Cerebral Blood Flow,Blood Volume,and Oxygen Metabolism Dynamics in Human Visual and Motor Cortex as Measured by Whole-Brain Multi-Modal Magnetic Resonance Imaging

Abstract

The development of neuroimaging methods to characterize flow-metabolism coupling is crucial for understanding mechanisms that subserve oxygen delivery. Functional magnetic resonance imaging (fMRI) using blood-oxygenation-level-dependent (BOLD) contrast reflects composite changes in cerebral blood volume (CBV), cerebral blood flow (CBF), and the cerebral metabolic rate of oxygen consumption (CMRO₂). However, it is difficult to separate these parameters from the composite BOLD signal, thereby hampering MR-based flow-metabolism coupling studies. Here, a novel, noninvasive CBV-weighted MRI approach (VASO-FLAIR with 3D GRASE (GRadient-And-Spin-Echo)) is used in conjunction with CBF-weighted and BOLD fMRI in healthy volunteers (n=7) performing simultaneous visual (8 Hz flashing-checkerboard) and motor (1 Hz unilateral joystick) tasks. This approach allows for CBV, CBF, and CMRO₂ to be estimated, yielding (mean±s.d.): ΔCBF=63%±12%, ΔCBV=17%±7%, and ΔCMRO₂=13%±11% in the visual cortex, and ΔCBF=46%±11%, ΔCBV=8%±3%, and ΔCMRO₂=12%±13% in the motor cortex. Following the visual and motor tasks, the BOLD signal became more negative (P=0.003) and persisted longer (P=0.006) in the visual cortex compared with the motor cortex, whereas CBV and CBF returned to baseline earlier and equivalently. The proposed whole-brain technique should be useful for assessing regional discrepancies in hemodynamic reactivity without the use of intravascular contrast agents.

Introduction

Understanding the relationship between neuronal activity, hemodynamic response, and energy metabolism is central to relating functional neuroimaging findings to underlying neuronal processing. The majority of functional neuroimaging is conducted noninvasively using blood-oxygenation-level-dependent (BOLD) magnetic resonance imaging (MRI) (Ogawa et al, 1990). Although the BOLD signal is a surrogate marker of neuronal activity through composite adjustments in cerebral blood flow (CBF), cerebral blood volume (CBV), and the cerebral metabolic rate of oxygen consumption (CMRO₂) (Buxton et al, 1998; van Zijl et al, 1998), it is unclear how each of these components influences the BOLD response and to what extent the BOLD effect reflects neuronal activity.

It is possible to better understand the processes that drive BOLD contrast by separately measuring CBF and CBV changes, and using these measures together with the BOLD response to estimate CMRO₂ (Lu et al, 2004b). In this way, the temporal dynamics of individual hemodynamic parameters can be investigated and the relationship between hemodynamic and metabolic response assessed. However, such an approach requires accurate CBF and CBV measurements over a large spatial volume, in addition to a model relating BOLD signal changes to CBF, CBV, and CMRO₂. The emergence of novel MRI approaches such as CBF-weighted (CBFw) arterial spin labeling (ASL: (Williams et al, 1992)) and CBV-weighted (CBVw) vascular-space-occupancy (VASO: (Lu et al, 2003)) has made this approach, in principle, possible.

Here, using a novel 3D GRadient-And-Spin-Echo (GRASE) acquisition, we propose a whole-brain MRI protocol for measuring CBF, CBV, and BOLD reactivity in sequence without the use of intravascular contrast agents. The CBVw, CBFw, and BOLD dynamics in response to simultaneous visual stimulation (8 Hz flashing checkerboard) and motor stimulation (1 Hz unilateral joystick movement) are investigated and specific CBF, CBV, and CMRO₂ contributions to BOLD contrast reported. The proposed technique should expand on the available approaches for studying neurovascular and flow-metabolism coupling.

Materials and methods

Arterial-Spin-Labeling and Vascular-Space-Occupancy Magnetic Resonance Imaging

CBFw ASL has emerged as a technique for generating quantitative CBF measurements without the use of contrast agents (Williams et al, 1992). In ASL, blood water outside an imaging region is magnetically tagged with a radiofrequency pulse, after which blood water flows into the imaging region, exchanges with tissue water, and a ‘tag’ image is acquired. A CBFw image is generated by subtracting the above tag image from a ‘control’ image (no blood water tagged). Whole-brain ASL has been difficult because of a timing discrepancy in the so-called blood water ‘inversion time’ (TI), defined as the time between the labeling of the inflowing blood water and the acquisition of the image. As different slices are acquired at separate time points in standard 2D MRI acquisition routines, each slice corresponds to a different TI, complicating quantification. Recently, it was shown that whole-brain ASL measurements can be obtained by following the ASL blood labeling preparation with a 3D GRASE readout (Fernandez-Seara et al, 2005; Gunther et al, 2005). In 3D GRASE, a single excitation pulse is used, followed by a sub-second acquisition of the entire 3D volume simultaneously. Therefore, 3D GRASE ASL maintains an equal TI for all slices.

Vascular-space-occupancy has been proposed as a method for obtaining CBVw contrast without the use of a contrast agent (Lu et al, 2003). In VASO, the intravascular blood signal is nulled and the resulting image is derived from extravascular tissue. Vasodilatory CBV changes associated with increased neuronal activity are inferred from reductions in the detectable tissue signal. Whole-brain coverage with VASO remains problematic (Lu et al, 2004c; Scouten and Constable, 2007) as image acquisition in VASO must be performed at the precise null point of blood water, which is not possible in 2D routines that excite different imaging slices at different times. Therefore, this shortcoming is similar to the above-mentioned TI problem in ASL. As the 3D GRASE readout uses a single excitation pulse for the entire imaging volume, it may be possible to append the 3D GRASE readout to the VASO blood nulling preparation.

Quantification of Cerebral Blood Volume from Vascular-Space-Occupancy-Based Signal Changes

When BOLD, ASL-measured CBF, and VASO-measured CBV changes are obtained, it is possible to estimate CMRO₂. This approach has been largely outlined in the literature (Gu et al, 2006; Lu et al, 2004b); the salient aspects are reproduced here.

CBV change between baseline and an activated state can be estimated from the blood-nulled VASO signal change (Lu et al, 2003), or from VASO-Fluid Attenuation Inversion Recovery (VASO-FLAIR) images (blood water and CSF water both nulled simultaneously after double inversion) (Donahue et al, 2006). The VASO (V) or VASO-FLAIR (VF) spin-echo signal for an activated (act) or baseline (base) state can be written:

$$\begin{array}{rl}{S}_{V,VF}^j=& (1-X_{\mathrm{C}\mathrm{S}\mathrm{F}}^j)(C_{\mathrm{p}\mathrm{a}\mathrm{r}}-{\mathrm{C}\mathrm{B}\mathrm{V}}^j\cdot C_{\mathrm{b}})M_t{\mathrm{e}}^{-TE\cdot R_{2,\mathrm{t}}^j}\\ & +X_{\mathrm{C}\mathrm{S}\mathrm{F}}^j{C}_{\mathrm{C}\mathrm{S}\mathrm{F}}{M}_{\mathrm{C}\mathrm{S}\mathrm{F}}{\mathrm{e}}^{-TE\cdot R_{2,\mathrm{C}\mathrm{S}\mathrm{F}}}\end{array}$$

(1)

j=act or base.

Here, X_CSF is the voxel CSF fraction; C _par ≈0.89 mL water/mL parenchyma, C_b≈0.87 mL water/mL blood, and C_CSF≈1 mL water/mL CSF are the parenchyma, microvascular blood (Hct≈0.37), and CSF water densities, respectively (Herscovitch and Raichle, 1985; Zhao et al, 2007); CBV is the gray matter parenchyma CBV; M_t and M_CSF are the extravascular tissue and CSF longitudinal magnetizations, respectively; TE is the echo time; and R_2,t and R_2,CSF are the transverse relaxation rates of extravascular tissue and CSF water, respectively. Note that the CSF fraction, CBV, and R_2,t may change between baseline and activation (Donahue et al, 2006; Lu and van Zijl, 2005; Scouten and Constable, 2008).

We use a VASO-FLAIR preparation for simultaneous blood water and CSF nulling, as this approach is found (see Results) to have higher contrast-to-noise ratio (CNR) than conventional VASO. One important advantage of the VASO-FLAIR preparation is that CSF magnetization is nulled (M_CSF=0), thereby simplifying the VASO-FLAIR signal:

$$S_{VF}^j=(1-X_{\mathrm{C}\mathrm{S}\mathrm{F}}^j)(C_{\mathrm{p}\mathrm{a}\mathrm{r}}-CB{V}^j\cdot C_{\mathrm{b}})M_{\mathrm{t}}{\mathrm{e}}^{-TE\cdot R_{2,\mathrm{t}}^j}$$

(2)

The signal change between baseline and activation can therefore be written:

$$\frac{\mathrm{\Delta }{S}_{VF}}{S_{VF}^{\mathrm{b}\mathrm{a}\mathrm{s}\mathrm{e}}}=\frac{(1-X_{\mathrm{C}\mathrm{S}\mathrm{F}}^{\mathrm{a}\mathrm{c}\mathrm{t}})(C_{\mathrm{p}\mathrm{a}\mathrm{r}}-CB{V}^{\mathrm{a}\mathrm{c}\mathrm{t}}\cdot C_{\mathrm{b}})}{(1-X_{\mathrm{C}\mathrm{S}\mathrm{F}}^{\mathrm{b}\mathrm{a}\mathrm{s}\mathrm{e}})(C_{\mathrm{p}\mathrm{a}\mathrm{r}}-CB{V}^{\mathrm{b}\mathrm{a}\mathrm{s}\mathrm{e}}\cdot C_{\mathrm{b}})}{\mathrm{e}}^{-TE\cdot \mathrm{\Delta }{R}_{2,\mathrm{t}}}-1$$

(3)

It is then possible to solve explicitly for CBV^act:

$${\mathrm{C}\mathrm{B}\mathrm{V}}^{\mathrm{a}\mathrm{c}\mathrm{t}}=\frac{1}{C_{\mathrm{b}}}[C_{\mathrm{p}\mathrm{a}\mathrm{r}}-\frac{(1-X_{\mathrm{C}\mathrm{S}\mathrm{F}}^{\mathrm{b}\mathrm{a}\mathrm{s}\mathrm{e}})}{(1-X_{\mathrm{C}\mathrm{S}\mathrm{F}}^{\mathrm{a}\mathrm{c}\mathrm{t}})}(\frac{\mathrm{\Delta }{S}_{VF}}{S_{VF}^{\mathrm{b}\mathrm{a}\mathrm{s}\mathrm{e}}}+1)(C_{\mathrm{p}\mathrm{a}\mathrm{r}}-{\mathrm{C}\mathrm{B}\mathrm{V}}^{\mathrm{b}\mathrm{a}\mathrm{s}\mathrm{e}}\cdot C_{\mathrm{b}}){\mathrm{e}}^{TE\cdot \mathrm{\Delta }{R}_{2,\mathrm{t}}}]$$

(4)

As intravascular blood water is nulled in VASO-FLAIR experiments, only extravascular BOLD effects may cause a change in R_2,t. These extravascular effects may originate around small vessels (e.g., capillaries and venules) or larger draining veins. The 3D GRASE acquisition uses multiple spin-echo signal formations. In spin-echo sequences, extravascular BOLD effects around large vessels are minimal because of static refocusing (Yacoub et al, 2003); therefore, BOLD contrast should only consist of extravascular effects originating from small vessels. In light of this, we have assumed that ΔR_2,t≈0. $X_{\text{GSF}}^{\text{base}}\approx 0.107$ and $X_{\text{GSF}}^{\text{act}}\approx 0.101$ , measured in previous VASO-FLAIR studies (Donahue et al, 2006) at comparable spatial resolution, were assumed here. In Equation 4, $\mathrm{\Delta }{S}_{VF}/S_{VF}^{\text{base}}$ is measured, whereas the remaining parameters are assumed on the basis of literature values as defined above. Note that VASO and VASO-FLAIR are generally used to measure CBV reactivity (changes from an assumed baseline state) and we have not attempted to quantify baseline CBV here. Instead, a global gray matter parenchyma CBV^base=0.055 mL blood/mL parenchyma was assumed (Giovacchini et al, 2002). The implications of this assumption are addressed in the Discussion section.

Cerebral Metabolic Rate of Oxygen Consumption and Oxygen Extraction Fraction Quantification from Blood-Oxygenation-Level-Dependent, Vascular-Space-Occupancy, and Arterial Spin Labeling Signal Changes

A parenchymal voxel can be divided into a tissue (t) compartment and a microvascular blood compartment. The microvascular compartment contains arterioles (a), capillaries, and venules (v) in approximately a 21%, 33%, and 46% ratio, and is commonly further approximated in fractions of 30% arteriolar (f_a=0.3) and 70% venular (f_v=0.7) blood (van Zijl et al, 1998). Therefore, the BOLD gradient-echo signal (S_BOLD) can be compartmentalized:

$$\begin{array}{rl}{S}_{\mathrm{B}\mathrm{O}\mathrm{L}\mathrm{D}}^j=& S_{\mathrm{B}\mathrm{O}\mathrm{L}\mathrm{D},\mathrm{a}}^j+S_{\mathrm{B}\mathrm{O}\mathrm{L}\mathrm{D},\mathrm{v}}^j+S_{\mathrm{B}\mathrm{O}\mathrm{L}\mathrm{D},\mathrm{t}}^j\approx f_a{x}^j{M}_{\mathrm{a}}{\mathrm{e}}^{-R_{2,\mathrm{a}}^{\ast }\cdot TE}\\ & +f_v{x}^j{M}_v{\mathrm{e}}^{-R_{2,\mathrm{v}}^{\ast ,j}\cdot TE}+(1-x^j)M_{\mathrm{t}}{\mathrm{e}}^{-R_{2,\mathrm{t}}^{\ast ,j}\cdot TE}\end{array}$$

(5)

j = act or base, where M_a, M_v, and M_t are the longitudinal magnetizations for arteriolar, venular, and tissue water, respectively. The effective transverse relaxation rate of blood water in arterioles (R_2,a^∗) does not change between baseline and an activated state, as neither oxygenation nor hematocrit is expected to vary in the arteriolar compartment. The venular blood water (R_2,v^∗) and tissue water relaxation rate (R_2,t^∗) will change because of an increase in venous oxygenation (Y_v) associated with increased neuronal activity. Finally, x is the water fraction of blood within a voxel and can be expressed in terms of the CBV in dimensionless units of mL blood/mL parenchyma:

$$x^j=\frac{{\mathrm{C}\mathrm{B}\mathrm{V}}^j\cdot C_{\mathrm{b}}}{C_{\mathrm{p}\mathrm{a}\mathrm{r}}}$$

(6)

j=base or act.

Assuming a healthy CBV^base, CBV^act and the corresponding water fraction during an activated state can be calculated from the VASO-FLAIR signal change (Eq. 4).

Using Equation 5, the BOLD signal change (ΔS_BOLD/S_BOLD^base) between baseline and an activated state can be written explicitly:

$$\frac{\mathrm{\Delta }S}{S_{\mathrm{B}\mathrm{O}\mathrm{L}\mathrm{D}}^{\mathrm{b}\mathrm{a}\mathrm{s}\mathrm{e}}}=\frac{f_{\mathrm{a}}\cdot \mathrm{\Delta }x\cdot M_{\mathrm{a}}{\mathrm{e}}^{-R_{2,\mathrm{a}}^{\ast ,\mathrm{b}\mathrm{a}\mathrm{s}\mathrm{e}}\cdot TE}+f_v\cdot M_v(x^{\mathrm{a}\mathrm{c}\mathrm{t}}\cdot {\mathrm{e}}^{-R_{2,\mathrm{v}}^{\ast ,\mathrm{a}\mathrm{c}\mathrm{t}}\cdot TE}-x^{\mathrm{b}\mathrm{a}\mathrm{s}\mathrm{e}}\cdot {\mathrm{e}}^{-R_{2,\mathrm{v}}^{\ast ,\mathrm{b}\mathrm{a}\mathrm{s}\mathrm{e}}\cdot TE})+M_t{\mathrm{e}}^{-R_{2,\mathrm{t}}^{\ast ,\mathrm{b}\mathrm{a}\mathrm{s}\mathrm{e}}\cdot TE}[{\mathrm{e}}^{-\mathrm{\Delta }{R}_{2,\mathrm{t}}^{\ast }\cdot TE}(1-x^{\mathrm{a}\mathrm{c}\mathrm{t}})-(1-x^{\mathrm{b}\mathrm{a}\mathrm{s}\mathrm{e}})]}{f_{\mathrm{a}}\cdot x^{\mathrm{b}\mathrm{a}\mathrm{s}\mathrm{e}}\cdot M_{\mathrm{a}}{\mathrm{e}}^{-R_{2,\mathrm{a}}^{\ast }\cdot TE}+f_v\cdot x^{\mathrm{b}\mathrm{a}\mathrm{s}\mathrm{e}}\cdot M_v{\mathrm{e}}^{-R_{2,\mathrm{v}}^{\ast ,\mathrm{b}\mathrm{a}\mathrm{s}\mathrm{e}}\cdot TE}+(1-x^{\mathrm{b}\mathrm{a}\mathrm{s}\mathrm{e}})\cdot M_{\mathrm{t}}{\mathrm{e}}^{-R_{2,\mathrm{t}}^{\ast ,\mathrm{b}\mathrm{a}\mathrm{s}\mathrm{e}}\cdot TE}}$$

(7)

where Δx=(x^act–x^base). The remaining unknown parameters above are $R_{2,v}^{\ast ,\text{act}}$ , $R_{2,v}^{\ast ,\text{base}}$ , R_2,a^∗, and ΔR_2,t^∗. From Yablonskiy and Haacke (Yablonskiy and Haacke, 1994):

$$\begin{array}{rl}\mathrm{\Delta }{R}_{2,\mathrm{t}}^{\ast }& \approx f_v\gamma B_0\frac{4}{3}\pi \mathrm{\Delta }\chi Hct[{\mathrm{C}\mathrm{B}\mathrm{V}}^{\mathrm{a}\mathrm{c}\mathrm{t}}(1-Y_{\mathrm{v}}^{\mathrm{a}\mathrm{c}\mathrm{t}})\\ & -CB{V}^{\mathrm{b}\mathrm{a}\mathrm{s}\mathrm{e}}(1-Y_v^{\mathrm{b}\mathrm{a}\mathrm{s}\mathrm{e}})]\end{array}$$

(8)

Here, γ=42.58 MHz/T is the water proton gyromagnetic ratio, B₀=3.0T is the field strength, Δχ=0.3 ppm (Zhao et al, 2007) is the susceptibility difference between fully oxygenated and fully deoxygenated blood, Hct≈0.37 is the approximate hematocrit in the microvasculature (Zhao et al, 2007), and Y_v is the venous oxygenation.

R_2,i^∗ (i=arteriole or venule) depends on Hct and can be determined from blood oxygenation (Y_i) according to (3.0T; extrapolated for Hct=0.37) (Zhao et al, 2007):

$$R_{2,i}^{\ast ,j}=16.6+99.6(1-Y_i{)}^2$$

(9)

j=act or base.

R_2,v^∗,act is the only remaining unknown parameter that can be rewritten in terms of $Y_v^{\text{act}}$ (Eq. 9) and calculated from the BOLD signal change (Eq. 7). The relationship between Y_v, oxygen extraction fraction (OEF) and Y_a is well known and can be expressed as (van Zijl et al, 1998):

$$1-Y_v^j=1-Y_{\mathrm{a}}+OE{F}^j{Y}_{\mathrm{a}}$$

(10)

j=act or base.

Therefore, once $Y_v^{\text{act}}$ is calculated, it is possible to quantify OEF.

This calculated OEF can be written in terms of the CMRO₂ and CBF according to:

$$OEF=\frac{\mathrm{o}\mathrm{x}\mathrm{y}\mathrm{g}\mathrm{e}\mathrm{n}\mathrm{c}\mathrm{o}\mathrm{n}\mathrm{s}\mathrm{u}\mathrm{m}\mathrm{p}\mathrm{t}\mathrm{i}\mathrm{o}\mathrm{n}}{\mathrm{o}\mathrm{x}\mathrm{y}\mathrm{g}\mathrm{e}\mathrm{n}\mathrm{d}\mathrm{e}\mathrm{l}\mathrm{i}\mathrm{v}\mathrm{e}\mathrm{r}\mathrm{y}}=\frac{{\mathrm{C}\mathrm{M}\mathrm{R}\mathrm{O}}_2}{\mathrm{C}\mathrm{B}\mathrm{F}\cdot Y_{\mathrm{a}}\cdot H{b}_{\mathrm{T}\mathrm{o}\mathrm{t}}}$$

(11)

where Hb_Tot is the total hemoglobin content. Under conditions of constant arterial oxygen content (Y_a·Hb_Tot=Constant), the CMRO₂ change (ΔCMRO₂/CMRO₂^base) associated with a change between baseline and an activated state can be expressed:

$$(1+\frac{\mathrm{\Delta }OEF}{OE{F}^{\mathrm{b}\mathrm{a}\mathrm{s}\mathrm{e}}})(1+\frac{\mathrm{\Delta }CBF}{CB{F}^{\mathrm{b}\mathrm{a}\mathrm{s}\mathrm{e}}})-1=\frac{\mathrm{\Delta }{\mathrm{C}\mathrm{M}\mathrm{R}\mathrm{O}}_2}{{\mathrm{C}\mathrm{M}\mathrm{R}\mathrm{O}}_2^{\mathrm{b}\mathrm{a}\mathrm{s}\mathrm{e}}}$$

(12)

Therefore, knowledge of the OEF change (ΔOEF/OEF^base) used in conjunction with the ASL-measured ΔCBF/CBF^base can be used to calculate ΔCMRO₂/CMRO₂^base.

Experiment

All subjects (n=8; age=29±6 years; right-handed) provided informed, written consent and were scanned on a 3.0T scanner (Siemens, Erlangen, Germany) sequentially with gradient-echo BOLD EPI, 3D GRASE ASL, 3D GRASE VASO, and 3D GRASE VASO-FLAIR. One subject failed to complete the study because of discomfort and was excluded from the analysis.

Gradient echo BOLD contrast cannot be isolated with the 3D GRASE imaging module. Therefore, an EPI module was used for BOLD measurements. In VASO, a single inversion pulse was used to null blood water (Lu et al, 2003), whereas in VASO-FLAIR a second inversion pulse was added to null blood water and CSF water simultaneously (Donahue et al, 2006). Note that with VASO-FLAIR, a TR≥5 s is required to maintain an adequate signal-to-noise ratio (SNR), as steady-state tissue magnetization is small (∼12% of M₀) at the time of excitation. Therefore, a long TR=5 s VASO-FLAIR sequence (double inversion) provides a similar residual tissue magnetization as a short TR=2 to 3 s VASO (single inversion) experiment. Common scan parameters were TE=40 ms, bandwidth=2004 Hz/pixel, FOV=240 × 240 mm, imaging matrix=64 × 64, 22 slices, slice thickness=3.8 mm, spatial resolution=3.8 × 3.8 × 3.8 mm³. Sequence-specific scan parameters were BOLD 2D EPI: TR=3000 ms; ASL 3D GRASE: pulsed ASL with background suppression, TR/TI=2500/1600; VASO 3D GRASE: TR/TI=5000/1054 ms; VASO-FLAIR 3D GRASE: TR/TI1/TI2=5000/2256/737 ms. TI for VASO and VASO-FLAIR was chosen assuming a blood T₁=1624 ms. It is well known that blood T₁ will vary with oxygenation and Hct (Lu et al, 2004a). A recent study found that the T₁ variation over a Hct range of 0.37 to 0.42 and an oxygenation range of 0.69 to 0.92 produced approximately a 10% variation in blood T₁ and 8% variation in TI choice at TR=5 s, and led to no statistical difference in the VASO-quantified CBV change (Donahue et al, 2009b). Therefore, to be consistent with previous VASO studies, we have used a blood T₁=1624 ms.

The functional paradigm consisted of a 60s/30 s off/on flashing blue-yellow checkerboard (f=8 Hz) repeated four times. The order of BOLD, VASO, VASO-FLAIR, and ASL scans was randomized between subjects to minimize habituation trends. During visual stimulation, volunteers were instructed to move a joystick unilaterally (right hand; f=1 Hz), which resulted in two different neuronal tasks, in separate brain regions, being performed simultaneously.

Post-Processing

Data were corrected for motion and baseline drift. Spatial resolution and slice orientation were matched across scans; BOLD images (2D EPI) were co-registered and warped to the 3D GRASE images using FSL routines (Jenkinson and Smith, 2001). Subsequently, time courses in voxels meeting activation from a statistical z-test (Matlab, The MathWorks) in BOLD (P<0.05; SNR>50), ASL (P<0.05; SNR>2.5), VASO (P<0.05; SNR>50), and VASO-FLAIR (P<0.05; SNR>50) were calculated. Time courses, SNR, CNR, and the number of voxels meeting the activation criteria were recorded for each scan. The SNR was defined as the mean baseline signal divided by the standard deviation of the baseline signal and therefore was effectively a measure of the time-course SNR. The CNR was defined as the product of the time-course SNR and the absolute signal change. In ASL, SNR and CNR were calculated from the difference (ΔM) images. To control for partial volume variations, time courses were calculated for only the subgroup of voxels meeting activation criteria in all sequences. Activated voxels in the visual cortex and motor cortex were analyzed separately to allow for reactivity to be assessed between regions. In VASO and VASO-FLAIR, vasodilatory signal changes are negative; to make comparison with BOLD and CBFw courses more direct, VASO and VASO-FLAIR time courses were inverted.

For CMRO₂ quantification, CBF, CBV, and BOLD changes were recorded for all voxels meeting the activation criteria in the visual or motor cortex for each volunteer, and CMRO₂ was calculated on a subject-by-subject basis.

Results

Experiment

Figure 1 shows representative activation maps for CBVw VASO (a) and CBVw VASO-FLAIR (b) using the whole-brain 3D GRASE acquisition. VASO-FLAIR activation was found to be well localized to the visual and motor cortex, whereas VASO activation was less specific. More activated voxels were found using VASO-FLAIR (2193±461) than VASO (1616±272); however, this finding was not statistically significant (P=0.172). This high P value was due to nonspecific activation in regions lying outside the visual and motor cortex in the VASO acquisition. When only voxels in the motor and visual cortex were analyzed, it was found that VASO-FLAIR gave statistically more (P=0.001) activated voxels than did VASO. Of the voxels that did meet the activation criteria, time courses were found to give only small and variable changes in VASO images compared with VASO-FLAIR images (Figure 1C and 1D). Given the above findings, subsequent CBV analysis was carried out using activated voxels in VASO-FLAIR 3D GRASE images.

Figure 1

Representative VASO 3D GRASE (A) and VASO-FLAIR 3D GRASE (B) activation maps. Time courses of overlapping voxels in visual (C) and motor (D) cortex for VASO-FLAIR (blue) and VASO (gray). VASO and VASO-FLAIR time courses have been inverted; error bars represent standard error over seven subjects and black lines represent stimulus periods.

Figure 2 shows representative BOLD (a), CBFw ASL (b), CBVw VASO-FLAIR (c), and overlapping voxel (d) activation maps for a representative subject (central 12 slices shown). Note that BOLD, CBFw, and CBVw maps show good localization to the visual cortex and left motor cortex as expected for the task. Corresponding visual and motor time courses were calculated for the subgroup of voxels meeting the activation criteria in all three scans (Figure 2D). The BOLD (Figure 2E), CBVw (Figure 2F), and CBFw (Figure 2G) responses are consistently smaller in the motor cortex compared with the visual cortex (BOLD: P=0.005; CBVw: P=0.031; CBFw: P=0.006). The BOLD post-stimulus undershoot is larger (visual: −0.9%±0.2%; motor: −0.2%±0.1%; P=0.003) and endures longer (visual: 47±6 s; motor: 31±4 s; P=0.006) in the visual cortex compared with the motor cortex. In addition, although magnitude CBV and CBF changes are different in the visual and motor cortex, the time for return to baseline in the visual (CBVw: 20±3 s; CBFw: 20±3 s) and motor (CBVw: 21±3 s; CBFw: 23±5 s) cortex is not statistically different (P>0.05). The BOLD (Figure 2H), CBVw (Figure 2I), and CBFw (Figure 2J) show the time courses normalized for maximum intensity.

Figure 2

Representative BOLD (A), ASL (B), VASO-FLAIR (C) and overlapping voxels (D) for a single subject. Time courses (n=7) for overlapping voxels (D) for BOLD (E), VASO-FLAIR, (F) and ASL (G) are shown below, separately segmented into visual (blue) and motor (gray) voxels. Corresponding normalized BOLD (H), CBVw (l) and CBFw (J) time courses are shown in the bottom row. The gray box indicates the task period. Note the consistently larger signal changes in the visual cortex in all techniques. In addition, notice that the CBF and CBV courses return to baseline at the same time in both the visual and motor cortex, where the BOLD post-stimulus change endures much longer in the visual cortex than in the motor cortex.

Table 1 shows the SNR, CNR, and signal changes for each scan. In addition, P values from a paired t-test comparing the visual and motor values are shown. The SNR is higher in CBVw VASO than in CBVw VASO-FLAIR (P<0.001) because less residual tissue signal remains in VASO-FLAIR.

Table 1

SNR, CNR, and signal changes in multi-modal fMRI

	Visual	Motor	P
BOLD
SNR	136±25	155±15	0.259
∣CNR∣	2.4±0.4	2.2±0.1	0.491
ΔS/S % ∗	1.9±0.3	1.5±0.3	0.005
ASL
SNR	4±1	4±1	0.555
∣CNR∣ ∗	2.4±0.2	1.8±0.1	0.013
ΔS/S % ∗	63±12	46±11	0.006
VASO
SNR ∗	141±10	125±13	0.033
∣CNR∣	0.8±0.2	0.1±0.2	0.113
ΔS/S %	−0.6±0.4	−0.2±0.6	0.239
VASO-FLAIR
SNR ∗	78±3	91±4	0.042
∣CNR∣	1.2±0.1	1.1±0.1	0.142
ΔS/S % ∗	−1.6±0.3	−1.2±0.3	0.031

ASL=arterial spin labeling; BOLD=blood-oxygenation-level-dependent; CNR=contrast-to-noise ratio; fMRI=functional MRI; SNR=signal-to-noise ratio; VASO=vascular-space-occupancy.

The ASL SNR represents the SNR in the subtracted image. Table values were calculated in the subgroup of overlapping voxels that met activation criteria in BOLD, VASO-FLAIR, and ASL visual (99±30) and motor (70±22) cortex. P values correspond to significance level (paired t-test) between visual and motor values. ^∗Significance at P≤0.05.

CBF changes in the visual and motor cortex were (mean±s.d.) 63%±12% and 46%±11%, respectively. The CBV changes in the motor and visual cortex were (mean±s.d.) 17%±7% and 8%±3%, respectively. Hemodynamic changes were found to be significantly higher in the visual cortex compared with the motor cortex for both CBF (P=0.006) and CBV (P=0.031). It may be the case that the baseline CBV is lower in the motor cortex compared with the visual cortex (Giovacchini et al, 2002); therefore, we repeated the CBV quantification step in the motor cortex with a smaller baseline CBV of 0.05 mL/mL (compared with 0.055 mL/mL). In this case, the CBV change increased slightly to 9±4%, but was still significantly (P=0.048) less than the CBV change in the visual cortex.

Cerebral Metabolic Rate of Oxygen Consumption and Oxygen Extraction Fraction Quantification

The above results show the relationship between CBF, CBV, and BOLD, which when combined can be used to estimate CMRO₂.

Simulations were performed to show how changes in CBF and CBV would influence OEF, CMRO₂, and Y_v. Figure 3A shows the relationship between the BOLD signal change and Y_v^act for varying possible CBV changes assuming a healthy Y_v^base=0.61. For vasodilatory CBV changes, BOLD signal is negative when Y_v^base=Y_v^act. However, as Y_v^act increases beyond Y_v^base, the BOLD signal changes become positive (as is generally observed for neuronal tasks). In Figure 3B, the dependence of ΔOEF/OEF^base on Y_v^act for varying Y_v^base is shown. As expected, ΔOEF/OEF^base is negative for Y_v^act>Y_v^base. Figure 3C shows ΔCMRO₂/CMRO₂^base versus ΔOEF/OEF^base for varying CBF changes. Note that for physiological ΔOEF/OEF^base (−20% to −30%), ΔCMRO₂/CMRO₂^base is only positive for large CBF changes (>40%).

Figure 3

Simulations performed to show how changes in CBF and CBV will influence OEF, CMRO₂, and Y_v. (A) The relationship between the BOLD signal change and Y_v^act for varying possible CBV changes (assuming Y_v^base=0.61). (B) The dependence of ΔOEF/OEF^base on Y_v^act for varying physiological Y_v^base possibilities. (C) ΔCMRO₂/CMRO₂^base versus ΔOEF/OEF^base for varying CBF changes.

The quantification procedure incorporates the above model with the BOLD, VASO-FLAIR (CBVw), and ASL (CBFw) data and relies on several assumed parameters. Table 2 shows how the calculated OEF and CMRO₂ values will vary for a realistic error in each of the assumed parameters. The error range surrounding a Y_v, OEF, and CMRO₂ change between baseline and activation of approximately 0.17, −0.27, and 0.12, respectively, was calculated. As can be seen from Table 2, CMRO₂ changes generally have the largest uncertainty range. Note that the quantification is especially sensitive to the oxygenation levels (Y_v), baseline CBV, and water densities (C_par; C_b).

Table 2

Errors associated with parameter estimations

Parameter	Value assumed±error considered	ΔY v /Y v base	ΔOEF/OEF base	ΔCMRO 2 /CMRO 2 base
Y a	0.98±0.02	0.16 to 0.17	−0.26 to −0.29	0.10 to 0.15
Y v	0.61±0.09	0.13 to 0.22	−0.25 to −0.33	0.05 to 0.16
Hct	0.37±0.04	0.16 to 0.17	−0.27 to −0.28	0.12 to 0.14
CBV base	0.055±0.01 mL/mL	0.15 to 0.18	−0.25 to −0.30	0.09 to 0.16
fa/fv	0.43±0.04	0.16 to 0.17	−0.27	0.13
C par	0.89±0.09 mL/mL	0.16 to 0.17	−0.26 to −0.29	0.10 to 0.15
C b	0.87±0.09 mL/mL	0.15 to 0.18	−0.25 to −0.29	0.11 to 0.16
R 2,a	16.6±1.66 s−1	0.16 to 0.17	−0.27 to −0.28	0.12 to 0.13
R 2,v	31.8±3.18 s−1	0.14 to 0.20	−0.23 to −0.33	0.05 to 0.20
R 1,a	0.601±6.01 s−1	0.16 to 0.17	−0.27 to −0.28	0.12 to 0.13
R 1,v	0.631±6.31 s−1	0.16 to 0.17	−0.27 to −0.28	0.11 to 0.14

A representative 1.6%, −1.4%, and 54% BOLD, VASO-FLAIR and ASL signal change has been assumed, leading to a ΔY_v/Y_v^base=0.17, ΔOEF/OEF^base=−0.27, ΔCMRO₂/CMRO₂^base=0.12. Errors associated with a 10% possible error in each parameter are shown, separately for ΔY_v/Y_v^base, ΔOEF/OEF^base, and ΔCMRO₂/CMRO₂^base_. Note that error is highest for oxygenation uncertainties (Y_a, Y_v) and smallest for uncertainties in the MR relaxation rates. Error is also generally higher in ΔCMRO₂/CMRO₂^base estimation than in ΔOEF/OEF^base estimation.

ΔOEF/OEF^base is found to be −0.31±0.03 (visual) and −0.23±0.04 (motor); ΔCMRO₂/CMRO₂^base is found to be 0.13±0.11 (visual) and 0.12±0.13 (motor). Following neuronal activity, during the maximal period of the BOLD undershoot, ΔCMRO_2/CMRO₂^base=0.11±0.07 (visual) and ΔCMRO_2/CMRO₂^base=0.03±0.04 (motor). Figure 4 shows the OEF change versus time. Note that following cessation of the applied stimulus, the OEF change in the visual cortex becomes positive, whereas the OEF change in the motor cortex approaches its baseline level. During this period, the CBF and CBV have returned to baseline as well.

Figure 4

The calculated ΔOEF/OEF^base time course for the visual and motor cortex. The gray box demarcates the task period. Note that following cessation of the task, the OEF change becomes positive in the visual cortex. During this period, the CBF and CBV have returned to baseline levels (Figure 2).

Table 3 shows the quantified hemodynamic and metabolic parameters for separate visual and motor tasks.

Table 3

Quantified hemodynamic and metabolic parameters for separate visual and motor tasks

	ΔCBF/CBF base	ΔCBV/CBV base	ΔY v /Y v base	ΔOEF/OEF base	ΔCMRO 2 /CMRO base
Visual	0.63±0.12	0.17±0.07	0.18±0.02	−0.31±0.03	0.13±0.11
Motor	0.46±0.11	0.08±0.03	0.14±0.03	−0.23±0.04	0.12±0.13
P	0.006	0.031	<0.001	<0.001	0.731

CBF=cerebral blood flow; CBV=cerebral blood volume; CMRO₂=cerebral metabolic rate of oxygen consumption; OEF=oxygen extraction fraction.

Values are mean±s.d. over seven subjects. Therefore, the error represents the variation in the value over the seven subjects. P values denote significance level (paired t-test) between visual and motor values. Possible errors associated with the fitting routine is shown separately in Table 2.

Discussion

We explored the individual hemodynamic constituents of the BOLD signal by performing whole-brain CBFw and CBVw functional MRI (fMRI) in combination with BOLD fMRI. The results show that the CBV changes measured using VASO 3D GRASE with CSF suppression (VASO-FLAIR) are consistent with the independently measured CBF and BOLD changes.

Vascular-Space-Occupancy-FLAIR 3D GRadient-And-Spin-Echo

The primary goal of this study was to assess whether CBVw VASO-FLAIR with the 3D GRASE readout could be used to assess CBV adjustments in different brain regions simultaneously. We found larger CBV adjustments in the visual cortex (ΔCBV=17%±7%) compared with the motor cortex (ΔCBV=8%±3%), consistent with a range of previous studies (Vafaee and Gjedde, 2000; Vafaee and Gjedde, 2004). Using ¹¹CO PET in humans, ΔCBV=21%±5% in response to an 8 Hz flashing checkerboard was reported, which is consistent within the error of this study (Ito et al, 2001). Using iron oxide contrast media in alpha-chloralose-anesthetized rats, van Bruggen et al. reported an average CBV change of 9.1%±1.5% in rat brain in response to sensory stimulus (van Bruggen et al, 1998). Using two-photon microscopy in an anesthetized rat model, Hillman et al. reported a CBV change of 16.8%±5.5% in response to somatosensory stimulation, which was reported to occur in the arterioles (Hillman et al, 2007). Mandeville et al. found a CBV change of 17%±2% in response to forepaw stimulation in anesthetized rats (Mandeville et al, 1999). Scouten and Constable quantified CBV using a multi-slice VASO approach with multiple acquisition with global inversion cycling (Lu et al, 2004c) and reported a CBV change of 17%±8% in the visual cortex (Scouten and Constable, 2007). Initially, very large CBV changes (40% to 60%) were reported with VASO; however, with improvement in the understanding of the contrast mechanism, specifically by accounting for CSF changes and inflow effects (Donahue et al, 2009a; Scouten and Constable, 2008), magnitude VASO signal changes have been reported in the range of 1% to 2%, with a corresponding CBV change of 5% to 20%. Therefore, the VASO-FLAIR 3D GRASE CBV changes are consistent with much of the CBV changes in the literature. Small variations that remain could be due to variation in species (humans versus rats), absence or presence of anesthesia, spatial resolution, and baseline CBV.

The CBVw VASO sequences without CSF nulling showed poor sensitivity to the functionally eloquent regions associated with the task (Figure 1). As this sensitivity improved considerably with VASO-FLAIR (simultaneous blood and CSF nulling), we attributed this to CSF contamination in the voxel in the case of VASO. This CSF contamination is especially detrimental in the 3D GRASE readout, as the long refocusing train will lead to artifacts in voxels that partial volume with tissues of significantly different T₂ (e.g., CSF). CSF contamination is known to influence quantification of EPI-based VASO signal changes (Donahue et al, 2006; Scouten and Constable, 2008); however, CSF did not have such a detrimental influence on sensitivity, as was found with the 3D GRASE module.

Several technical advantages of the CBVw VASO-FLAIR 3D GRASE approach should be considered. First, as the 3D GRASE readout module uses a single excitation radiofrequency pulse per TR, an identical TI is maintained in all slices of the 3D volume, thereby maintaining blood water nulling throughout the long echo train. Second, the 3D GRASE readout is fundamentally a spin-echo sequence. Extravascular BOLD effects arising from large diameter veins are reduced in spin echo, and therefore BOLD sensitivity in 3D GRASE ASL and VASO-FLAIR sequences is reduced relative to gradient echo. Third, the VASO-FLAIR sequence does not require the injection of contrast agents and is therefore entirely noninvasive. This could make VASO-FLAIR 3D GRASE a useful approach for tracking CBV reactivity over time where restrictions on intravascular contrast agent doses may disallow invasive approaches.

Several technical drawbacks of the proposed CBVw approach should also be considered. First, as VASO-FLAIR uses two inversion pulses, the residual tissue magnetization at the time of imaging is lower than in VASO. This observation was reflected in the lower SNR of VASO-FLAIR compared with VASO. However, CNR of VASO-FLAIR 3D GRASE was higher than that of VASO 3D GRASE. Second, the 3D GRASE readout contains a long echo train (∼400 ms) that leads to some signal smearing in the slice direction. This effect is not likely to confound the results in this study as we took care to orient the slices such that the motor and visual cortexes were in different planes. These smearing phenomena can be reduced by using shorter echo train readout times (e.g., fewer slices), refocusing pulses with <180° flip angle, or by incorporating parallel imaging (Feinberg et al, 2009). Furthermore, the reduction in readout time associated with parallel imaging will likely offset any reduction in SNR attributed to the parallel imaging module. Regarding the ASL measurements, both CBF and arterial transit time (ATT), defined as the time for the labeled blood water to reach the capillary exchange site in the imaging slice, can influence quantification (Gonzalez-At et al, 2000). We assumed that ATT did not change between baseline and the activated period. This assumption was based on separate functional ASL measurements in the visual and motor cortex performed at multiple TI values (250 to 2250 ms). We found only a small change in ATT of ∼3% and a similar CBF value to those reported in this single TI study. At the relatively long TI=1600 ms where experiments in this study were performed, an ATT reduction of 3% will have only a small influence on the CBF change (<4%), which is within the error of the values we report.

The OEF and CMRO₂ quantification procedure shows promise, yet possible errors in the assumptions do influence the absolute numbers (Table 2). However, arterial oxygenation (Y_a) can easily be measured with pulse oximetry. Recently, it was shown that Y_v can be estimated noninvasively using a new tissue-relaxation-under-spin-tagging MRI approach (Lu and Ge, 2008). Finally, baseline CBV can in principle be measured, most reliably with intravascular contrast agents. Thus, by incorporating independent measures of Y_v, Y_a, and CBV^base, the error associated with the CMRO₂ quantification procedure can be reduced. Such measurements would add an additional 5 to 10 mins to the total scan duration.

Blood-Oxygenation-Level-Dependent Time-course Dynamics

Following elevated neuronal activity, VASO-FLAIR-measured CBV and ASL-measured CBF return to baseline at approximately the same time, irrespective of the two functional regions studied (Figure 2F and 2G). During this period, the BOLD post-stimulus undershoot is larger and endures longer in the visual than in the motor cortex. This observation is consistent with a persisting change in OEF and CMRO₂ (Figure 4).

There is considerable debate in the MR literature regarding the physiological origins of the BOLD post-stimulus undershoot. This undershoot has been attributed to a slow return to baseline of venous CBV (Buxton et al, 1998), as well as to persisting elevated CMRO₂ (Donahue et al, 2008; Frahm et al, 2008; Harshbarger and Song, 2008; Lu et al, 2004b). The majority of studies that relate elevated metabolism to the BOLD undershoot were conducted in the visual cortex, whereas much of the literature linking elevated CBV with the undershoot was conducted in the somatosensory and motor cortex where metabolic increases may be smaller (Donahue et al, 2008; Frahm et al, 2008; Harshbarger and Song, 2008; Jin and Kim, 2008; Lu et al, 2004b; Mandeville et al, 1999; Uludag, 2008). Therefore, it is possible that CBF/CBV and CMRO₂ mismatches following elevated neuronal activity are more detectable in brain regions that are very energetically demanding (e.g., the visual cortex). In addition, Yacoub et al. showed using MION in an anesthetized cat model at 9.4T that different physiological mechanisms may govern the BOLD post-stimulus undershoot (Yacoub et al, 2006). Specifically, they found that CBV in surface vessels returned to baseline quickly after cessation of a visual stimulus, whereas CBV in tissue voxels returned to baseline more slowly. At 9.4T and high spatial resolution (2 × 2 × 2 mm³), Jin and Kim showed a cortical layer dependence of CBV and CBF dynamics, showing that flow-volume coupling is both spatially and temporally dependent in the brain (Jin and Kim, 2008). The whole-brain protocol proposed here might be useful for studying such effects in the future.

It should be noted that both the CBFw and CBVw time courses return to baseline after 15 to 20 s regardless of the magnitude of the CBF and CBV change. Conversely, the BOLD post-stimulus undershoot returns to baseline more quickly in the motor cortex than in the visual cortex, and this return is equivalent to the CBF and CBV return. Therefore, the BOLD post-stimulus undershoot clearly varies regionally in the brain in response to these two separate tasks, and yet the post-stimulus CBF and CBV responses are largely unchanging. The CBV weighting in VASO may be more sensitive to pre-capillary CBV changes than to post-capillary CBV changes. This is due to blood/tissue water exchange influencing the nulling of blood water in the veins. Therefore, it is possible that small CBV adjustments in the venous vasculature are less detectable with VASO. However, the VASO signal changes are consistent with the BOLD signal changes during neuronal activity and the BOLD contrast is clearly derived from capillary and post-capillary blood oxygenation changes. Therefore, VASO is likely sensitive to some venous CBV effects.

Finally, an ability to measure flow-metabolism coupling and individual CBF, CBV, and CMRO₂ reactivity may have a role in the clinical management of patients with cerebrovascular disease; however, this has not yet been shown conclusively. Rossini et al. found consistent magnetoencephalography signals in both the affected and unaffected hemispheres of patients with cerebrovascular disease, but a lack of BOLD fMRI signal changes in a subgroup of patients (Rossini et al, 2007). This was attributed to the altered vasomotor reactivity with the conclusion that many mental and physical activities for patients with cerebrovascular disease may appear normal, but could be hemodynamically fatiguing at a brain tissue level. The CBF, CBV, and CMRO₂ reactivity will influence the BOLD signal in different ways, and certain combinations of abnormal reactivity could elicit no BOLD response as was found in the above study. Others have reported normal neuronal activity in patients with carotid artery occlusion despite reduced CBF (Powers et al, 1988). Similar studies correlating neuronal activity with BOLD fMRI responses have found reduced BOLD signal changes in the face of preserved neuronal activity (Bilecen et al, 2002; Rother et al, 2002). An ability to characterize individual constituents of BOLD contrast may therefore be useful in distinguishing between patients with no or limited hemodynamic reactivity and patients with hemodynamic reactivity that fails to elicit a BOLD response. Similarly, measuring regional changes in perfusion and metabolism may prove superior over BOLD-fMRI when investigating post-stroke cerebral plasticity. As patients recover, new cortical areas might be recruited to take over the task of damaged areas. As stroke is often caused by atherosclerosis affecting several vessels, recruited areas potentially suffer from less vascular reserve capacity, and thus may elicit only a weak BOLD signal.

Conclusion

By combining the CBVw VASO-FLAIR magnetization preparation with a 3D GRASE readout, we show that it is possible to obtain whole-brain CBVw functional images without the use of intravascular contrast agents. Furthermore, the CBV values obtained from this new approach were consistent with separately measured CBF and BOLD changes. The goal of this study, to generate a whole-brain imaging protocol for identifying the different hemodynamic constituents of BOLD contrast, has been tested by identifying known CBV adjustments and flow-metabolism variations in the visual and motor cortices.