Image-Guided PO2Probe Measurements Correlated with
Parametric Images Derived from18F-Fluoromisonidazole
Small-Animal PET Data in Rats
Rachel M. Bartlett1, Bradley J. Beattie1, Manoj Naryanan2, Jens-Christoph Georgi3, Qing Chen1, Sean D. Carlin1,
Gordon Roble4, Pat B. Zanzonico1, Mithat Gonen5, Joseph O’Donoghue1, Alexander Fischer3, and John L. Humm1
1Department of Medical Physics, Memorial Sloan-Kettering Cancer Center, New York, New York;2Philips Research NA, Briarcliff
Manor, New York;3Philips Technologie GmbH Forschungslaboratorien, Molecular Imaging Systems (MIS), Aachen, Germany;
4Research Animal Resource Center (RARC), Sloan-Kettering Institute, New York, New York; and5Epidemiology-Biostatistics,
Memorial Sloan-Kettering Cancer Center, New York, New York
18F-fluoromisonidazole PET, a noninvasive means of identifying
hypoxia in tumors, has been widely applied but with mixed
results, raising concerns about its accuracy. The objective of
this study was to determine whether kinetic analysis of dynamic
18F-fluoromisonidazole data provides better discrimination of
tumor hypoxia than methods based on a simple tissue-to-
plasma ratio. Methods: Eleven Dunning R3327-AT prostate tu-
mor-bearing nude rats were immobilized in custom-fabricated
whole-body molds, injected intravenously with18F-fluoromiso-
nidazole, and imaged dynamically for 105 min. They were then
transferred to a robotic system for image-guided measurement
of intratumoral partial pressure of oxygen (PO2). The dynamic
18F-fluoromisonidazole uptake data were fitted with 2 variants
of a 2-compartment, 3-rate-constant model, one constrained to
have K1equal to k2and the other unconstrained. Parametric
images of the rate constants were generated. The PO2measure-
ments were compared with spatially registered maps of kinetic
rate constants and tumor-to-plasma ratios. Results: The con-
strained pharmacokinetic model variant was shown to provide
fits similar to that of the unconstrained model and did not
introduce significant bias in the results. The trapping rate con-
stant, k3, of the constrained model provided a better discrimi-
nation of low PO2than the tissue-to-plasma ratio or the k3of the
unconstrained model. Conclusion: The use of kinetic modeling
on a voxelwise basis can identify tumor hypoxia with improved
accuracy over simple tumor-to-plasma ratios. An effective
means of controlling noise in the trapping rate constant, k3,
without introducing significant bias, is to constrain K1equal
to k2during the fitting process.
Key Words: hypoxia; [18F]-FMISO; [18F]-fluoromisonidazole;
kinetic modeling; parametric images; OxyLite probe
J Nucl Med 2012; 53:1–8
The presence of hypoxia in tumors is associated with
radio- and chemotherapy resistance, enhanced angiogenesis
and metastatic potential, and, in general, poor clinical out-
come (1–4). However, direct measurement of tumor oxygen-
ation using oxygen microelectrodes is subject to several
limitations (5,6). Direct measurement is restricted to easily
accessible lesions, and the limit on the number of measure-
ments that can practically be made generally results in
undersampling of the tumor. Moreover, such measurements
cannot distinguish between necrotic and viable but hypoxic
intratumoral regions. A noninvasive imaging method of iden-
tifying tumor hypoxia has therefore been a long-sought goal.
Of the various nuclear, MRI, and optical methods that
have been devised to image tumor hypoxia, one of the most
extensively evaluated clinically is18F-fluoromisonidazole
PET.18F-fluoromisonidazole is a 2-nitroimidazole com-
pound, subject to bioreduction by intracellular nitroreduc-
tases. In the presence of sufficient intracellular O2, the
nitroimidazole is back-oxidized to the parent compound,
which readily diffuses from the cell. Otherwise, further
irreversible reductions occur and the reactive metabolites
covalently bind to intracellular macromolecules, leading, in
the case of18F-fluoromisonidazole, to accumulation of the
18F radiolabel within the hypoxic tissue (7,8).
In humans,18F-fluoromisonidazole is metabolized slowly
and clears from the body with a half-time of approximately
6 h. It is freely and homogeneously distributed, achieving
tissue-to-plasma (T:P) ratios of about unity in most normal
tissues. Koh et al. (7) proposed that an18F-fluoromisonidazole
T:P ratio greater than 1.4 at 2 or more hours after injection is
indicative of viable hypoxic tissue. Also, several in vitro stud-
ies show a near-binary pattern of18F-fluoromisonidazole up-
take for O2levels above or below a defined threshold (9),
supporting the use of a simple ratio for clinical imaging.
However, a recent study examining voxelized tumor-to-
muscle ratios (similar to T:P) failed to show a correlation
with spatially registered Eppendorf partial pressure of oxy-
gen (PO2) measurements (10).
Received Jan. 24, 2012; revision accepted May 31, 2012.
For correspondence or reprints contact: John L. Humm, Memorial Sloan-
Kettering Cancer Center, 1275 York Ave., New York, NY 10065.
Published online nnnn.
COPYRIGHT ª 2012 by the Society of Nuclear Medicine and Molecular
18F-FLUOROMISONIDAZOLE PET DATA AND PO2 • Bartlett et al.1
jnm103523-sn n 8/29/12
Copyright 2012 by Society of Nuclear Medicine.
Journal of Nuclear Medicine, published on August 29, 2012 as doi:10.2967/jnumed.112.103523
One possible explanation for this negative finding is that
T:P ratios for18F-fluoromisonidazole change over time in
a perfusion-dependent manner. Low blood perfusion
reduces the rate of18F-fluoromisonidazole supply, and thus
uptake in such regions may achieve a 1.4 T:P threshold only
at later times, if at all. In fact, neither the time after in-
jection nor the threshold ratio recommended by Koh et al.
was defined with reference to hypoxic tissue. Instead, these
were based on normal-tissue equilibration rates and the
level of noise in the images, such that normoxic voxels
would be reliably below the defined threshold.
An alternative approach is to use pharmacokinetic model-
ing to estimate a parameter more directly dependent on
oxygen levels. Several models describing the uptake of18F-
fluoromisonidazole have been proposed (11–14), requiring
between 4 and 7 free parameter values to be determined dur-
ing the data-fitting process. Three of these models (12–14)
have been used to generate parametric images, but no one
has yet validated the pertinent oxygen-dependent parameter
against direct PO2probe measurements.
Our goal in this work was to determine whether pharma-
cokinetic modeling of18F-fluoromisonidazole uptake can
provide an oxygen-dependent rate constant that is better than
T:P ratio at demarcating hypoxic and nonhypoxic regions
within tumors. The studies were conducted on nude rats
bearing subcutaneous tumors. Robotically controlled PO2
probe measurements were taken at points spatially regis-
tered to18F-fluoromisonidazole dynamic PET image data.
In a subset of tumor-bearing animals, the hypoxia marker
pimonidazolewas coadministered with18F-fluoromisonidazole
and the perfusion marker Hoechst 33342 administered
shortly before sacrifice. Subsequently, autoradiographic
(18F-fluoromisonidazole) and fluorescence (pimonidazole,
Hoechst 33342) images of tumor tissue sections were ac-
quired and compared.
MATERIALS AND METHODS
All animal experiments and procedures were approved by the
Memorial Sloan-Kettering Cancer Center Institutional Animal
Care and Use Committee in compliance with National Institute of
Health regulations on the research use of rodents.
Eleven male nude rats (average weight, ;300 g) were injected
subcutaneously in the left hind limb with approximately 106Dun-
ning R3327-AT prostate tumor cells. Imaging was performed be-
tween 2 and 3 wk after tumor inoculation. Substantial hypoxic
tumor volumes were found in tumors larger than 2,000 mm3,
calculated as volume 5 ðp
3 orthogonal dimensions of the tumor as measured with a caliper.
Animals were anesthetized using a mixture of isoflurane (5%
induction, 1.5% maintenance) and air for the entire duration of the
study and were euthanized at its completion in accordance with the
guidelines of the Institutional Animal Care and Use Committee.
Throughout the study, animals were kept immobilized in a prone
position in a custom-fabricated, animal-specific foam mold (Soule
Medical) within a hemicylindric acrylic couch (15). Tumors from 4
rats were excised after sacrifice for further analysis.
6Þ · x · y · z; where x, y, and z are the
18F-fluoromisonidazole (specific activity, 370 MBq/mg) was
prepared according to the method of Grierson et al. (16) by the
Cyclotron and Radiochemistry Service at Memorial Sloan-Kettering
Cancer Center. Animals were positioned in a small-animal scanner
(microPET Focus 120; Concorde Microsystems) with tumors cen-
tered in the field of view.
Before the18F-fluoromisonidazole injection, a 2-min static PET
image was acquired of the rat on the couch together with a remov-
able registration plate (
Fig. 1A) containing 4 wells each loaded
with 10 mL of an approximately 370 kBq/mL (10 mCi/mL) solu-
tion of18F. This image was used to spatially register the subse-
quently acquired PET images with the robotically guided PO2
probe measurements as described elsewhere (17–19).
The animals were then injected via the tail vein with
approximately 55 MBq (;1.5 mCi) of18F-fluoromisonidazole
and imaged dynamically over 105 min, beginning at the time of
injection. Dynamic images were subsequently reconstructed off-
line using maximum a posteriori estimation into a 128 · 128 · 95
matrix (voxel dimensions, 0.87 · 0.87 · 0.79 mm) and time-
binned into 2 · 6, 4 · 12, 1 · 60, 9 · 120, 10 · 180, and 11 ·
300 s frames.
Immediately after the dynamic acquisition, a 5-min static PET
scan was acquired and reconstructed. This scan was used for
image guidance of the PO2probe measurements.
foam mold, positioned on robot platform. Registration plate is cen-
tered over rat tumor. (B) Top image is rat tumor with OxyLite probe
sampling PO2along predefined track. Bottom image is target region
for sampling PO2as defined on 5-min static PET image.
(A) Anesthetized rat immobilized in custom-fabricated
2THE JOURNAL OF NUCLEAR MEDICINE • Vol. 53 • No. 10 • October 2012
jnm103523-sn n 8/29/12
PET–to–Robot Coordinate Registration
On completion of imaging, the couch with the anesthetized
immobilized animal in place was transferred to the image-guided
robotic system. The initial and final static PET scans (for
registration and target definition, respectively) were loaded into
the robot application software (3D Slicer; www.slicer.org, Engi-
neering Research Center for Computer Integrated Surgical Sys-
tems and Technology, Johns Hopkins University). This registration
process and its accuracy are described in detail elsewhere (17–19).
Sets of trajectories (vertical tracks) were defined on the late
static PET image (Fig. 1B). The robot then moved the PO2probe
(OxyLite 4000; Oxford Optronix) to a location directly above each
trajectory. A needle was used to puncture the skin and fascia
covering the tumor. The probe was then advanced through an in-
dwelling cannula (used to improve mechanical stability) until con-
tact was made. Subsequent probe penetration of the tissue was
performed under robot control. Measurements of PO2were per-
formed at 0.5-mm intervals along each probe trajectory (initial
advance of 0.8 mm followed by 0.3-mm retraction to relieve pres-
sure at the probe tip). This process was repeated for each defined
trajectory. After the last PO2measurement, animals were eutha-
nized in place by isoflurane overdose (;3 h after injection).
Tissue Processing for Microscopic Analysis
At the time of18F-fluoromisonidazole administration, 4 animals
were coinjected with the hypoxic cell marker pimonidazole hydro-
chloride ([1-[(2-hydroxy-3-piperidinyl)propyl]-2-nitroimidazole hy-
drochloride; 20 mg/mL in normal saline; 80 mg/kg; Chemicon
International). These same animals were also injected with the
fluorescent dye Hoechst 33342 (5 mg/mL in normal saline; 15
mg/kg; Sigma-Aldrich) via the tail vein 1 min before sacrifice.
Immediately after sacrifice, a set of fiduciary angiocatheters was
inserted into the tumor perpendicular to the coronal imaging
plane. The tumor was then excised, frozen on dry ice, embedded
in optimal cutting medium (OCT 4583; Sakura Finetek) and
mounted on the planchet of a Microm HM500 cryostat microtome
(Microm International GmbH) such that the plane of tissue sec-
tions was cut parallel to the PET coronal imaging plane. The
angiocatheter needles were then completely retracted, leaving only
the plastic sleeve in place. Multiple sets of contiguous 10-mm-thick
tissue sections were acquired at 0.5-mm intervals within the tumor
block using the microtome digital readout to track perpendicular
Autoradiography and Fluorescence Images
Digital autoradiograms of
tained by placing a tumor section from each contiguous set against
a Fujifilm BAS-MS2325 imaging plate (Fuji Photo Film Co.) in a
light-tight cassette. Plates were exposed overnight and read by
a Fujifilm BAS-1800II bioimaging analyzer (Fuji Photo Film
Co.), generating digital images with pixel dimensions of 50 ·
Images of the distributions of pimonidazole and Hoechst 33342
were obtained after completion of18F-fluoromisonidazole digital
autoradiogram exposures. To eliminate possible misregistration,
we used the same tumor sections as were used for digital autora-
diography. Sections were fixed in a 4% paraformaldehyde solution
for 12 min, followed by incubation in Superblock/PBS (Thermo
Scientific) for 30 min at room temperature. Immunofluorescence
18F-fluoromisonidazole were ob-
staining for pimonidazole was performed as described previously
Images of tumor sections were acquired at 100· magnification
using a fluorescence microscope (Nikon Diaphot 300) equipped
with a computer-controlled, motorized stage and digital camera
for image capture (Photometrics Coolsnap EZ). Pimonidazole and
Hoechst 33342 were imaged using green and blue filters, respec-
tively. Composite images of whole tumor sections were generated
by stitching together individual microscopic images using Image-
Pro software (Image-Pro Plus, version 7.0; MediaCybernetics).
Input Function Determination. Amira 5.2.2 (Mercury Com-
puter Systems) was used for initial dynamic small-animal PET
image visualization and processing. Image-derived input functions
were obtained from a volume of interest (VOI) placed over the
proximal end of the ventral caudal artery (identified from
a summed image of the first 24 s of the study) for each individual
rat. Image-derived input functions were corrected for partial-
volume effects (21). A recovery coefficient (RC) of 0.302 was
estimated by convolving the known small-animal PET point
spread function (full width at half maximum of 1.8 mm) with
a published mean artery diameter of 0.7 mm (22,23). This RC
was used to obtain the corrected activity concentration for each
rat, derived according to:
CTrue5CMeasured2 ð1 2 RCÞCBackground
where CTrue, CMeasured, and CBackgroundare the true activity con-
centration, the VOI-derived ventral caudal artery activity concen-
tration, and the VOI-derived background activity concentration,
respectively. The background VOI was drawn over normal muscle.
The image-derived partial-volume–corrected input function
was then fit to a heuristically derived sum of a series of functions,
each consisting of the product of a power and an exponential
function (Eq. 2). In this equation, Cb(t) is the activity con-
centration in the blood as a function of time, t. The remaining
terms represent fitting parameters without particular interpretation.
The number of functions summed together when performing the fit
was selected according to the Bayesian information criterion.
CbðtÞ 5 +
?t 2 t0
?Bi e2Ciðt 2 t0Þ=t:
Voxelized Tumor Kinetic Parameters. Whole-tumor VOIs were
delineated from a summed image of the last 20 min of the
dynamic PET data. Each frame of the dynamic data was smoothed
using a 1.2-mm 3-dimensional gaussian. Kinetic parameters, K1
(mL min21g21, equivalent to min21assuming unit density tis-
sue), k2(min21), and k3(min21) were estimated for each voxel
within the tumor VOI by fitting an irreversible 2-tissue-compart-
ment model using the Voxulus software package (Phillips Re-
search Europe). Two variants of this model were examined. In
one variant, k2was constrained to be equal to K1(assuming unit
density tissue), leaving just 2 free parameters (K1,k3) to be deter-
mined during the fit. In the other variant, the 3 rate constants were
unconstrained. These 2 variants will henceforth be referred to
simply as the constrained and unconstrained models, respectively.
The Voxulus software was previously validated by Wang et al.
(14). The model used here differs slightly from that used by Wang
18F-FLUOROMISONIDAZOLE PET DATA AND PO2 • Bartlett et al.3
jnm103523-sn n 8/29/12
et al. in that the current model does not include a contribution from
the vasculature, as our unpublished studies on these tumors show
this space to be negligible and leaving it in would likely have
reduced the robustness of the other parameter values because of
covariance among the parameters.
The irreversible 2-tissue-compartment model is described by
the following equations:
C1ðtÞ 5 K1e2ðk21 k3Þt5CbðtÞ
?1 2 e2ðk21 k3Þt?5CbðtÞ;
where Cb(t), C1(t), and C2(t) are time-dependent activity concen-
trations in the blood, reversible tissue compartment, and trapped
tissue compartment, respectively, and 5 denotes convolution. The
activity concentration in each voxel is the sum of C1(t) and C2(t).
The fitting procedure involved first segmenting the tumor into 4
clusters determined by a k-means clustering. For each of these
mean profiles, the kinetic rate constants were estimated using
a Levenberg–Marquardt least-squares optimization. These values,
in turn, were used as the initial parameter values in a second round
of model fitting, this time applied to individual voxel time profiles.
The resultant rate constants for each voxel were then arranged so
as to form a set of parametric images, K1, k2, and k3, for each
tumor. The steady-state uptake rate constant Ki5 K1k3/(k21 k3)
and T:P ratio parametric images were also calculated. Bayesian
information criterion values were calculated for each voxel when
fit by each of the 2 models.
Comparison of Parametric Images and PO2
Parametric image voxels along the PO2probe trajectories were
identified and their numeric values recorded. Because of the dif-
ference in distance between PO2probe measurements (0.5 mm)
and the PET voxel size (0.79 mm), the PO2data were resampled at
the voxel interval. This was done by averaging neighboring PO2
measures along the track, weighted by the small-animal PET
point-spread function. Corresponding pairs of image voxel and
resampled PO2values were tabulated.
A receiver-operating-characteristic analysis was applied to
these data using the Youden index (24) to compare the relative
utility of the various parameters (k3, Ki, T:P ratio) in distinguishing
hypoxic from nonhypoxic tissues, taking measured PO2as the
reference standard. Data points for which the PO2 was below
a given cut point and the parametric variable (e.g., k3) above
a given cut point were defined as true-positives, or TPs (i.e., pos-
itive by both measures). Conversely, points above the PO2cut point
and below the parameter cut point were defined as true-negatives,
or TNs. Similarly, false-positives (FPs) and false-negatives (FNs)
were defined as, respectively, both below or both above the re-
spective cut points. The Youden index, J, may be defined as the
maximal difference between TP and FP rates encountered along
the receiver-operating-characteristic curve. Thus, for any given
PO2cut point, the Youden index selects a corresponding “optimal”
parameter cut point. We plotted the Youden index as a function of
PO2cut point. The location of the maxima in this curve corre-
sponds to candidate PO2cut points where discrimination of hyp-
oxic and nonhypoxic voxels is maximized. When comparing the
relative utility of the parameters (k3, Ki, T:P ratio), we made use of
the SE in the Youden index, calculated as:
TP ? FN
ðTP 1 FNÞ31
FP ? TN
ðFP 1 TNÞ3
The significance of differences in Youden indices calculated at
a selected PO2cut point was assessed using a 2-tailed z test.
Fits of the unconstrained and constrained models had
similar rate constant estimates, and their respective Bayes-
ian information criterion values showed no clear model
preference, as illustrated by a representative voxel fit in
Figure 2. Also depicted in Figure 2 is the associated partial-
volume–corrected input function and fit for this same ani-
mal. K1 and k2 values calculated by the unconstrained
model were highly correlated (r2, ;0.6), with a slope close
to unity (0.9). Thus, constraining K1equal to k2would not
be expected to introduce significant bias. The main appar-
ent difference between the models was reduced noise in the
parametric images for the constrained case, presumably
a consequence of the reduced covariance.
A transaxial image, from a single animal, summed over
the last 40 min of image data is shown with overlays of the
fit from this same animal. (A) Data and fitted curve obtained using unconstrained model. (B) Data and fitted curve obtained using con-
strained model. (C) Partial-volume–corrected blood data and fit plotted on log linear scale such that peak of input function can be seen.
Plots of data and fitted curve from single voxel of individual rat and corresponding partial-volume–corrected input function and
4THE JOURNAL OF NUCLEAR MEDICINE • Vol. 53 • No. 10 • October 2012
jnm103523-sn n 8/29/12
tumor ROI and a PO2probe trajectory in
shown are the associated parametric images (T:P ratio,
K1, and k3) in gray scale along with color overlays describ-
ing the measured PO2levels. Of note in these images is the
general negative spatial correlation between the K1and k3
maps and the broad correspondence between low PO2and
high k3/low K1.
We also observed visually good spatial correlation
between pimonidazole immunofluorescence and the
digital autoradiography (though the degree varied among
tumors) and an apparent inverse spatial relationship be-
tween pimonidazole and Hoechst 33342 fluorescence
½Fig: 4? Fig. 4). These results are in broad concordance with the
relationships seen between parametric images of Figure 3.
characteristic analyses in terms of the Youden indices
plotted against the PO2cut point for the 3 potential hypoxia-
discriminating parameters k3, Ki, and T:P ratio, where larger
Youden index values indicate better discrimination. While
interpreting the results of this analysis, the reader should note
that random noise added to a given discriminating parameter
will tend to decrease its Youden index values. Thus, the
relatively noise-free T:P ratio outperforms the unconstrained
model’s Ki, which in turn outperforms the still noisier k3
parameter (Fig. 5A). However, for the parameters deter-
mined by the constrained model this order is reversed such
that now k3outperforms Kiand both do better than T:P ratio
(Fig. 5B) and better than their unconstrained counterparts.
Therefore, even though noisier than either Kior T:P ratio,
the constrained model k3outperforms them in demarcating
hypoxic from nonhypoxic tissue.
It is also clear from these plots that there is no obviously
optimal PO2cut point that maximizes the discrimination
between hypoxic and nonhypoxic tissue, although local
Figure 3. Also
Figure 5 shows the results of the receiver-operating-
maxima appear at about 0.3 mm Hg and 3.4 mm Hg. The
lack of a clear peak, we believe, is due to the high noise
(60.7 mm Hg) in the PO2measurements at these low levels
(25). The local maximum at 3.4 mm Hg corresponds to the
PO2value providing half-maximal radiobiologic hypoxic
protection (26) and is consistent with various in vitro mea-
surements of 2-nitroimidazole binding (9). Therefore, for
purposes of illustrative and statistical comparison, we se-
lected 3.4 mm Hg as a discriminatory cut point defining
tissue hypoxia. A statistical comparison of the Youden in-
dex values at this PO2threshold shows that k3of the con-
strained model outperforms T:P ratio (P 5 0.138) and is
significantly better than both k3and Kiof the unconstrained
model (P , 0.001 and P 5 0.011, respectively) but not
significantly better than constrained Ki.
Figure 6 shows scatterplots of k3versus K1and of each
potential image-derived hypoxia-discriminating parameter
(k3, Ki, T:P ratio) versus PO2for data pooled from all 11
tumors. All these plots show negative relationships. The
negative correlation between k3and K1provides some mea-
sure of general support for the observations of Figures 3
and 4 of an inverse relationship between hypoxia and per-
fusion. The parameter cut points (thresholds) shown as red
lines in Figures 6B–6D are the values chosen by the Youden
indices for a PO2cut point of 3.4 mm Hg defining tissue
hypoxia. Also shown in Figure 6D is the T:P threshold of
1.4 (blue) recommended by Koh et al. (7) for discriminating
In this work, we directly compared the PO2levels mea-
sured with an accepted reference standard device (OxyLite
probe) against voxelwise measures of18F-fluoromisonidazole
last 40 min, with ROI used to segment data
for kinetic analysis overlaid in red and PO2
probe trajectory overlaid in blue. Images on
right are voxelwise kinetic maps for same
slice and ROI shown on left. Overlaid onto
each of these maps are corresponding PO2
concentration measurements made at each
of those corresponding voxel locations.
On left is image averaged over
18F-FLUOROMISONIDAZOLE PET DATA AND PO2 • Bartlett et al.5
jnm103523-sn n 8/29/12
uptake. Although similar correlative studies have been un-
dertaken by others (10,27), to our knowledge ours is the
first comparison using pharmacokinetic model–derived
metrics. This is an important advance over previous work
because the pharmacokinetic model allowed us to tease out
a parameter, k3, that is directly dependent on the oxygen
level, effectively removing the flow- and transport-related
confounds present in other uptake metrics (e.g., T:P ratio,
Ki, and standardized uptake value).
The results of the analysis suggest that the k3parameter
provides the best discrimination of tissue hypoxia in terms
of producing the greatest Youden index from receiver-op-
erating-characteristic analysis. However, this was true only
when the kinetic model was constrained by setting K15 k2,
effectively removing a degree of freedom from the data-
fitting process. In biologic terms, this constraint is equivalent
to assuming that the transmembrane transport of18F-fluoro-
misonidazole is a passive diffusion process and that, in the
absence of bioreductive metabolism, the steady-state intra-
cellular and extracellular concentrations would be equal.
This assumption is consistent with the typical observation
of T:P ratios for18F-fluoromisonidazole of approximately 1
in most normal tissues, seen both clinically and in our
Although our analyses suggest that k3is the optimal
discriminator of tissue hypoxia (when noise in its measure
is sufficiently limited), the correspondence was far from per-
fect. Specifically, there were low k3values in tissue voxels
for which correspondingly low PO2 measurements were
made. One possible explanation is that k3can be high only
in viable hypoxic tissue and not in regions of necrosis (28).
This explanation is plausible, as bioreduction of the parent
18F-fluoromisonidazole molecule requires a certain level of
biochemical functionality (nitroreductase activity) more
likely to be found in intact cells and is consistent with the
enhanced concentration of pimonidazole adducts typically
seen in perinecrotic regions of tumor tissue sections but
not in regions of frank necrosis. Another possible explana-
tion for imperfect correspondence between k3and PO2is
spatial misregistration of probe- and image-based values
due to deformation of the probe trajectory (29). The OxyLite
probe is a thin (200-mm diameter), flexible optic fiber and
may deviate from the plotted trajectory given an alternative
“path of least resistance.”
The results of this study, exemplified by both Figure 3
and Figure 6, show that the k3parametric image can pro-
vide a sharp, almost binary, representation of tumor hyp-
oxia. This representation is consistent with the typically
rapid increase in18F-fluoromisonidazole binding rate over
a limited PO2range seen with cells in culture. In contrast,
other metrics such as Kior T:P ratio show a more subtle
gradation, partly related to the confounding influences of
tracer delivery or transport.
tumor from each of 4 rats from which histology was obtained. Top
row (blue-stained sections) shows histologic images of Hoescht
33342. Middle row (green-stained sections) shows histologic
images of pimonidazole. Bottom row shows digital autoradiographs
Registered histology images from single slice of rat
strained model (A) and constrained model (B).
Plots of Youden index vs. PO2 threshold: uncon-
6THE JOURNAL OF NUCLEAR MEDICINE • Vol. 53 • No. 10 • October 2012
jnm103523-sn n 8/29/12
In a commonly used approach, a T:P ratio above 1.4 later
than 2 h after injection is taken to indicate hypoxia, and in
several studies this criterion was used to conclude that some
hypoxic tumors were rendered normoxic after radiotherapy
(30–33). However, meaningful correlations with tumor pro-
gression or other measures of outcome have not been forth-
coming. This, combined with a lack of correspondence
18F-fluoromisonidazole uptake and direct mea-
sures of PO2, has cast some doubt on the utility of18F-fluoro-
misonidazole as a hypoxia marker (10,34–36). Previous
studies by our group have shown that interventions that
affect vascular functionality can lead to a dissociation be-
tween tumor hypoxia and18F-fluoromisonidazole net up-
take (37), indicative of a 2-fold problem associated with
using a static net uptake metric. First, the slow rate of
18F-fluoromisonidazole blood clearance and its high distri-
bution volume leads to a high background level of18F-
fluoromisonidazole in normoxic tissues. Second, reduced
or compromised perfusion in hypoxic regions can result
in18F-fluoromisonidazole uptakes that are not significantly
above (and may in fact be below) normoxic tissue levels.
Consistent with this view is the suggestion that it would be
better to measure18F-fluoromisonidazole uptake at longer
times after injection (4 or more hours) (38). Such longer
times would allow greater diffusion into poorly perfused
regions, greater time for irreversible binding, and greater
clearance from normoxic tissue leading to improved hyp-
oxic contrast. Conversely, waiting an additional18F half-
life would mean reduced counts and greater noise in the
images, thus increasing the threshold required to define
a significant difference above background.
The alternative approach of kinetic modeling of18F-fluoro-
misonidazole uptake, first suggested by Thorwarth (13),
provides a means of distinguishing regions where18F-fluoro-
misonidazole is accumulating (hypoxia) from those where
it is destined to clear (normoxia). The results of the current
study provide preliminary proof of the concept that this can
improve the correspondence between hypoxia image and
As is the case with all pharmacokinetic compartmental
modeling, it is important to not mix together tissues having
differing kinetics when applying the model. In the case of
18F-fluoromisonidazole, mixing tissues having similar perfu-
sion and transport properties but differing in k3would result
in an averaged k3value. Although it is perhaps reasonable to
assume perfusion and transport are roughly homogeneous
within a voxel-sized region, this is almost certainly not the
case for many whole-tumor regions. Therefore, it is impor-
tant that pharmacokinetic modeling be applied on a voxel-
wise basis or, at minimum, that subregions be defined on the
basis of similarity in the shape of tissue uptake curves.
Voxelwise estimates of k3are better at identifying low
tissue oxygen levels than are other measures of18F-fluoro-
misonidazole uptake. An effective means of controlling
noise in k3, without introducing significant bias, is to con-
strain K1to be equal to k2during the fitting process.
The costs of publication of this article were defrayed in
part by the payment of page charges. Therefore, and solely
ters estimated from constrained model (K15
k2) vs. corresponding PO2 measure. Red
lines define optimal cut points selected by
Youden indices. (A) k3vs. K1. (B) k3vs. PO2.
(C) K1vs. PO2. (D) T:P ratio vs. PO2; blue line
shows standard 1.4 threshold commonly
applied in assessment of hypoxia.
Scatterplots of kinetic parame-
18F-FLUOROMISONIDAZOLE PET DATA AND PO2 • Bartlett et al.7
jnm103523-sn n 8/29/12
to indicate this fact, this article is hereby marked “adver-
tisement” in accordance with 18 USC section 1734.
We gratefully acknowledge the contributions of Valerie
Longo and the technical services provided by the Memorial
Sloan-Kettering Cancer Center Small-Animal Imaging Core
Facility. This work was supported in part by NIH grant
P01 CA115675 on Hypoxia Imaging, NIH Small-Animal
Imaging Research Program (SAIRP) grant R24 CA83084,
and NIH Center grant P30 CA08748. Philips Medical
provided $10,000 for the cost of the radiotracer. No other
potential conflict of interest relevant to this article was
1. Brizel DM, Scully SP, Harrelson JM, et al. Tumor oxygenation predicts for the
likelihood of distant metastases in human soft tissue sarcoma. Cancer Res.
2. Hockel M, Schlenger K, Aral B, Mitze M, Schaffer U, Vaupel P. Association
between tumor hypoxia and malignant progression in advanced cancer of the
uterine cervix. Cancer Res. 1996;56:4509–4515.
3. Mottram JC. A factor of importance in the radio-sensitivity of tumours. Br J
4. Teicher BA, Holden SA, al-Achi A, Herman TS. Classification of antineoplastic
treatments by their differential toxicity toward putative oxygenated and hypoxic
tumor subpopulations in vivo in the FSaIIC murine fibrosarcoma. Cancer Res.
5. Nozue M, Lee I, Yuan F, et al. Interlaboratory variation in oxygen tension
measurement by Eppendorf “Histograph” and comparison with hypoxic marker.
J Surg Oncol. 1997;66:30–38.
6. Olive PL, Banath JP, Aquino-Parsons C. Measuring hypoxia in solid tumours: is
there a gold standard? Acta Oncol. 2001;40:917–923.
7. Koh WJ, Rasey JS, Evans ML, et al. Imaging of hypoxia in human tumors with
[F-18]fluoromisonidazole. Int J Radiat Oncol Biol Phys. 1992;22:199–212.
8. Valk PE, Mathis CA, Prados MD, Gilbert JC, Budinger TF. Hypoxia in human
gliomas: demonstration by PET with fluorine-18-fluoromisonidazole. J Nucl
9. Chapman JD, Lee J, Meeker BE. Adduct formation by 2-nitromidazole drugs in
mammalian cells: optimization of markers for tissue oxygenation. In: Adams
GE, Breccia A, Fielden EM, Wardman P, eds. Selective Activation of Drugs by
Redox Processes. New York, NY: Springer; 1991:313–323.
10. Mortensen LS, Buus S, Nordsmark M, et al. Identifying hypoxia in human
tumors: a correlation study between18F-FMISO PET and the Eppendorf oxy-
gen-sensitive electrode. Acta Oncol. 2010;49:934–940.
11. Casciari JJ, Graham MM, Rasey JS. A modeling approach for quantifying tumor
hypoxia with [F-18]fluoromisonidazole PET time-activity data. Med Phys.
12. Kelly CJ, Brady M. A model to simulate tumour oxygenation and dynamic18F-
Fmiso PET data. Phys Med Biol. 2006;51:5859–5873.
13. Thorwarth D, Eschmann SM, Paulsen F, Alber M. A kinetic model for dynamic
[18F]-Fmiso PET data to analyse tumour hypoxia. Phys Med Biol. 2005;50:2209–
14. Wang W, Georgi JC, Nehmeh SA, et al. Evaluation of a compartmental model for
estimating tumor hypoxia via FMISO dynamic PET imaging. Phys Med Biol.
15. Zanzonico P, Campa J, Polycarpe-Holman D, et al. Animal-specific positioning
molds for registration of repeat imaging studies: comparative microPET imaging
of F18-labeled fluoro-deoxyglucose and fluoro-misonidazole in rodent tumors.
Nucl Med Biol. 2006;33:65–70.
16. Grierson JR, Link JM, Mathis CA, Rasey JS, Krohn KA. A radiosynthesis of
fluorine-18 fluoromisonidazole. J Nucl Med. 1989;30:343–350.
17. Kazanzides P, Chang J, Iordachita I, Li J, Ling CC, Fichtinger G. Design and
validation of an image-guided robot for small animal research. Med Image
Comput Comput Assist Interv. 2006;9:50–57.
18. Kazanzides P, Chang J, Iordachita I, Li J, Ling CC, Fichtinger G. Development
of an image-guided robot for small animal research. Comput Aided Surg.
19. Chang J, Wen B, Kazanzides P, et al. A robotic system for18F-FMISO PET-
guided intratumoral pO2 measurements. Med Phys. 2009;36:5301–5309.
20. Carlin S, Pugachev A, Sun X, et al. In vivo characterization of a reporter gene
system for imaging hypoxia-induced gene expression. Nucl Med Biol.
21. Kessler RM, Ellis JR Jr, Eden M. Analysis of emission tomographic scan data:
limitations imposed by resolution and background. J Comput Assist Tomogr.
22. Bao JY. Rat tail: a useful model for microvascular training. Microsurgery.
23. Blain B, Zhang F, Jones M, et al. Vascular grafts in the rat model: an anatomic
study. Microsurgery. 2001;21:80–83.
24. Youden WJ. Index for rating diagnostic tests. Cancer. 1950;3:32–35.
25. Griffiths JR, Robinson SP. The OxyLite: a fibre-optic oxygen sensor. Br J Radiol.
26. Hall EJ, Giaccia AJ. Radiobiology for the Radiologist. 6th ed. Philadelphia, PA:
Lippincott Williams &Wilkins; 2006:85–103.
27. Sørensen M, Horsman MR, Cumming P, Munk OL, Keiding S. Effect of intra-
tumoral heterogeneity in oxygenation status on FMISO PET, autoradiography,
and electrode Po2 measurements in murine tumors. Int J Radiat Oncol Biol Phys.
28. Krohn KA, Link JM, Mason RP. Molecular imaging of hypoxia. J Nucl Med.
29. Elas M, Ahn KH, Parasca A, et al. Electron paramagnetic resonance oxygen
images correlate spatially and quantitatively with Oxylite oxygen measurements.
Clin Cancer Res. 2006;12:4209–4217.
30. Hicks RJ, Rischin D, Fisher R, Binns D, Scott AM, Peters LJ. Utility of FMISO
PET in advanced head and neck cancer treated with chemoradiation incorporat-
ing a hypoxia-targeting chemotherapy agent. Eur J Nucl Med Mol Imaging.
31. Lin Z, Mechalakos J, Nehmeh S, et al. The influence of changes in tumor
hypoxia on dose-painting treatment plans based on F-18-FMISO positron emis-
sion tomography. Int J Radiat Oncol Biol Phys. 2008;70:1219–1228.
32. Thorwarth D, Eschmann SM, Scheiderbauer J, Paulsen F, Alber M. Kinetic
analysis of dynamic18F-fluoromisonidazole PET correlates with radiation treat-
ment outcome in head-and-neck cancer. BMC Cancer. 2005;5:152–161.
33. Thorwarth D, Eschmann S-M, Holzner F, Paulsen F, Alber M. Combined uptake
of18F FDG and18F FMISO correlates with radiation therapy outcome in head-
and-neck cancer patients. Radiother Oncol. 2006;80:151–156.
34. Allemann K, Wyss MT, Wergin M, et al. Measurements of hypoxia (18F-FMISO,
18F-EF5) with positron emission tomography (PET) and perfusion using PET
(15O-H2O) and power Doppler ultrasonography in feline fibrosarcomas*. Vet
Comp Oncol. 2005;3:211–221.
35. Rajendran JG, Wilson DC, Conrad EU, et al. F-18 FMISO and F-18 FDG PET
imaging in soft tissue sarcomas: correlation of hypoxia, metabolism and VEGF
expression. Eur J Nucl Med Mol Imaging. 2003;30:695–704.
36. Roels S, Slagmolen P, Nuyts J, et al. Biological image-guided radiotherapy in
rectal cancer: is there a role for FMISO or FLT, next to FDG? Acta Oncol.
37. Oehler C, O’Donoghue JA, Russell J, et al.18F-fluromisonidazole PET imaging
as a biomarker for the response to 5,6-dimethylxanthenone-4-acetic acid in co-
lorectal xenograft tumors. J Nucl Med. 2011;52:437–444.
38. Abolmaali N, Haase R, Koch A, et al. Two or four hour18F FMISO-PET in
HNSCC. When is the contrast best? Nuklearmedizin. 2011;50:22–27.
8THE JOURNAL OF NUCLEAR MEDICINE • Vol. 53 • No. 10 • October 2012
jnm103523-sn n 8/29/12