Comparison of Dynamic Contrast Enhanced MRI and Quantitative SPECT in a Rat Glioma Model

Abstract

Pharmacokinetic modeling of dynamic contrast enhanced (DCE)-MRI data provides measures of the extracellular volume fraction (v_e) and the volume transfer constant (K^trans) in a given tissue. These parameter estimates may be biased, however, by confounding issues such as contrast agent and tissue water dynamics, or assumptions of vascularization and perfusion made by the commonly used model. In contrast to MRI, radiotracer imaging with SPECT is insensitive to water dynamics. A quantitative dual-isotope SPECT technique was developed to obtain an estimate of v_e in a rat glioma model for comparison to the corresponding estimates obtained using DCE-MRI with a vascular input function (VIF) and reference region model (RR). Both DCE-MRI methods produced consistently larger estimates of v_e in comparison to the SPECT estimates, and several experimental sources were postulated to contribute to these differences.

Introduction

Dynamic contrast enhanced magnetic resonance imaging (DCE-MRI) is a technique often employed to characterize the microvascular environment in tumor tissues (1,2). This quantitative imaging technique provides a measure of the volume transfer constant (K^trans) and the volume fraction of extracellular-extravascular tissue space (v_e) via tracer kinetic modeling of an MRI contrast agent (CA), typically Gd-DTPA. The most commonly used DCE metric is K^trans, which reports on perfusion, but accurate measure of v_e may also be of clinical importance in assessing tumor type and response to treatment. For example, it has been shown that the relative size of v_e can help distinguish between gliomas, meningiomas, and metastases (1) in brain tumors. The pharmacokinetic modeling necessary for DCE-MRI reduces a complex tissue micro-environment into a relatively simple compartmental model (3), which has known limitations (4–7). Also, parameter estimation is sensitive to the sometimes difficult to acquire vascular input function (VIF) (8–10) or the more easily acquired reference region (RR) signal (11–12). Given the array of potential sources of error in DCE-MRI, the interpretation of fitted model parameters as physical characteristics of tissue remains a question.

One of the DCE-MRI model parameters, v_e, can potentially be measured directly with single photon emission computed tomography (SPECT). To this end, a dual-isotope SPECT approach was developed and implemented on rats to obtain estimates of v_e that did not require dynamic tracer modeling, assumptions about water dynamics, or a VIF or RR signals. In a orthotopic rat glioma tumor model, these SPECT estimates of v_e were compared with estimates derived from DCE-MRI using both the VIF and RR approaches, with the objective of elucidating some of the sources of error in the DCE-MRI.

Methods

Approximately two weeks prior to imaging, female Sprague-Dawley rats (n=8, 234–270g) were anesthetized, given analgesic, and inoculated with C6 glioma cells (1×10⁵ cells) via stereotaxic injection. Two jugular catheters were placed in each rat, up to 24 hours before imaging. One catheter was used for CA injection while the other was used for blood sampling. Each rat was anesthetized using ≈ 2% isoflurane in oxygen for all surgical and imaging procedures. For MR imaging procedures body temperature was maintained near 37° C by a flow of warm air directed over the animal. Respiration was monitored using a pneumatic pillow placed on the side of the animal near the abdomen. Rats were imaged using a DCE-MRI protocol and were allowed to recover after the scan. Dual-isotope SPECT measurements were performed the following day, within 24 hours of the completion of the DCE-MRI measurements. Prior to SPECT, a bi-lateral nephrectomy was performed to prevent washout of the radiotracer, as was required for analysis of the SPECT data. The animals were anesthetized before the nephrectomy and remained anesthetized throughout SPECT imaging and euthanasia by overdose of isoflurane. All procedures were approved by Vanderbilt University’s institutional animal care and use committee.

DCE-MRI

MRI was performed on a 9.4T horizontal-bore magnet (Agilent, Santa Clara, CA). The animal’s head was positioned in a 38 mm diameter Litz quadrature coil (Doty Scientific, Columbia, SC) and was secured by a bite bar apparatus that also served to deliver the anesthetic gas. Scout images were collected to identify a 2 mm thick imaging slice through the center of the brain tumor and a second slice in the neck containing the major vessels feeding the brain.

Prior to DCE-MRI, an inversion-recovery snapshot experiment—TR = 12 s, TI (inversion time) = 0.250-11 s, 2 averaged excitations, 128 × 128 samples, 32 mm × 32 mm FOV—was used to generate data to produce a pre-contrast R₁ (longitudinal relaxation rate = 1/T₁) map, R₁₀. A three parameter fit to these data was used to estimate R₁ in the presence of imperfect inversion. The DCE-MRI protocol employed a standard spoiled gradient-echo (SPGE) sequence—TR = 10 ms, TE = 2.1 ms, flip angle (α) = 15°, 96×96 samples, and 2 averaged excitations—over the prescribed slices. Approximately 40 images were collected before a 200 µl bolus of Gd-DTPA (0.05 mmol/kg) was delivered over 5 s via the jugular catheter. After injection of the contrast agent, dynamic images were collected for 20 minutes (~2s per image) and were reconstructed to 128 × 128 samples to match the spatial resolution of the R₁₀ map. Following the completion of the DCE-MRI experiment, a 3D SPGE—TR = 20 ms, TE = 2.1 ms, α = 20°, 128 × 128 × 64 samples, 32 × 32 × 32 mm³ slab—was acquired to provide a reference for image registration between MRI and SPECT images.

Bi-lateral Nephrectomy

Under anesthesia, two parallel incisions, one on each side, were made on the lower back of the rat. Using a swab, the kidneys were exposed and 3-0 silk sutures were slipped around each kidney and tightened around the renal pedicles, preventing clearance through the kidney. The kidneys were returned to their original position and the incisions were sutured shut. The total time for the procedure was approximately 20 minutes.

SPECT

Radionuclide imaging was performed on a NanoSPECT/CT (Bioscan, Washington DC, USA) with image reconstruction carried out using HiSPECT in the corresponding InVivoScope software. Under anesthesia, an extracellular radiotracer, ¹¹¹In-DTPA (~19 MBq), was injected first, followed by a saline flush, and allowed to equilibrate for approximately 30 minutes (longer than the known wash-in time for Gd-DTPA). An intravascular tracer, ^99mTc-RBCs (Ultratag®) (~130 MBq), was then injected. The higher activity of ^99mTc compared with ¹¹¹In was used to minimize scatter of ¹¹¹In emitted photons into the ^99mTc energy window. After injection of the second tracer, images were collected (48 projections (helical SPECT), 40 s per projection, reconstructed volume = 67 × 67 × 35 mm³). Three energy windows (width=18%) were set about the ^99mTc and ¹¹¹In photopeaks (140 keV and 171/245 keV, respectively), and projection data were recorded independently for each energy window using a 9-pinhole aperture collimator (pinhole diameter = 2.0 mm) on each of four detector heads. After collecting the brain images a blood sample (500 µL) was obtained from the unused jugular catheter and imaged with the previously described dual-isotope protocol. An X-ray CT scan was also collected for registration purposes (imaging dimensions same as SPECT). All SPECT images were reconstructed with 0.4 × 0.4 × 0.4 mm³ voxels to match the CT image.

To test the dual-isotope SPECT method, ^99mTc and ¹¹¹In solutions were prepared across a range of concentrations, and activities of each were measured in a gamma counter. These solutions were then used to make five mixed-isotope imaging phantoms (500 µL) with activity ratios (i.e. A_99mTc / A_111In) similar to those expected in the blood samples. The dual-isotope SPECT protocol was used to measure activity of each isotope in each imaging phantom, and the results were compared to the gamma counter derived measures from single isotope samples.

Data Analysis

SPECT-MRI Registration

All MR and SPECT/CT imaging data were imported into MATLAB (Mathworks, Natick, MA) for processing. Prior to any quantitative analysis, the CT data were registered to the 3D SPGE data set using a rigid registration algorithm. All SPECT data were subsequently registered to the MR-space as the SPECT images were implicitly co-registered to the CT scan. This process allowed co-registration of the single slice dynamic data (from the brain tumor) to the SPECT images.

DCE-MRI

The R₁₀ map was used to identify a tumor region of interest (ROI) within which an R₁ tissue time course, R_1t(t), was determined for each voxel using:

$$R_1(t)=\frac{1}{\mathit{\text{TR}}}\text{ln}\left[\frac{S_0·\text{sin }\alpha -S(t)·\text{cos }\alpha }{S_0·\text{sin }\alpha -S(t)}\right],$$ [1]

where S₀ was defined as

$$S_0=S_{-}·\left[\frac{1-e^{-\mathit{\text{TR}}·R_{10}}·\text{cos }\alpha }{(1-e^{-\mathit{\text{TR}}·R_{10}})·\text{sin }\alpha }\right],$$ [2]

S₋ was the measured SPGE signal intensity prior to contrast agent injection, and S(t) was the measured signal intensity at a time t after contrast injection. Similarly, from a 5–10 voxel ROI in the linguofacial artery in the neck an image derived VIF was defined as R_1b(t). The concentration of CA in the plasma space (C_p) was then determined using an assumption of fast water exchange across red blood cell membranes,

$$C_p(t)=(R_{1b}(t)-R_{10b})/r_1·(1-h)$$ [3]

where the hematocrit (h) was 0.46, r₁ (the relaxivity of Gd-DTPA at 9.4T) was assumed to be 3.8 mM⁻¹s⁻¹ (13,14), and the pre-contrast R₁ of blood was assumed to be R_10b = 0.435 s⁻¹ (15,16).

K^trans, v_e, and v_p (plasma volume fraction) were then estimated by fitting C_p(t) and R_1t(t), with Eqs. [4] and [5] (5):

$$R_{1t}(t)=R_{10t}+r_1·v_e·C_t(t)$$ [4]

and

$$C_t(t)=K^{\mathit{\text{trans}}}·{\displaystyle \underset{0}{\overset{t}{\int }}}{C}_p(T)·\text{exp}(-(K^{\mathit{\text{trans}}}/v_e)·(t-T))\mathit{\text{dt}}+v_p{C}_p(t).$$ [5]

In these equations, C_t(t) is the concentration of the CA in tumor tissue as represented by the extended Tofts model (3). Subsequently, in this paper, K^trans and v_e estimates from this VIF-based analysis are defined as $K_{\mathit{\text{VIF}}}^{\mathit{\text{trans}}}$ and v_e,VIF, respectively. Non-linear fitting, here and below, used the default settings of the lsqcurvefit function in MATLAB R2011a (Natick, MA) with fitted parameters (K^trans, v_e and v_p) constrained to non-negative values.

In addition to analysis using the VIF, the dynamic data were also analyzed using a RR model (11). In the RR analysis, a reference tissue ROI was selected in the temporal muscle surrounding the skull. Two parameters, K^trans,t and v_e,t, were then estimated by fitting the R₁ dynamic time course from each tumor voxel and the RR with

$$R_{1,t}(t)=R\times (R_{1,\mathit{\text{ref}}}(t)-R_{10,\mathit{\text{ref}}})+R[(K^{\mathit{\text{trans}},\mathit{\text{ref}}}/v_{e,\mathit{\text{ref}}})-(K^{\mathit{\text{trans}},t}/v_{e,t})]\times {\displaystyle \underset{0}{\overset{t}{\int }}}(R_{1,\mathit{\text{ref}}}(T)-R_{10,\mathit{\text{ref}}})·\text{exp}((-K^{\mathit{\text{trans}},t}/v_{e,t})\times (t-T))\mathit{\text{ dt}}+R_{10,t}$$ [6]

where subscripts t and ref indicate tumor and reference regions, R=K^trans,t/K^trans,ref, and K^trans,ref = 0.05 min⁻¹ and v_e,ref = 0.11 were set from literature values (8,11). Subsequently, in this paper, K^trans,t and v_e,t estimates from this RR-based analysis are defined as $K_{\mathit{\text{RR}}}^{\mathit{\text{trans}}}$ and v_e,RR, respectively.

SPECT

The nephrectomy procedure allowed the extracellular radiotracer to reach equilibrium in the vascular and the extracellular-extravascular space such that

$$C_{111\mathit{\text{In}},\mathit{\text{blood}}}\approx C_{111\mathit{\text{In}},\mathit{\text{ECEV}}}.$$ [7]

The concentrations of tracers in the blood sample were determined from the SPECT image of the blood sample as

$$C_{99\mathit{\text{mTc}},\mathit{\text{blood}}}=A_{99\mathit{\text{mTc}},\mathit{\text{blood}}}/V_{\mathit{\text{blood}}},\text{ and}$$ [8]

$$C_{111\mathit{\text{In}},\mathit{\text{blood}}}=A_{111\mathit{\text{In}},\mathit{\text{blood}}}/V_{\mathit{\text{blood}}},$$ [9]

where V_blood was the known volume of the blood sample and A was the measured activity of the specified radiotracer (in MBq) from the entire sample in the SPECT image. Similarly, the activity in the tumor tissue was described by

$$A_{99\mathit{\text{mTc}},\mathit{\text{tumor}}}=C_{99\mathit{\text{mTc}},\mathit{\text{blood}}}·V_{\mathit{\text{IV}}},\text{ and}$$ [10]

$$A_{111\mathit{\text{In}},\mathit{\text{tumor}}}=C_{111\mathit{\text{In}},\mathit{\text{blood}}}·V_{\mathit{\text{IV}}}+C_{111\mathit{\text{In}},\mathit{\text{ECEV}}}·V_{\mathit{\text{ECEV}}},$$ [11]

where V_IV and V_ECEV are the volumes of the intravascular space and extracellular extravascular space, respectively, in the tumor tissue. Given the measured activities in a given tumor voxel and tracer concentrations in the blood from Eqs. [8] and [9], Eqs. [7], [10] and [11] were solved to calculate V_IV and V_ECEV. From here, v_e was estimated voxel by voxel as

$$v_e=V_{\mathit{\text{ECEV}}}/V_{\mathit{\text{tissue}}},$$ [12]

where V_tissue was the known SPECT voxel volume. Subsequently, in this paper, v_e estimates from the SPECT measurements are defined as v_e,SP.

Results

Figure 1 shows the results of the phantom SPECT measurements, demonstrating good accuracy of the dual-isotope SPECT protocol across a wide range of activities. Figure 2a shows an example of the MRI slice used to visualize and select an ROI for the VIF; a typical VIF is shown in Fig. 2b. Figure 3 shows a typical tumor ROI as manually selected on a pre-contrast T₁ map (Fig. 3a) and a corresponding example voxel dynamic R₁ time course along with the fit to the RR model (Fig. 3b). Figure 3a also shows a reference region (skeletal muscle) ROI and the inset in Fig. 3b shows its R₁ time course, which was used for the DCE-MRI RR model fitting. All DCE-MRI fitting showed qualitatively good fits and quantitative estimates of parameter uncertainties that were much less than the spatial and inter-animal variations in model parameters. Figure 4 shows results of the parametric maps from the RR model fitting and the dual-isotope SPECT analysis from two example rats (left and right). On the left, the tumor is relatively uniformly perfused (K^trans, Fig 4a) with moderate estimates of v_e (DCE-MRI, Fig 4c; SPECT, Fig 4e). In contrast, on the right, the tumor exhibits more heterogeneous maps of K^trans (4b), v_e,RR (4d), and v_e,SP (4f). In this case, although the colorbar indicates a maximum v_e,RR = 1, in the tumor core some voxels exhibited v_e > 1 (sometimes >> 1), indicating a breakdown of the model for accurately describing the DCE data. In order to avoid extreme outliers, only voxels exhibiting v_e < 2 and K^trans < 0.5 min⁻¹ were included in subsequent analysis. This restriction resulted in discarding 2 - 30% (range, across animals) of tumor voxels from VIF analysis and 0.5 – 8% of tumor voxels from RR analysis.

Fig. 1.

Mixed isotope phantom validation for activity measurement with the dual-isotope protocol. a) ^99mTc activity measures per volume from the SPECT image and gamma counter. b) ¹¹¹In activity measures per volume from the SPECT image and gamma counter. Error bars were determined from the standard deviation across voxels in the SPECT images, and the uncertainties in the gamma counter derived measures were smaller than the marker size. Linear regression analysis also shown.

Fig. 2.

a) Example of ROI placement for the VIF in the linguofacial artery. ROIs contained between 5 – 10 voxels. b) Example VIF time-course (in [CA]) recorded over an approximately 20 minute time period. Note the rapid rise and washout of the CA.

Fig. 3.

a) An example T₁₀ map showing the location of the manually defined brain tumor ROI and reference tissue ROI in the temporal muscle. Note the elevated T₁ times in the tumor region. b) Corresponding R₁ time course including RR model fit. The RR model shows a good fit to the dynamic data. The inset displays the dynamic time course of the reference tissue used for RR modeling.

Fig. 4.

Parametric maps from RR model fitting and co-registered SPECT/MR images for two different rats (one per column). (a,b) $K_{\mathit{\text{RR}}}^{\mathit{\text{trans}}}$ maps, (c,d) v_e,RR maps, and (e,f) v_e,SP maps. Note the pseudocolor scale for v_e is maximum at 1, apparent in the central region of the tumor in frame d). The SPECT maps in e) and f) display the tumor ROI as well as a ROI in the muscle. The apparent discrepancy between spatial location of of high v_e,RR and high v_e,SP might reflect limitation in the MRI-SPECT registration.

Figure 5 shows the tumor ROI mean ± standard deviation (SD) of estimates of v_e (top) and K^trans (bottom) from both DCE-MRI methods and SPECT (for v_e only) for all eight animals. In general, the v_e estimates from the VIF analysis exceeded those of the RR analysis, and both DCE-MRI analyses of v_e were higher than those found in the dual-isotope SPECT. Computed voxel-by-voxel from the tumor ROIs in all eight animals, v_e,_VIF - v_e,SP = 0.57 (±0.01 standard error (SE)), and the difference was significant (t-test, p<0.0001) in all eight animals. Similarly, across all tumor ROI voxels, v_e,RR - v_e,SP =0.14 (±0.005 SE), and the difference was significant (t-test, p<0.01) in 7 of 8 animals. Comparing the DCE-MRI analysis techniques, v_e,VIF - v_e,RR =0.43 (±0.008 SE), and again, the difference was significant (t-test, p<0.0001) in all eight animals.

Fig. 5.

a) Comparison of v_e for VIF, RR, and SPECT analysis for each animal studied. b) Comparison of K^trans for VIF and RR analysis. Solid bars represent mean values while error bars represent the standard deviations. Note that error bars are somewhat underestimated because outlier voxels exhibiting v_e > 2 or K^trans > 0.5 min⁻¹ were not included in analysis.

Although there was no comparable measure for K^trans from the SPECT data, it is worth noting that $K_{\mathit{\text{VIF}}}^{\mathit{\text{trans}}}>K_{\mathit{\text{RR}}}^{\mathit{\text{trans}}}$ across all eight animals (Fig. 5b). Also, although not shown in Fig. 5, a similar overestimation of blood volume fraction by VIF analysis of DCE-MRI (range across animals of mean = 0.04–0.18 and of SD = 0.03–0.15) compared with SPECT measures (range across animals of mean = 0.019–0.036 and of SD = 0.005–0.011) was found. Note, however, that constraining v_p to a priori values similar to those found by SPECT had little effect on the fitted values of v_e by DCE-MRI.

Discussion

To the authors’ knowledge, this is the first account of an in vivo radionuclide measure of the extracellular-extravascular volume fraction; although previous such measures have been made ex vivo via whole tissue gamma counting or autoradiography (17,18). While the radionuclide imaging used in this study is not practical for routine pre-clinical use because it involves surgical intervention to allow tracer equilibration, it is insensitive to water dynamics and does not require a vascular input function. Regarding surgical effects on parameter estimates, two previous studies showed no significant changes in plasma space or tissue water content of normal rat brain over the time, post-nephrectomy, during which the current measurements were recorded (19,20). Phantom tests (see Fig. 1) demonstrated the accuracy of this dual-energy SPECT method for measuring concentrations of ^99mTc-RBCs and ¹¹¹In-DTPA, and the in vivo results demonstrate a reasonable correspondence of both v_e (0.10 ± 0.04) from skeletal muscle (17,18, 21) and blood volume (0.03 ± 0.01) in tumors to literature values (22). Thus, although no independent validation of the in vivo SPECT measurements from tumor have been made, we interpret the observed differences in SPECT and MRI measures of v_e (Fig. 5) to predominantly reflect an overestimation of v_e from DCE-MRI.

The explanation for why DCE-MRI overestimated v_e is not entirely clear, but is apparent that DCE analysis using a VIF resulted in a much greater overestimation than did analysis using the reference region. Averaging of the vascular signal over space with non-vascular signal or over time during the peak of the VIF will result in a diminished VIF and, in turn, an increased estimate of v_e (8, 23). However, in this study, the VIF was acquired at relatively high spatial and temporal resolutions, so these factors are unlikely to explain the differences between v_e,VIF and v_e,_RR. The VIF amplitude may also be reduced by T₂^* effects, but again, in this study, with TE = 2.1 ms and peak C_p ~0.2 mM, a relaxivity of ~10 s⁻¹/mM results in < 1 % signal reduction. A third potential source of error in the VIF is due to the inflow effect on the T₁-weighting image contrast, which has been studied in detail recently (24, 25) and an earlier study showed roughly the same scale of v_e overestimation when inflow effects were not avoided (26).

Inflowing blood which did not experience the previous RF excitation pulses is not T₁-weighted and results in an overestimation of S₀ (Eq. [2]) and an underestimation of the change in S(t) (Eq [1]). Both of these factors diminish the VIF and, in turn, result in an overestimation of v_e. In the context of the present study, an arterial blood velocity of 5 cm/s (27) dictates that ≈¼ of the blood in a 2 mm imaging slice is replaced with inflowing blood every TR (10 ms). Crudely then, one can consider the blood signal to be 75% from magnetization at steady state and 25% from magnetization at equilibrium. Then, given R_10b = 0.435 s⁻¹ and α = 15°, Eq. [2] will overestimate S_0,blood by ≈3× and, for a peak C_p ~0.2 mM, the net effect is to overestimate v_e by ≈2×. This model of inflow agrees well with the observations (not shown) that S_0,blood ≈ 3 × S_0,muscle and roughly with the observations of v_e,VIF compared with v_e,RR (Fig. 5), thereby suggesting that inflow is the dominant source of discrepancy between the VIF and RR results in this study. The simplest strategy to avoid inflow effects on the VIF is to use a 3D acquisition protocol, but this is difficult in rodents due to the high temporal sampling requirements of the VIF (28). Various other strategies to minimize the inflow effect include: deriving the VIF from small vessels with slowly flowing blood and/or vessels lying in the imaging plane (29); through appropriate RF preparation of out of slice magnetization (26, 30); or through the use of a reference tissue for calibration of the AIF (22,29). Alternatively, the reference region method can be used to avoid the VIF altogether.

The reference region analysis provided values of v_e,_RR that were closer to but still consistently larger than v_e,_SP (Fig. 5). The DCE model used in this study assumed fast exchange of water between intra- and extra-cellular tissue spaces, which may not be a valid assumption, particularly with the introduction of extra-cellular contrast agent (4). However, in the case where fast exchange is not valid, the effect on the DCE analysis is to underestimate not overestimate v_e, so while water exchange may be affecting the DCE analysis it does not explain the observations here. In addition to water dynamics, CA dynamics may also affect DCE parameter estimation. Pellerin et al. postulated that in tumors CA diffuses from well-vascularized regions to adjacent poorly perfused regions over the time-course of a DCE experiment (7). To account for the slow arrival of CA to poorly perfused tissues, the Tofts model assigns very low (non-zero) values to K^trans and overestimates of v_e (7,31). If this phenomenon were contributing to overestimates of v_e,RR, the greatest effect would be expected in necrotic tumor regions, which is consistent with previous DCE observations of large v_e in necrotic tissue by Donahue et al. (17). Indeed, the central tumor voxels of rats 7 and 8 exhibited a large v_e,SP (e.g., Fig. 4f) perhaps indicating a necrotic core, and the DCE analysis showed low K^trans and v_e,RR > 1 from the same voxels. (Of course, slow arrival of ¹¹¹In-DTPA to necrotic regions could also result in a reduced value of v_e,SP, although more time was allowed for tracer distribution in the SPECT studies.) Also, Sourbron and Buckley recently demonstrated that DCE analysis with the Tofts model is sensitive to variations in vascularization and perfusion, independent of CA diffusion between voxels (6). Their analysis shows that in intermediate-to-highly vascularized tissues or tissues that are poorly perfused, v_e may be overestimated by DCE-MRI, which may explain the observations here of v_e,RR > v_e,SP in non-necrotic tumor regions.

The interpretation that DCE-MRI may overestimate v_e might also explain previous studies that have found, surprisingly, only weak to moderate correlations between v_e and the apparent diffusion coefficient (ADC) of water (32–34) in tumors. While DCE-MRI measures of v_e in tumors has been found to shrink in response to treatment (35, 36), water ADC has been shown to increase in response to treatment (36). Perhaps this apparent inconsistency reflects a change in the state of tumor vascularization or perfusion with treatment that results in a more accurate measure of v_e. In this case, the observed v_e from DCE-MRI may go down in response to treatment simply because it is less overestimated, while in fact the extracellular volume fraction goes up, resulting in an increased water ADC. Regardless of the exact source of error in measures of v_e, the wide range of v_e values observed in this study indicates that one should be cautious about an absolute quantitative, physical interpretation of v_e particularly if an in-flow sensitive VIF is used.

Abbreviations Used

DCE

dynamic contrast enhanced

v_e

extracellular-extravascular volume fraction

K^trans

volume transfer constant

VIF

vascular input function

RR

reference region

CA

contrast agent

SPECT

single photon emission computed tomography

TI

inversion time

SPGE

spoiled gradient-echo

Gd-DTPA

gadopentate dimeglumine

¹¹¹In-DTPA

indium 111 labeled diethylene triamine pentaacetic acid

^99mTc-RBCs

technecium 99 metastable labeled red blood cells

CT

computed tomogrphy

v_p

plasma volume fraction

ADC

apparent diffusion coefficient