn WA , DeGruttola VG , DeMets DL , Downing GJ , Hoth DF , et al. ( 2001 ) Biomarkers and surrogate endpoints: preferred definitions and conceptual framework . Clin Pharmacol Ther 69 : 89 - 95 . PMID: 11240971
27. Waterton JC ( 2013 ) Translational Magnetic Resonance Imaging and Spectroscopy: Opportunities and Challenges . In: Garrido L, Beckmann N , editors. New Applications of NMR in Drug Discovery and Development . Cambridge: The Royal Society of Chemistry . pp. 333 - 360 .
28. Parker GJ , Roberts C , Macdonald A , Buonaccorsi GA , Cheung S , Buckley DL , et al. ( 2006 ) Experimentally-derived functional form for a population-averaged high-temporal-resolution arterial input function for dynamic contrast-enhanced MRI . Magn Reson Med 56 : 993 - 1000 . PMID: 17036301
29. Rijpkema M , Kaanders JH , Joosten FB , van der Kogel AJ , Heerschap A ( 2001 ) Method for quantitative mapping of dynamic MRI contrast agent uptake in human tumors . J Magn Reson Imaging 14 : 457 - 463 . PMID: 11599071
30. Pickup S , Zhou R , Glickson J ( 2003 ) MRI estimation of the arterial input function in mice . Acad Radiol 10 : 963 - 968 . PMID: 13678084
31. Ragan DK , Lai SY , Bankson JA ( 2011 ) Fast, reproducible measurement of the vascular input function in mice using constrained reconstruction and cardiac sampling . NMR Biomed 24 : 373 - 384 . doi: 10. 1002/nbm.1601 PMID: 20862661
32. Checkley D , Tessier JJ , Kendrew J , Waterton JC , Wedge SR ( 2003 ) Use of dynamic contrastenhanced MRI to evaluate acute treatment with ZD6474, a VEGF signalling inhibitor, in PC-3 prostate tumours . Br J Cancer 89 : 1889 - 1895 . PMID: 14612898
33. Ng CS , Waterton JC , Kundra V , Brammer D , Ravoori M , Han L , et al. ( 2012 ) Reproducibility and comparison of DCE-MRI and DCE-CT perfusion parameters in a rat tumor model . Technol Cancer Res Treat 11 : 219 - 288 .