Vascular space occupancy (VASO)-dependent fMRI is a novel imaging technique that is capable of spatially localizing changes in the microvascular cerebral blood volume (CBV) (1). During neuronal activation, smooth muscle lining the walls of small, intraparenchymal vessels relaxes, thereby facilitating increased blood delivery to tissue. Since activation-related vasodilation is specific to small vessels and not large sympathetically regulated vessels (1), the VASO fMRI signal reports on microvascular effects. Consequently, VASO has the ability to provide improved spatial localization of neuronal activation compared to blood oxygenation level-dependent (BOLD) fMRI (2), which may suffer from signal contamination in and around large draining veins. The VASO technique has been tested using different stimulus paradigms and field strengths (1, 3, 4), giving results consistent with comparable tasks investigated using BOLD and cerebral blood flow (CBF)-based techniques.
Despite the proven sensitivity of VASO fMRI for CBV changes, the signal mechanism is not completely understood. During visual stimulation, absolute VASO signal changes of 2–3% (roughly corresponding to CBV increases of 40–60%) were originally found at 1.5T and 3.0T, at a resolution of 2 × 2 × 5 mm3 (1, 3). Comparable CBV increases (∼30%) were reported in a previous study that used a paramagnetic contrast agent at somewhat lower spatial resolution (5). However, data reported by us and others at a recent conference (6, 7) reveal that absolute VASO signal changes of 4–7% can be found reproducibly at very high spatial resolution (0.78 × 0.78 × 3 mm3), high magnetic field strength (3.0T), and short repetition time (TR = 2 s) (Fig. 1). Such a large signal change corresponds to a CBV increase of more than 80%, which is unlikely in view of the current knowledge of vascular dynamics (1, 8–12). We therefore investigated the VASO contrast mechanism by first modeling potential contributions and subsequently validating these predictions through fMRI experiments performed as a function of TR, TE, in-plane resolution, and signal-to-noise ratio (SNR) threshold for voxel selection. The choice of an appropriate SNR threshold is very important for VASO experiments because some voxels may have very low signal content due to overlap with a large vessel with nulled signal. The results of simulations and experiments demonstrate that the physiological origins of the VASO contrast mechanism are considerably more elaborate than initially proposed.
Multicomponent Biophysical Model for VASO Signal Changes
VASO fMRI is achieved through nonselective (NS) inversion followed by image acquisition at an inversion time (TI) when the longitudinal component of the blood water magnetization is zero (Fig. 2a). Because the T1 of gray matter (GM) is shorter than that of blood, the GM signal is slightly positive at TI, and thus the VASO signal will be sensitive to changes in tissue volume in accordance with changes in the CBV. Similarly to other methods, such as BOLD and arterial spin labeling (ASL), other signals may contribute due to partial-volume effects. For VASO, such signals can originate from white matter (WM), the T1 of which is short compared to GM, and CSF, the T1 of which is long compared to GM (Table 1). VASO fMRI experiments are commonly performed using a TR range of 2000–4000 ms (1, 3, 4), corresponding to a steady-state blood-nulling TI range of 710–994 ms at 3.0T, at which time there will be a negative CSF signal contribution and a positive WM signal contribution (Fig. 2a).
Table 1. MR Relaxation Times at 3.0T and Water Densities (C) Used in the Model Fitting
C (ml H20/ml compartment)
Measured here from the TE-dependence of the extravascular signal.
measured using the same TE, TR parameters as the study of Lu et al. (39).
Lu et al. (14), taken equal for arterial and venous blood.
Unpublished data of transverse relaxation rates as a function of oxygenation and Hct at 3.0T (provided by Clingman, van Zijl). Derived assuming Hct = 0.44 in large vessels, corresponding to Hct = 0.37 in microvessels.
The extent of the GM parenchyma (SGM), WM parenchyma (SWM), and CSF (SCSF) signal contributions to the total MRI signal (STotal) can be described using a three-compartment slow-exchange biophysical model with magnetization fractions X for each compartment:
By assuming that 1) the CSF and parenchymal fractions remain unchanged during the physiological response to stimulus, and 2) the CBV and CBF within GM parenchyma increase, only the GM parenchyma signal will change between rest (S) and activation (S). Thus, the total normalized VASO signal change can be expressed as:
Notice that this is different from BOLD, where susceptibility effects may change signal outside the GM. The water signal in the CSF compartment is given by
while the MR water signals in the parenchymal compartments (extravascular tissue signal + microvascular blood signal) are
where b is blood, MNS(TR,TI) is the longitudinal magnetization after NS inversion, T2 is the transverse relaxation time, and CBV is the CBV fraction in units of ml blood/ml parenchyma (VASO units). The units for MR signal are magnetization/ml tissue for a given voxel, based on the units of M (magnetization/ml water), X (dimensionless), and the water density C (ml water/ml tissue; Table 1).
Under steady-state conditions, the effect of NS inversion is described by
Equation  always applies to CSF, which has no blood. Under steady-state conditions of blood nulling in VASO, M(TR,TI) is zero and the blood signal contribution in Eq.  vanishes, which was always assumed in previous papers. However, it is important to realize that blood effects can contribute to the VASO signal through other means. First, the NS inversion pulse labels blood water in a manner similar to ASL experiments. Second, a steady-state situation may not apply. For instance, fresh blood that flows into the slice within a particular TR may remain and become inverted in the next TR. Thus, a fraction of the blood water is in steady state while the remaining fraction is fresh, with the relative contributions being dependent on TR.
Steady-State Blood Condition Including ASL
When ASL effects occur, the parenchymal steady-state term becomes (see Appendix, Eq. [A6]):
where T1,i is the longitudinal relaxation time for the specific parenchymal tissue, fi is CBF in units of ml/g/s, λ = 0.9 ml/g is the blood–brain partition coefficient (13), and T1,b is the longitudinal relaxation constant for arterial blood water (14), which is approximately equal to that of venous blood water at 1.5T and 3.0T (14), but may start to differ at higher fields.
Fresh Blood Condition Including ASL
Although blood was assumed to be in steady state in previous VASO papers (1, 3), this may not be true depending on the volume of spins inverted by the transmission coil. For instance, if the coil is very small and the TR is sufficiently long, all blood will be fresh. When a large transmit coil and short TR are used, the fresh-blood fraction will likely be small but potentially significant. Under such fresh-blood conditions, the extravascular tissue magnetization in the parenchymal signal (Eq. ) must be expanded to include the effects of both steady-state (nulled) and fresh (non-nulled) blood water:
in which M(TR,TI) can be evaluated by choosing TR approaching ∞ in Eq. . X is the fraction of blood in steady state. In addition, the intravascular blood magnetization in the parenchymal signal (Eq. ) is given by Eq.  for TR approaching ∞ and is weighted by (1 − X). It is assumed for simplicity that X will be identical for intravascular blood and extravascular tissue.
A second important issue when considering inflow of fresh spins is that one average microvascular blood volume compartment can no longer be considered. In reality, microvasculature consists of arteriolar, capillary, and venular contributions, in an approximately 21%, 33%, and 46% contribution, respectively, which is often approximated by a two-compartment model consisting of a 30%/70% arteriolar/venular ratio. Upon inflow, fresh-blood water spins will first replace the arteriolar compartment, followed by capillary and venular compartments. For the TI range used in most VASO experiments (710–1106 ms), it may be assumed that no fresh blood has entered the venular space during the experiments, and that no fresh blood has reached the imaging slice in the first scan. Thus, simulations can be performed for fresh (a)rteriolar space and steady-state (v)enular space according to the two-compartment model, i.e., CBV =0.3 · CBVa + 0.7 · CBVv, with the appropriate relaxation times (Table 1). At very long TR, the steady-state and fresh-blood water magnetizations converge.
Experimental Estimation of CSF
It is assumed in Eq.  that CSF may contribute to the baseline MR signal, but that the CSF fraction does not change between rest and activation. While this is assumed in most fMRI models (15), more information would be useful. To estimate CSF changes experimentally, we acquired an image at the point in time when blood and CSF were simultaneously nulled. This two-inversion-pulse experiment (Fig. 2b) is a combination of VASO and the fluid attenuated inversion recovery (FLAIR) technique, and is here dubbed VASO-FLAIR. The magnetization term for such an experiment is
TI1 and TI2 are shown in Fig. 2b. When these experiments are performed at a resolution that is sufficiently high to neglect WM contributions, the results of this procedure can be combined with the VASO fMRI data to estimate CSF volume fraction changes between rest and activation. At long TR, ASL effects may be neglected and the absolute VASO signal can be approximated by
where A is a constant that gives the MRI signal per unit volume of water protons at equilibrium. Note that because T1,CSF > T1,GM (Table 1), M(TR,TI) is negative at TI for blood nulling. Therefore, absolute value signs must be included in Eq.  to render it correct for voxels in which the negative CSF term outweighs the positive GM term (i.e., voxels that are primarily CSF).
The VASO-FLAIR signal can be approximated by
in which M(TR,TI) is expressed in Eq. . The constant, A, can be calculated using the maximum signal intensity in ventricles, S, as a reference (pure CSF). Equation  can then be rearranged to solve explicitly for A given XCSF = 1:
Once A has been calculated, it is possible to solve Eqs.  and  simultaneously for the two remaining unknowns: CBV and XCSF. While the above equations are only approximate, they allow for a magnitude estimation of CSF changes between activation and rest.
MATERIALS AND METHODS
The dependence of VASO signal on the MR parameters TR and TE were first simulated from the slow-exchange biophysical model proposed above assuming healthy physiological values for CBF and CBV. GM CBV values have been reported to be in the range of 0.048–0.055 ml/ml, varying slightly with the imaging modality and technique employed (16–20), and therefore an approximate mean literature value of 0.0525 ml/ml was chosen for our simulations. Similarly, GM CBF has been consistently reported to be in the range of 43–65 ml/100 g/min (16, 20), and an approximate average value of 55 ml/100 g/min was chosen. GM CBF increases of 60–150% during neuronal activation have been reported (21), and we chose an intermediate GM CBF increase of 91% (corresponding to CBF = 105 ml/100 g/min), as measured in a previous study using ASL during fMRI tasks (22). When we determined the ASL contribution, we assumed that capillary permeability remains unchanged during activation. Using knowledge about GM and WM CBF ratios from the positron emission tomography (PET) literature (18), we assumed that WM CBF can be approximated by
Furthermore, CBF and CBV changes are often related through Grubb's empirical equation:
α was previously fitted for and found to be 0.38 in monkey experiments (11). However, the same data can also be satisfactorily fitted using a factor α = 0.50 (23), which would reflect a vascular model of expanding infinitely long random cylinders. We modeled both approximations in this study: CBV = 0.0725 (α = 0.5), and CBV = 0.067 (α = 0.38). Other physiological values included CBVWM = 0.033 ml/ml (Eq. ) and CBFWM = 22 ml/100 g/min (Eq. ). In partial-volume simulations, a CSF fraction of 15% and a WM fraction of 5% were chosen. Clearly there is a range of healthy physiological values, and, as discussed below, different choices will affect the simulation results. To be as objective as possible, we simply chose the above mean literature values to demonstrate the model. The CSF fraction is most variable and therefore measurements of this fraction were performed.
Finally, it is possible that spin-echo (SE)-BOLD effects may have a significant influence on VASO signal changes, especially at long TE. Using the approximation that the SE-BOLD effect is approximately 1/3 (24) that of the extravascular gradient-echo (GRE)-BOLD effect, which was previously measured to be ΔR = −0.38s−1 (3), it is possible to incorporate SE-BOLD effects into the model. When ΔR2 = −0.127s−1 was combined with the extravascular GM T2 (Table 1), it was found that T = 71.4 ms, which can be used in the TE-dependent term of Eq.  to simulate SE-BOLD effects in VASO.
The above parameters were inserted into the slow-exchange biophysical model, and VASO signal changes were simulated for a TR range of 2000–7000 ms and TE range of 24–54 ms. CBV, CBF, SE-BOLD, WM, CSF, and fresh-blood contributions were simulated successively to show how each contribution influenced the VASO signal change.
All experiments were performed on a 3.0T MRI scanner (Philips Medical Systems, Best, The Netherlands) using body coil transmission and sensitivity encoding (SENSE) head coil reception. Written informed consent was obtained from all subjects in accordance with institutional review board guidelines. In all experiments the subject's head was immobilized with a Velcro® strap and foam padding to minimize unwanted motion during the scan. A flashing blue-yellow checkerboard (visual angle = 25°, frequency = 10.05 Hz) was used for visual stimulation. The subject was instructed to fixate on a cross sign in the center of the screen during rest periods. Five separate fMRI experiments were performed: 1) high-spatial-resolution (0.78 × 0.78 × 3 mm3) VASO, 2) VASO with varying TR, 3) VASO with varying TE, 4) VASO with varying in-plane resolution, and 5) VASO followed by VASO-FLAIR for CSF analysis. The experimental conditions are described separately below.
High-spatial-resolution VASO fMRI data were first acquired at submillimeter resolution to investigate whether previous lower-spatial-resolution VASO fMRI studies gave results consistent with the high-resolution acquisitions. Visual stimulation consisted of a 30-s on/30-s off flashing checkerboard sequence repeated six times. This entire scan was repeated four times and acquisitions were averaged to obtain a sufficient SNR. A two-shot, segmented-EPI SE readout was used with FOV = 112 × 112 mm2, imaging matrix = 144 × 144 and slice thickness = 3 mm, for an in-plane resolution of 780 × 780 μm2. Other imaging parameters included adiabatic inversion, SENSE factor = 2.5, and TR/TI/TE = 3000/889/32 ms.
We then investigated the possible ASL contribution to the VASO effect by performing experiments for varying TR. To enable this to be done within a reasonable time frame, the minimum spatial resolution was changed from the ultrahigh 780 × 780 μm2 in-plane (30:00-min acquisition) to 1.89 × 1.89 mm2 in-plane (5:30-min acquisition). This reduced resolution is still much higher than the approximately 3 × 3 mm2 in-plane acquisition typically used for CBF, BOLD, and VASO fMRI studies at 3.0T (3). For the TR study, single-slice VASO fMRI was performed on five healthy volunteers (age range = 23–41 years). A standard T2-weighted anatomical image was first acquired for the purpose of locating the calcarine sulcus. Six separate VASO scans were performed for TR = 2000–7000 ms in 1000-ms increments. The order of the scans was pseudo-randomized for each subject. Each trial consisted of 30 s of stimulation followed by 60 s of baseline cross-sign fixation. This paradigm was repeated three times for each TR value and averaged to improve the SNR. In each scan, a delay of 60 s was used before the first stimulus to allow the subjects to reach a hemodynamic steady state (data acquired during this time were not analyzed). The TI was varied so as to keep blood water signal nulled in all scans, giving TI = 710, 889, 994, 1054, 1088, and 1106 ms for TR = 2000, 3000, 4000, 5000, 6000, and 7000 ms, respectively. Other imaging parameters included FOV = 212 × 212 mm2, image matrix = 112 × 112, SENSE factor = 2.5, SE acquisition (TE = 44 ms), in-plane resolution = 1.89 × 1.89 mm2, and slice thickness = 3 mm.
There was a concern that SE-BOLD effects would occur at such a long TE, and therefore TE-dependent data were acquired on the same volunteers using an identical paradigm as in the TR study. At 1.89 × 1.89 mm2 in-plane resolution, the minimum TE for a single-shot EPI readout is approximately 44 ms. For this reason, segmented two-shot EPI was employed to allow measurement over a TE range of 24–54 ms in 10-ms increments (four TEs per subject). Long-TR acquisitions yield high SNR, but because of the long time between acquisitions, very few images are averaged during times of activation. By contrast, short-TR scans allow for the acquisition of many images, but the SNR is low compared to that of long-TR scans. Therefore, TR = 3000 ms (TI = 889 ms) was chosen as a good intermediate value for the TE analysis.
To assess the influence of partial-volume effects, we performed experiments on the same five subjects in the TR and TE studies using a constant slice thickness of 3 mm and varying in-plane voxel resolutions of 1.89 × 1.89, 2.21 × 2.21, 2.65 × 2.65 and 3.31 × 3.31 mm2, corresponding to volumes of 10.7, 14.7, 21.1, and 32.9 μl, respectively. These experiments were performed at TR = 2000 ms and 7000 ms using a constant FOV = 212 × 212 mm2, but an increasing imaging matrix. Other imaging parameters were identical to those used in the TR experiments described above.
Finally, we performed VASO-FLAIR experiments on five subjects sequentially with VASO experiments at a voxel resolution of 1.89 × 1.89 × 3 mm3 to assess whether the CSF volume fractions changed between rest and activation. Prior to fMRI, we checked for CSF nulling by performing VASO-FLAIR experiments as a function of TI1 and TI2, and analyzing CSF signal in the ventricles (results not shown). The imaging parameters were TI/TR = 1106/7000 ms for VASO, and TI1/TI2/TR = 2893/849/7000 ms for VASO-FLAIR. A TR of 7000 ms was chosen to allow for a reasonable number of images to be acquired within the 5:30-min fMRI paradigm and to ensure that a sufficiently high remaining GM magnetization would be present at acquisition (∼10% in VASO-FLAIR vs. ∼20% in VASO).
Images were realigned using standard 2D rigid body registration Automated Image Registration (AIR) routines (25). For each voxel the time course was corrected for linear baseline drift using a cubic spline interpolation algorithm. Fractional signal change was calculated from the average signal intensity of all activated voxels.
Recent data suggest that the CBV-based VASO hemodynamic response may take as long as 10 s to return to baseline after visual stimulation (12). Therefore, data acquired less than 10 s after conclusion of the flashing checkerboard were not used in the determination of S. In addition, the first and last data points within the stimulation period were not analyzed in the determination of S. This was done to avoid potential errors in calculating signal change, as it eliminates transition periods when the subject may not be in a pure resting or activated state.
A region of interest (ROI) encircling the occipital lobe was drawn, and voxels within this region were analyzed using a cross-correlation (cc) test for activation consisting of cluster size ≥ 15, and cc < –0.15 (TR, TE, VASO-FLAIR, and resolution experiments) or cc < –0.20 (submillimeter-resolution experiment). Note that the cc threshold is negative because GM signal decreases in VASO experiments (1). Since noise may vary between subjects, as well as with MR parameter choice and spatial resolution, the effect of SNR on ΔS/S was studied separately, and a threshold of SNR ≥ 20 was determined to be necessary for voxels to be reliably analyzed (see Results).
For analysis of CSF volume fractions during rest and activation, we solved Eqs.  and  voxel-by-voxel for XCSF by combining the VASO and VASO-FLAIR fMRI signals. Next, voxels that showed activation in the VASO-FLAIR experiment were overlaid on the calculated CSF maps. For all voxels that met the activation criteria, the average CSF fraction was calculated during rest and activation periods.
Using Eqs.  and , and the assumptions described in the Simulation section above, contributions for the three voxel compartments (GM, WM, and CSF) were determined as shown in Fig. 3a and b. Over the TR and TE ranges shown, the GM and WM signals are always positive, whereas the CSF signal is always negative.
The simulations in Fig. 3c show that the addition of the ASL contribution in GM produces a strong TR dependence for the VASO signal change. This simulation was performed by replacing the parenchyma magnetization term in Eq.  with the parenchyma magnetization term given by Eq. . We term this effect “vascular space labeling” (VASL). At longer TR, the VASL contribution becomes small and the GM signal plateaus to be predominantly due to CBV. Because a TR of 3 s was used, the CBF effect also makes the VASO signal changes in the TE experiments more negative (Fig. 3d), although there is no TE-dependent effect due to CBF.
Figure 3e and f depict how SE-BOLD, CSF, and WM effects will contribute to the CBF+CBV response shown in Fig. 3c and d, given the multicompartment model described by Eq. . The SE-BOLD effect, although small, reduces the VASO effect, which is expressed as a baseline contribution to the TR-dependent data (Fig. 3e) and a TE-dependent effect in the TE curve (Fig. 3f). The CSF contribution (simulated here as 15%) has a large baseline effect at long TR and TE. Therefore, if CSF contributes significantly to the voxel (i.e., there is a large partial-volume effect), this will become visible by the VASO signal change becoming more negative at long TR and TE. The simulation suggests that this partial-volume effect can actually undo the small SE-BOLD effect and lead to a negative slope in the TE dependence (Fig. 3f). Partial-volume effects with WM (simulated here for 5%) will slightly reduce the magnitude of the VASO effect as well.
The effect of fresh blood on the VASO signal (Eq. ) is shown in Fig. 3g and h. This effect is highly TR dependent (Fig. 3g), with a signal change of approximately –32% for pure fresh blood and of approximately –11% for a compartment composed of 30% arteriolar (fresh) and 70% venular (steady-state) blood water. In addition, the TE dependence is more pronounced when the fresh-blood contribution is introduced (Fig. 3h), with the signal changes varying by almost 4% over the 24–54 ms range, compared to only a 0.5% variation for the steady-state assumption.
Figure 4 shows the relationship between the VASO signal change and SNR threshold used for selecting voxels. The data show that low SNR produces erroneously large negative signal changes, necessitating the use of SNR constraints for the acquisition of reproducible VASO data. The results indicate that the VASO signal change largely plateaus above SNR = 20, and the selection criterion for activated voxels therefore required SNR ≥ 20.
Visual activation maps (acquired from the multi-TR VASO study) overlaid on VASO-fMRI images at TR = 2000 ms and TR = 7000 ms (Fig. 5a and b) show that activation is largely localized to regions of GM, in line with the microvascular origin of the VASO contrast. The corresponding time courses in Fig. 5c and d illustrate that the absolute value of the VASO signal change at TR = 2000 ms (6–7%) is much larger than at TR = 7000 ms (3–4%), in line with the model prediction that signal changes would be most negative at short TR due to CBF effects being most pronounced in this range.
Figure 6a and b show results from typical VASO and VASO-FLAIR experiments, respectively. These images, acquired at TR = 7000 ms, were used to calculate CSF volume fractions (Eqs. 11, 12) in voxels that met the activation criteria (Fig. 6c). The resulting CSF map for this example is shown in Fig. 6d. Notice the CSF nulling (which is most obvious in the ventricles) of the VASO-FLAIR image and the large CSF volume fractions calculated in the ventricles, as expected. The results for calculated CSF volume fractions during periods of rest and activation are presented in Table 2. All subjects show a slight reduction in CSF fraction upon activation; however, this effect was not statistically significant when a paired t-test was applied across the five subject averages (P-value = 0.61; confidence interval (CI) = –1.95 to 3.14). The effect of the average CSF volume fraction change from 10.7% to 10.1% would lead to the VASO signal change becoming less negative by only 0.05% and 0.2% at TR = 2 s and TR = 7 s, respectively. This small variation is well within experimental error, and therefore the assumption that CSF volume fractions do not change between rest and activation is supported by these results.
Table 2. CSF Volume Fractions for Five Subjects During Rest and Activation Periods*
Individual subjects are mean CSF fraction in all voxels showing activation ± standard error. Mean is average of five subject CSF fraction ± SD. Standard error was used for individual subject error analysis since subjects had varying numbers of activated voxels. When applying a paired t-test to the mean CSF fractions, the change was found to be statistically insignificant (P-value = 0.61; confidence interval =–1.95 to 3.14).
XCSF Rest %
11.1 ± 0.4
7.77 ± 0.2
11.1 ± 0.4
10.9 ± 0.3
12.8 ± 0.5
10.7 ± 1.8
XCSF Act %
10.7 ± 0.4
7.31 ± 0.2
10.9 ± 0.4
10.2 ± 0.3
11.6 ± 0.5
10.1 ± 1.5
The measured dependencies of the SE-VASO signal change on TR and TE for voxels that met the activation criteria (see Data Processing section) averaged over five volunteers are shown in Fig. 7. The slightly negative TE dependence indicates that the SE-BOLD effect is small, in line with the expectation that SE-BOLD effects at this field strength mainly have intravascular contributions (23, 26), which are ideally nulled in the VASO experiment. Overlaid on the TR and TE experimental data is the steady-state blood model assuming typical physiological CBF and CBV values for α = 0.5 (solid line) and α = 0.38 (dashed line). An average CSF fraction of 10.5% was chosen based on the VASO-FLAIR results, and a small WM volume fraction of 2% was incorporated as well. The TR dependence and TE dependence agree well with these model predictions.
Spatial resolution data for voxels that met the activation criteria (Fig. 7c) show that VASO signal changes become less negative as voxel size increases. Assuming the healthy physiological values described above at the highest resolution, we performed a two-variable nonlinear generalized reduced gradient fit for XCSF and XWM separately for each of the lower-resolution data points (voxel volumes 14.7–32.9 μl). The fit results revealed that XCSF = 13.6%, 13.8%, and 13.8%, and XWM = 9.3%, 13.0%, and 17.8% for the voxel volumes of 14.7, 21.1, and 32.9 μl, respectively. These fit results are only an approximation and will vary with the CBF and CBV values used. However, the results suggest that the WM partial-volume effect increases with increasing voxel volume, whereas the CSF partial-volume effect increases only slightly.
Our results indicate clearly that, depending on the repetition time chosen, the VASO signal change does not arise predominantly as a result of changes in a single physiological parameter. First, in addition to the originally proposed CBV effect (1), an NS inversion pulse followed by a waiting time is also sensitive to CBF changes and possible inflow of fresh-blood water spins. The CBF dependence is due to the occurrence of ASL during this inversion pulse and is most prominent in short-TR acquisitions. The fresh spin effect is due to the limited excitation range of the coil and may confound the steady-state assumptions at shorter TR. Second, similarly to other functional imaging modalities that are acquired at rather coarse spatial resolution, the primary effects of CBV and CBF are combined with partial-volume effects from CSF and WM. Third, simulations show that when longer TE values are used, the extravascular BOLD effect is expected to reduce the VASO signal changes. Fortunately, one can vary the contributions of these different effects simply by changing the MR acquisition parameters (TR for CBF and inflow, inversion range for inflow of fresh spins, spatial resolution for partial-volume effects, and TE for extravascular BOLD). Thus, the VASO sequence may provide a versatile means of noninvasively investigating CBV and CBF, as well as the mechanisms of the BOLD effect and the relationships among CBV, CBF, and oxygen metabolism.
Effect of CBF Changes on the VASO Signal
The simulations and experimental data show that the CBF contribution to the VASO experiment (i.e., VASL) enhances the signal changes at short TR. This effect probably was not noticed in a previous study at 1.5T (1) because the low spatial resolution typically used in fMRI experiments causes partial-volume averaging with nonactivated regions, which makes the magnitude of the actual VASO effects appear smaller and within the expectations of published CBV changes. In the present study at twice the field strength and SNR, we were able to use a higher spatial resolution, which led to noticeable nuances. Even though a measurement dependence on two physiological parameters may at first be construed as being a disadvantage, for a number of reasons this is not entirely the case. First, by imaging at long TR, contrast that is predominantly dependent on CBV can be obtained. As recently shown, when such long TR measurements are performed before and after a gadolinium bolus, absolute CBV measurements can be achieved (17). Second, at short TR the VASL effect actually increases the contrast-to-noise ratio (CNR) in VASO activation studies. Third, the ASL contribution may provide investigators an opportunity to determine blood flow changes during physiological perturbations by simply varying the TR. However, a detailed analysis and validation of the potential to measure CBF is beyond the scope of this paper. We therefore restrict ourselves to commenting on some of the potentially interesting information that such an approach might provide compared to existing ASL approaches (27–30). The spin labeling that occurs during the NS pulse in the VASO-VASL experiment is fundamentally similar to that in pulsed ASL (PASL), and the theory (31) used in our model was similar, except for the inclusion of steady-state blood (Appendix). However, the derivation of CBF values would be very different from ASL, in which a reference scan with a different spatial saturation profile is required to define an input bolus that travels to the tissue, analogously to a PET tracer. This makes ASL dependent on the transit time between the front edge of the bolus and the tissue, and thus makes it difficult to study WM (long transit time) and pathological conditions, such as ischemia, in which the input function is unknown. Also, outside the brain, when vessels leading to tissue are not as easily labeled as those feeding the brain, ASL becomes problematic. In a VASL approach, all vascular spins are labeled, and consequently inverted blood water will immediately start to exchange with tissue water, making the arterial input function (AIF) effectively instantaneous. On the other hand, there are many variables to be fit in the VASL approach. Currently, studies are being designed to address and validate the possibility of fitting the proposed biophysical model to a VASO-VASL experiment acquired at multiple TR values to quantitatively determine CBF and CBV.
In addition to a dependence on two physiological parameters, the VASO experiment has several experimental limitations that may affect signal changes as well. Two issues that deserve attention are 1) the efficiency of the NS inversion pulse for nulling the blood signal, and 2) the assumption of a steady state for blood nulling (the accuracy of which depends on the volume over which the adiabatic inversion pulse is applied).
The quality of blood nulling in an inversion-recovery approach depends on the inversion efficiency as well as proper knowledge of blood T1. To ensure the former, adiabatic inversion pulses, which are known to have very high inversion efficiency, were used in all of the presented experiments. This inversion efficiency was measured in voxels within ventricles, assuming pure CSF and known T1. A comparison of CSF signals at short and very long TI indicated that we achieved nearly perfect inversion efficiency. It is assumed that the inversion efficiency is similar between blood water and CSF; however, it is possible that the inversion efficiency for blood water is very slightly lower due to the higher velocity of protons in blood. T1 for blood water was recently measured at 3.0T (14), although this value may vary slightly depending on hematocrit and, to a lesser extent, blood oxygenation (e.g., arteriole, capillary, and venule). This oxygenation dependence is quite small at 1.5T and 3.0T, but may become a factor at higher fields. For simplicity, we assumed negligible effects of these parameters in the current paper. However, it is important to keep in mind that, as with other ASL sequences, small errors in blood T1 and inversion efficiency will affect quantification.
A related issue arises as a result of the finite volume of the body coil used for inversion. When adiabatic pulses are used, a typical body coil will invert a volume of 400–450 mm around its isocenter, which suggests that only 200–225 mm, on average, will be inverted proximal to the imaging slice (toward the heart). For finite TR values, fresh-blood water that is not in steady state will flow into the inversion volume. In the next TR period, this fresh-blood water will have experienced only one inversion pulse—not two as assumed by the steady-state equations. Since optimal TI values were calculated for blood in steady state, fresh blood will contribute additional effects to the VASO signal changes, as simulated in Fig. 3g and h.
Also, in addition to parenchymal space contracting to accommodate fresh blood water, the tissue magnetization (WM and GM) for fresh blood must be modeled differently as well (Eq. [A6]). The extent to which steady-state and fresh-blood water contribute to VASO signal changes at short TR is largely speculative at this point. The contributions will vary with the size of the subject relative to the body coil and the precise volume of the body coil itself, which will vary with the MR system manufacturer. The ASL literature reports that, when using a labeling slab with trailing edge around 92.5 mm below the slice, it takes 1500 ms for fresh blood water to reach the capillary exchange site (32). In addition, it may take an additional 800 ms once blood water reaches the capillary exchange site for it to reach the venules. Therefore, it is likely that at short TR, the majority of blood water is in steady state, with the exception of a small fraction in arterioles. When the measured data are studied, it can be seen that at both TR = 2 s and TR = 3 s blood water appears to be largely in steady state, in agreement with expectations. At long TR, a much greater fraction of the blood is likely to be fresh. However, since the expressions for steady-state and fresh-blood water magnetization converge at long TR (Fig. 3g), this observation is not as important as the steady-state/fresh-blood considerations at short TR, where the signal contributions may vary considerably (Fig. 3g). It should be noted that the model curves will also shift depending on the CBF, CBV, and MR constants (T1, T2, etc.). Therefore, care should be taken in interpreting steady-state and fresh-blood volume fractions at this point. What should be noted at this time is that VASO signal changes are subject to greater error at short TR due to uncertainty in steady-state and fresh-blood water assumptions. This important point is currently being studied in detail. Before absolute CBF and CBV can be quantified through a VASL approach (see above), a thorough understanding of the fresh/steady-state blood issue must be achieved.
Similarly to other fMRI approaches that employ large voxels, VASO is susceptible to partial-volume averaging of GM with neighboring CSF and WM. We obtained an estimate of the magnitude of the CSF fractions during rest and activation using a VASO-FLAIR sequence. The data show that CSF fractions are approximately 10–11%, on average, at a high fMRI resolution of 1.89 × 1.89 × 3 mm3 and that the fractional changes between rest and activation are minimal (<1% generally). Furthermore, by performing fMRI experiments as a function of in-plane resolution, we found that signal changes became less negative by approximately 1% and 2.5% when voxel resolution was decreased from 10.7 μl to 32.9 μl at a TR of 7 s and 2 s, respectively. Our model suggests that this variation is largely due to an increase in WM partial-volume effects, which are of opposite sign to CSF effects.
SE vs. GRE BOLD Effects
The long TEs required for high-spatial-resolution VASO-fMRI raise the possibility of contributing SE-BOLD effects. Since VASO images are acquired at the null time of blood, intravascular BOLD effects are minimized. In reality, however, small errors in inversion efficiency may allow for some intravascular blood signal contribution. Because T2 > T, SE sequences are likely to give more intravascular blood signal than gradient echoes, an effect that becomes especially relevant in capillaries where oxygenation is high (3). However, for most of the venular system, extravascular BOLD effects are known to be about one-third smaller than those from GRE experiments (24), which suggests that SE-VASO should be relatively insensitive to changes in blood oxygenation at 3.0T. This was confirmed in our TE experiments, in which an increase in VASO effect was observed. This is because the small residual extravascular SE-BOLD effect is undone by the contribution from CSF signal (most noticeably at long TR). This is actually a fortunate circumstance because the use of a SE sequence was important for the present experiments to accommodate the long readout times needed for the high spatial resolution without too much signal loss in single-shot EPI acquisitions. The occurrence of a slope opposite of that expected for BOLD effects also eliminates the likelihood of a substantial GRE contribution that may occur due to the use of a long readout period around the TE.
Correct Value of the Grubb Exponent
In this study the proposed slow-exchange biophysical model was simulated for healthy physiological GM CBF = 55 ml/100 g/min and CBV = 0.0525 ml/100 g/min. For these values, a Grubb exponent of 0.38 (CBV = 0.067 ml/ml) models the data much worse than a Grubb exponent of 0.50 (CBV = 0.0725 ml/ml). However, since the TR dependency is CBF dependent, one could reverse this simply by choosing a different resting GM CBF, which would still be well within a healthy physiological range. Therefore, no definitive information concerning the correct value of the Grubb exponent can be learned from this study.
Assumptions Underlying the VASO Experiment
Two issues that have not yet been completely detailed are the physiological explanation for the VASO effect, and the effect of exchange on this phenomenon. It is important to stress that experiments in several VASO-based studies showed beyond a doubt that tissue signal is lost and vascular space increases (1, 3, 12). However, some uncertainty remains about which “tissue water” is lost, and where it goes. Experimental evidence from the present study indicates that CSF space shows a trend for reducing slightly (10.7–10.1% for GM); however, even though this was found in all subjects, the reduction was not significant. Therefore, in the model used in this paper, total CSF, GM, and WM contributing to the voxel are unchanged (Eq. ), while the water shifts are assumed to occur within the parenchyma. At this point we need to describe how “parenchyma” is defined. In our definition we include GM tissue and microvessels (arterioles, capillaries, and venules) up to about 200 μm in size. Thus, the CBV is the sum of this microvascular blood volume, including most vessels that can expand and cause vasodilatation during focal activation, especially the pial arteries and arteriolar spaces, and probably also capillaries and venules (33). Given that it is difficult to reduce water volume under pressure, one possible mechanism for vasodilatation during activation is through relaxation of smooth muscle and pericytes, with a water shift from these tissue cells to the vasculature. This effect is represented in Eq. , which shows that CBV increases and tissue volume reduces, while a constant total GM fraction is maintained in Eq. . In this respect it is important to note that smooth muscle and pericyte water are included in the tissue fraction (9), indicating that most of the expansion probably occurs in the pial arteries, arterioles, and capillaries. However, experimental evidence for this is not conclusive, and even though expansion of the diameter may be smaller in venules, the total volume is about twice that of the arterioles, which may in large part compensate for the smaller expansion. We therefore used a simplified single microvascular space model. Future research, perhaps at higher fields (where T1 of arteries and veins may become different) and at higher spatial resolution (where parenchyma and vessels can be separated out) may shed light on this issue. A final possibility is that some of the vascular expansion is compensated for by the large veins downstream—a contribution that was not included in the model. However, in the first VASO paper (1) it was shown that the tissue signal reductions are well localized in the GM, so this latter explanation, even though possible, seems less likely.
The second issue of importance is how to properly include the effects of exchange between tissue and capillary space (34). In an appendix to the first VASO paper (1), we estimated the effect of exchange on the VASO signal change and found it to be negligible. In agreement with this, St. Lawrence and Ye (35) also reported simulations that showed a negligible effect of exchange on the VASO effect. When the ASL effect is included, the exchange is assumed to be fast for simplicity, which may at first seem contradictory to the separate compartments in the VASO equation. However, previous in-depth studies and simulations of the ASL effect (35–37) have shown that the use of equations for fast and slow exchange give the same results at very low magnetic fields (1.5T, 3T), and that more detailed equations are needed only at higher field (38). Thus, even though the equations would be more complete if exchange were taken into effect, the VASO and ASL literature suggests that the exchange effects on the resulting parameters are probably small at the field strength used in the current study.
The data in this paper illuminate a variety of complex processes underlying VASO-fMRI contrast, and ultimately reveal this method to be a more complex, yet more versatile imaging technique than originally thought. It was demonstrated that the VASO signal mechanism arises from multiple physiological contributions, including those from CBV, CBF, and, as shown previously at longer TE (3), extravascular BOLD effects. In addition, inflow of fresh spins may contribute to the VASO effect, partial-volume effects with CSF and sometimes WM are significant, and the use of a SNR threshold is needed to avoid spurious activation in voxels that partial volume with large vessels. The CBF contribution, inflow of fresh spins, and partial-volume effects with CSF enhance the VASO effect and make it easier to detect when using short TR and long TE in an SE sequence. At sufficiently high spatial resolution, the multiple effects contributing to VASO can be largely separated by varying the MR acquisition parameters, which may allow individual physiological parameters to be determined in the future.
This work provides an explanation for the larger signal changes recently observed in VASO-fMRI, and a framework for quantifying important physiological information from VASO experiments in the future.
The authors are grateful to Terri Brawner, Kathleen Kahl, and Joe Gillen for experimental assistance. Dr. Jones is supported by a grant from Philips Medical Systems to the Kennedy Krieger Research Institute. Dr. van Zijl is a paid lecturer for Philips Medical Systems. This arrangement has been approved by Johns Hopkins University in accordance with its conflict of interest policies.
Solution to the Flow-Modified Bloch Equation Assuming a Steady-State Blood Water Contribution
The Bloch equation, modified to include perfusion, for spin-lattice relaxation of tissue water is given by
where the tissue, arterial blood, and venous blood water magnetizations are given by M(t), Ma(t), and Mv(t), respectively; t is the time between NS inversion and excitation; TR is the time between inversion pulses in consecutive dynamic scans; M0 is the steady-state equilibrium value of tissue magnetization; and, assuming a well-mixed blood compartment, Mv(TR,t) = MNS,st-state (TR,t)/λ. The condition for steady-state arterial blood magnetization is given by
where T1,b is the transverse relaxation time of arterial blood water. Notice that Ma(TR,t) is commonly assumed to be in equilibrium, the requirement for completely refreshed blood in each TR (TR→∞, in which case the final term in Eq. [A2] vanishes). Upon substitution of Eq. [A2], Eq. [A1] can be rewritten as
where T1,app is defined in the text. The solution of Eq. [A3] is found to be
where Meq(t) is the solution to Eq. [A1] assuming arterial blood in equilibrium (Eq. [A2] with TR→∞) that is most commonly reported,
and M(0) = M0(e−1) for i = GM or WM. Upon substitution of Eq. [A5] into Eq. [A4], MNS,st-state(TR,t = TI) can be rewritten as