Measurement of Parenchymal Extravascular R₂* and Tissue Oxygen Extraction Fraction Using Multi-echo VASO MRI at 7 Tesla

Abstract

Parenchymal extravascular R₂* is an important parameter for quantitative blood-oxygenation-level-dependent (BOLD) studies. Total and intravascular R₂* values and changes in R₂* values during functional stimulations have been reported in a number of studies. The purpose of this study was to measure absolute extravascular R₂* values in human visual cortex and estimate the intra- and extravascular contributions in the BOLD effect at 7T. Vascular-space-occupancy (VASO) MRI was employed to separate out the extravascular tissue signal. Multi-echo VASO and BOLD fMRI with visual stimulation were performed at 7T for R₂* measurement at a spatial resolution of 2.5×2.5×2.5 mm³ in healthy volunteers (n = 6). The ratio of changes in extravascular and total R₂* (ΔR₂*) was used to estimate the extravascular fraction of the BOLD effect. Extravascular R₂* were found to be 44.66 ± 1.55 s⁻¹ and 43.38 ± 1.51 s⁻¹ (mean ± SEM, n = 6) at rest and activation, respectively, in human visual cortex at 7T. The extravascular BOLD fraction was estimated to be 91 ± 3%. Parenchymal oxygen extraction fraction (OEF) during activation was estimated to be 0.24 ± 0.01 based on the R₂*measurements, indicating an approximately 37% decrease compared to OEF at rest.

Introduction

Parenchymal extravascular R₂* is an important parameter for quantitative blood-oxygenation-level-dependent (BOLD) studies. It is well known that the BOLD effect increases with field strength, and the BOLD signal from extravascular tissue becomes more dominant at higher field, as the intravascular BOLD signal contribution is significantly reduced due to the faster R₂* decay of venous blood [1-9]. The relative contribution of the extravascular BOLD effect can be estimated using the ratio of extravascular and total R₂* changes during neuronal activation. Furthermore, R₂* changes can also be used to estimate changes in physiological parameters such as venous oxygenation (Y_v) and tissue oxygen extraction fraction (OEF) during brain activation.

Total and intravascular R₂* values at various field strengths have been reported in a number of studies in animals and humans [2, 7-22]. However, reports on extravascular R₂* values remain scarce in the literature (Table 1), as it is not trivial to measure mainly due to the difficulty to separate out the extravascular and intravascular signals in parenchyma. Duong et al. [1] used diffusion gradients to suppress the intravascular BOLD signal in order to investigate the microvascular contribution in the BOLD effects at 4T and 7T. Van der Zwaag et al. [8] excluded the intravascular BOLD effects in large veins using high resolution anatomical scans and reported R₂* values in human motor cortex at 1.5, 3 and 7T at a spatial resolution of 1×1×3mm³. Lu et al. [22] determined parenchymal extravascular R₂* values at 1.5T and 3T using multi-echo vascular-space-occupancy (VASO) MRI, which eliminates the intravascular signal based on different T₁ relaxation times of blood and tissue [23]. Donahue et al. [13] employed bipolar crushing gradients to suppress fast flowing blood signal (intravascular), and measured extravascular R₂* change (ΔR₂*) during visual stimulation in human brain at 1.5T, 3T and 7T. To date, however, the combined BOLD/VASO method for determination of parenchymal extravascular R₂* values [22] has not yet been used in human brain at 7T.

Table 1.

Comparison of extravascular and total parenchymal R₂* values at 1.5T, 3T and 7T.

		Resolution (mm3)	Extravas. R2,rest* (s−1)	Extravas. ΔR2* (s−1)a	Total R2,rest* (s−1)	Total ΔR2* (s−1)	Extravas. ΔR2* fraction (%)b
1.5 T	Ref. [22] Ref. [13] Ref. [8]	2×2×2 3.5×3.5×3.5 1×1×3	16.14±0.64 --- ---	−0.25±0.02 −0.28±0.07 ---	16.78±0.65 --- 11.6±0.3	−0.57±0.10 −0.61±0.10 −0.51±0.06	47±7 45±13 ---
3.0 T	Ref. [22] Ref. [13] Ref. [8]	2×2×2 3.5×3.5×3.5 1×1×3	21.15±0.66 --- ---	−0.38±0.05 −0.52±0.07 ---	22.06±0.84 --- 18.1±0.4	−0.58±0.09 −0.74±0.05 −0.98±0.08	67±6 70±11 ---
7.0 T	Ref. [13] Ref. [8]	3.5×3.5×3.5 1×1×3	--- ---	−1.25±0.11 ---	--- 30.8±1.0	−1.37±0.34 −2.55±0.22	91±11 ---
	This study	2.5×2.5×2.5	44.66±1.55	−1.27±0.14	45.05±1.34	−1.40±0.16	91±3

mean ± standard error (SEM)

^aΔR₂* = R_2,act* − R_2,rest*.

^bExtravas. ΔR₂* fraction = 100 × (Extravas. ΔR₂* / Total ΔR₂*) %.

In this study, we applied multi-echo BOLD and VASO fMRI with visual stimulation to measure total and extravascular R₂* values in the visual cortex in human brain at 7T. By comparing total and extravascular ΔR₂* during visual stimulation, the intra- and extravascular parenchymal contributions to the BOLD signal at 7T could be assessed. Y_v and OEF changes during activation were estimated from the R₂* measurements.

Methods

The protocol was approved by the Internal Review Board of the Johns Hopkins University. Seven healthy subjects gave informed written consent before participating this study. From these, only six were included in the final report due to the fact that the relative extravascular ΔR₂* from subject 7 was more than two standard deviations larger than the averaged value from the other six subjects. Experiments were performed on a 7T human MRI scanner (Philips Healthcare, Best, The Netherlands), using a quadrature transmit head coil (10 inch or 25.4 cm in foot-head coverage) and a 32-channel phased array receive coil (Nova Medical, Wilmington, MA, USA). A single slice was carefully placed to cover the calcarine fissure. Three pseudo-randomized fMRI scans including two VASO (repetition time or TR = 4 s, inversion time or TI = 1293 ms) and one BOLD (TR = 2 s) scans were performed on each participant with visual stimulation (yellow/blue flashing checkerboard, 40s off 24s on, 4 blocks, 1 extra off period in the end). The blood nulling TI in VASO was calculated using blood T1 = 2212 ms, measured from bovine blood with 79% oxygenation and Hematocrit (Hct) = 0.43 at 7T [24]. The spatially nonselective inversion pulse in VASO was optimized for 7T in previous work (hyperbolic secant adiabatic pulse, duration = 20 ms, peak B1 = 15 μT, bandwidth = 1050Hz, >95% inversion at half of the maximum B1 in phantom and brain) [25, 26]. A magnetization reset module (90° RF pulse followed by spoiler gradients) [27] was applied immediately after the readout in both VASO and BOLD scans to suppress inflow effects from non-inverted spins. For both BOLD and VASO fMRI, single-shot gradient echo (GE) echo-planar-imaging (EPI) readout was used with four echoes acquired at TE = 9, 27, 45 and 63 ms. Common imaging parameters included: flip angle (FA) = 67° (Ernst angle for BOLD scan based on a grey matter T1 of 2132 ms [28]), field of view (FOV) = 192×192mm², spatial resolution = 2.5×2.5×2.5 mm³, parallel imaging (SENSE) acceleration factor = 4, partial Fourier fraction = 0.6. Second-order shimming was applied in both BOLD and VASO scans, for which a water line width of <60Hz was achieved in all scans.

All fMRI images were corrected for motion and baseline drift using Statistical Parametric Mapping (SPM8, University College London, UK) and Matlab R2009b (Mathworks, Natick, MA, USA). BOLD and VASO images at different TEs were co-registered. The two VASO scans were averaged to improve signal-to-noise ratio (SNR) [22]. VASO images from all four echoes were used to extrapolate to an effective TE of 0 ms to minimize BOLD contamination. A general linear model (GLM) was used to detect activated voxels in VASO (TE = 0 ms) and BOLD (TE = 27 ms, second echo) scans. Temporal SNR (tSNR) was calculated as the voxel-wised average signal divided by standard deviation along the time course during the rest periods (excluding the data acquired during 20 s at the start of each rest period). The criteria for activation were t-score > 1.5 (BOLD), t-score > 1 (VASO), adjusted p-value < 0.05, cluster size > 4 and tSNR > 20 [29]. Only voxels activated in both modalities were used for R₂* calculation so that voxels containing large vessels are excluded and signals are predominantly localized in the parenchyma [23, 29, 30]. Signal intensities (S) at the four echo times were numerically fitted as a function of TE (S =

S₀·exp(-TE·R₂*)) to obtain S₀ and R₂* values on a voxel-wise basis.

Using the calculated R₂* values, Y_v and parenchymal OEF can be quantified using the equations for the static dephasing regime described in [22, 31, 32]:

$$\Delta R_{2t}^{‡}\approx x_v\cdot {\omega }_0\cdot \frac{4\pi }{3}\Delta {\chi }_{deoxy}\cdot Hct\cdot \left\{CB{V}^{act}\left(1-Y_v^{act}\right)-CB{V}^{rest}\left(1-Y_v^{rest}\right)\right\}$$ [1]

where ΔR_2t* = R_{2t, act}* - R_{2t, rest}* (t denotes extravascular tissue), x_v = 0.7 is the venous fraction of cerebral blood volume (CBV) including capillaries [22], “act” and “rest” denote values during activation and rest, respectively. The reported susceptibility difference between fully oxygenated and deoxygenated blood (Δχ_deoxy) varies from 0.18 ppm to more than 0.3ppm in the literature [32-36]. Here, we adopted a value of 0.27 ppm from most recent studies [34, 35]. Other parameters assumed were: microvascular hematocrit (Hct) = 0.356 [37] and CBV^rest = 0.052ml blood/ml tissue [29]. CBV^act can be calculated from the VASO signal change [23]. By assuming Y_v^rest = 0.61, Y_v^act can be calculated from Eq. [1], and OEF can be quantified using [38]:

$$1-Y_v=1-Y_a+OEF\cdot Y_a$$ [2]

where the arterial oxygenation (Y_a) was taken to be 0.98, corresponding to a resting state OEF of 0.38.

Results

Figs. 1a-b show representative activation maps (thresholded) superimposed on BOLD and VASO images from one subject. The VASO signal changes displayed are those found when extrapolating to TE = 0 ms. Figs. 1c-d show the corresponding t-score maps. Note that the t-scores for activated voxels in VASO (negative signal change upon activation) were positive as well, because the contrast itself was reversed when performing GLM analysis. The peak activated voxels were well localized in the visual cortex. For the purpose of this study, activations that were clearly outside the visual regions were excluded in the analysis. Figs. 1e-f show the time courses of fractional signal changes averaged over activated voxels in the BOLD and VASO scans, respectively. As expected, negative signal changes in VASO were observed upon neuronal activation because the concomitant vasodilatation results in tissue signal reduction at TE = 0ms [23].

Figure 1.

Representative BOLD (TE = 27 ms) and VASO (extrapolated TE = 0 ms) fMRI results from one subject. a-b: Activation maps (thresholded) superimposed on BOLD and VASO images. c-d: Corresponding t-score maps. As the contrast in VASO and BOLD scans were reversed when performing GLM analysis, the t-scores for activated voxels in both scans were positive. e-f: Corresponding time courses of fractional signal changes from BOLD and VASO activation maps. The horizontal bars indicate the periods of visual stimulation. The blue and red points were used to calculate signals during rest and activation, respectively. Error bars show standard error over the included voxels within this subject.

Typical BOLD and VASO images at all four echo times are shown in Figs. 2a-b. As second order shimming was applied, and a single slice was acquired in the occipital lobe (which has a relatively homogeneous B0 field), the distortion found in these images was small. In addition, images at different TEs were co-registered before further analysis. The average tSNR (n = 6) of images at the longest echo time (TE = 63ms) were 16.0±1.9 and 12.8±1.7 for BOLD and VASO, respectively, which is considered sufficient for robust R₂* fitting. (Please note that the tSNR threshold of 20 mentioned in Methods was only applied on the extrapolated (TE = 0 ms) VASO images and BOLD images at the second echo time during functional analysis. Here, the tSNR of images at the longest TE shows that it is sufficient for R2* fitting.) Figs. 2c-d show the averaged result (n = 6) of TE-dependence curves of relative and absolute signal changes for both BOLD and VASO fMRI. Only the commonly activated voxels in both modalities were selected. The absolute BOLD signal change (Fig. 2d) was the largest at the second echo time (TE = 27 ms), consistent with the notion that the optimal TE for BOLD contrast should be around tissue T₂*[39]. The well-known linear relationship between ΔS/S and TE in BOLD and VASO was fitted (Fig. 2c). The VASO signal change is negative for very short TE, but reverses sign at longer TEs. This is expected as the extravascular BOLD effects (positive) become quite large at longer TE, which counteract the negative VASO signals. This also stresses the importance of extrapolating to TE = 0 or using a readout with very short TE [25] for VASO fMRI at 7T.

Figure 2.

Typical BOLD (a) and VASO (b) images at all four echoes (same scale). Group aver aged (n = 6) relative (c, ΔS/S) and absolute (d, arbitrary unit) fMRI signal changes versus TE in voxels that are both activated in BOLD (TE=27ms) and VASO (TE=0) methods. Error bars represent inter-subject variation. Lines in (c): results from linear fitting.

The fitted total and extravascular R₂* values during rest and activation in six subjects are summarized in Table 2. Only the voxels that were activated in both BOLD and VASO scans were included. The ratio of extravascular ΔR₂* to total ΔR₂* was 91 ± 3% (n = 6, mean ± SEM). Figs. 3a-b show the R₂* time courses from one subject.

Table 2.

Extravascular and total BOLD effects measured in gray matter parenchyma at 7T.

	Extravas. R2,rest* (s−1)	Extravas. R2,act* (s−1)	Extravas. Δ2* (s−1)a	Total R2,rest* (s−1)	Total R2,act* (s−1)	Total ΔR2* (s−1)	Extravas. ΔR2* fraction (%)b
Subject 1	40.38	39.08	−1.30	42.68	41.10	−1.57	83
Subject 2	46.61	45.16	−1.37	47.16	45.78	−1.45	95
Subject 3	40.46	39.73	−0.74	40.56	39.75	−0.81	91
Subject 4	45.18	43.89	−1.29	44.72	43.21	−1.51	86
Subject 5	45.02	43.26	−1.77	45.22	43.26	−1.96	90
Subject 6	50.34	49.17	−1.17	49.94	48.81	−1.13	103
Mean	44.66	43.38	−1.27	45.05	43.65	−1.40	91
SEM	1.55	1.51	0.14	1.34	1.33	0.16	3

SEM: inter-subject standard error.

^aΔR₂* = R_2,act* − R_2,rest*.

^bExtravas. ΔR₂* fraction = 100 × (Extravas. ΔR₂* / Total ΔR₂*) %.

Figure 3.

Representative time courses of total (a) and extravascular (b) parenchymal R₂* averaged over voxels activated in both BOLD (TE=27ms) and VASO (TE=0) scans from one subject. R₂* values were fitted from data acquired at all four TEs. The horizontal bars indicate the periods of visual stimulation. Error bars show inter-voxel standard errors within the subject. The dash-dot lines depict average R₂* values at baseline.

CBV increased by 35.8 ± 3.2% (n = 6, mean ± SEM) during visual stimulation, calculated from a -1.94 ± 0.17% (n = 6, mean ± SEM) VASO signal change using the extrapolated VASO images at TE = 0 ms and assuming a baseline CBV value of 0.052ml blood/ml tissue[29]. Using Eqs. [1,2], Y_v^act and OEF during activation were quantified to be 0.75 ± 0.01 and 0.24 ± 0.01 (n = 6, mean ± SEM), respectively, indicating an approximately 37% OEF decrease during visual stimulation.

Discussion

In this study, we applied multi-echo VASO fMRI to remove the intravascular signal and measured the extravascular (tissue) R₂* values in human visual cortex at 7T. Table 1 compares these data with the total and extravascular parenchymal R₂* values reported at various field strengths (1.5T, 3T and 7T) in the literature. The R₂* values measured here are in good agreement with the Donahue study [13], but differ considerably from the ones reported in the van der Zwaag study [8]. One plausible explanation may be that both the Donahue and current studies were conducted in the visual cortex with comparable spatial resolution, whereas the van der Zwaag study measured R₂* values in the motor cortex with a much finer spatial resolution (1×1×3 mm³). As expected, both total and extravascular R₂* increase with field strength. The absolute and relative R₂* changes (ΔR₂* and ΔR₂*/R₂*) during activation also increase with the field, indicating better sensitivity for BOLD fMRI as predicted. Extravascular ΔR₂* shows a linear trend with field strength (ΔR₂* = -0.196·B₀+0.114, correlation coefficient = 0.99, p-value = 0.09), consistent with the theoretical calculations for extravascular BOLD effects in the static dephasing regime model proposed by Yablonskiy and Haacke [32]. The relative contribution from the extravascular component in the total BOLD contrast becomes larger at higher field, as indicated by the larger ratio of extravascular ΔR₂* to total ΔR₂*. Our results showed that the BOLD effect is dominated by the extravascular component (91%) at 7T, in line with the results from Duong et al.[1] and Donahue et al.[13] that used crushing gradients to suppress the intravascular signal. These experimental results are consistent with the theoretical calculations that the intravascular BOLD signal will be significantly reduced relative to the extravascular signal at higher field due to the faster R₂* decay of venous blood [1-9]. It is important to mention that the intravascular effects are actually larger than extravascular at short TE [7, 17, 21]. However, when using typical TEs for BOLD fMRI (which are usually comparable to GM T2*), the relative contribution from the intravascular venous compartment to the overall parenchymal effect is small.

When assuming resting OEF and CBV values from the literature, multi-echo VASO fMRI also allows the measurement of changes in physiological parameters such as Y_v and OEF in the parenchyma during neuronal activation. OEF reduction upon visual stimulation measured here at 7T agrees reasonably well with previously reported values: 33% [40, 41], 39% [22], 53% [33] at 1.5T and 45% [22] at 3T. A number of neurophysiology studies suggest that the vasodilation during functional stimulation occurs predominantly in arterioles and capillaries that are very close to the active neural tissue [42], indicating an improved spatial specificity for CBV weighted VASO fMRI, as demonstrated by several animal and human fMRI studies [23, 29, 30]. Therefore the OEF changes measured here were expected to be predominantly localized in the parenchyma by taking overlapping voxels activated in both BOLD and VASO scans, while some previous measurements were made mainly in the draining veins [33, 40, 41, 43-46].

In this 7T study, a blood T1 of 2212 ms was used to calculate the blood nulling time (TI) in VASO fMRI, which was measured from 79% oxygenation bovine blood (Hct = 0.43) at 7T[24]. Although this is within typical physiological range in normal human brains, it is known that blood T1 is sensitive to hematocrit and, to a lesser extent, to oxygenation [47-49]. This may affect the quantification of extravascular R₂* values and relative CBV changes during functional activation. The oxygenation level of 79% corresponds mostly to an estimated average of the blood in capillaries and venules [38, 50]. As only voxels activated from both the BOLD and VASO scans were used in the calculations, blood signals in veins should be largely excluded (if blood in veins is not completely nulled in VASO scans, it would reduce or cancel out the negative VASO changes during activation, thus be excluded in the voxel selection). Therefore, only incompletely nulled blood in arteries and arterioles is most likely to affect the extravascular R₂* quantification. However, since the R₂* values are comparable in arterial blood (about 40 s⁻¹, unpublished data, experimental setup same as in [49]) and in extravascular tissue (Table 2) at 7T, this should not result in major biases in the extravascular R₂* estimation. Gu et al.[51] showed that the VASO signal change (ΔS/S, thus ΔCBV) during visual stimulation plateaus over a range of about 150 ms around the theoretical blood nulling TI. Donahue et al.[52] also demonstrated that at typical spatial resolution for fMRI (around 3mm) and at relatively long TRs (>4s), the variation of blood T1 over a range of 100 ms has very limited influence on VASO signal change and ΔCBV estimation. It is therefore reasonable to expect that the blood T1 variation over typical physiological ranges in normal human brains should not have substantial effects on the ΔCBV estimation here. Moreover, the relative difference of T1 values decreases with field strength[28, 30], which should further reduce the potential biases in R₂* and ΔCBV estimation resulted from blood T1 variations.

The contribution from physiological noises in fMRI signals has been shown to increase with field strength and voxel size [53]. In order to check this for our study, we measured the coefficients of variation for theR2* values (Total R2*: 0.073; Extravascular R2*: 0.085), which were found to be comparable to those reported at 3T (Total R2*: 0.076; Extravascular R2*: 0.063) and 1.5T (Total R2*: 0.077; Extravascular R2*: 0.079) in a previous study adopting similar methodology [22]. However, a larger voxel size was used in that study (2×2×5=20mm³ versus 2.5×2.5×2.5=15.625mm³ here). From this we conclude that the noise contribution in the R2* measurements at 7T would be larger than 3T and 1.5T if the same voxel size were used. We also noticed that the total R2* values measured in subjects 4 and 6 were slightly lower than the corresponding extravascular R2* values (total R2* is expected to be higher than extravascular R2* due to additional blood contributions), although the difference was within noise range. This also occurred for one subject reported in previous 3T data but not 1.5T [22]. We attribute this to the many possible contributions to R2*, making it difficult to measure this parameter with great accuracy. More repeats and averaging may be necessary for robust R2* measurement on a single-subject level.

A potential source of errors when estimating Y_v^act and OEF using Eqns. 1 and 2 comes from the literature values assumed for the model parameters. While this has been investigated in previous works [54-56], here again we performed an error analysis by estimating OEF using parameter values (assumed in Methods) over the normal physiological range: baseline CBV from 0.045 to 0.055 ml/ml, microvascular Hct from 0.38 to 0.46 and Δχ_deoxy from 0.20 to 0.27 ppm. Less than 8% estimated OEF differences were found between the two ends of both Hct and Δχ_deoxy ranges, whereas about 14% difference were observed when varying baseline CBV values. The assumed CBV^rest value of 0.052ml/ml in this study is approximated [29] based on the reported 0.048 - 0.055 ml/ml range in the literature [57-61]. Although this error analysis shows that the estimated OEF values in this study are only moderately affected by these assumptions, the measurement of these physiological parameters in each participant would certainly improve the accuracy for OEF quantification.

Conclusions

Total and extravascular R₂* values in the parenchyma in human visual cortex were measured using multi-echo VASO and BOLD fMRI with visual stimulation at 7T. The parenchymal extra-vascular R₂* value was 44.66 ± 1.55 s⁻¹ at rest, and the ratio of extravascular ΔR₂* to total ΔR₂* was 91 ± 3% at 7T, confirming a predominant contribution from the extravascular component of the BOLD effect. A 37% decrease in parenchymal OEF during stimulation was estimated based on these measurements, consistent with values reported at lower field strengths.

Abbreviations

BOLD

blood oxygenation level dependent

VASO

vascular space occupancy

CBV

cerebral blood volume

OEF

oxgyen extraction fraction

fMRI

functional MRI

GE

gradient echo

EPI

echo planar imaging

SENSE

sensitivity encoding

SNR

signal-to-noise ratio

GLM

general linear model