Implementation of Vascular-space-occupancy (VASO) MRI at 7 Tesla

Abstract

VASO-MRI exploits the difference between blood and tissue T₁ to null blood signal and measure cerebral blood volume (CBV) changes using the residual tissue signal. VASO imaging is more difficult at higher field because of sensitivity loss due to the convergence of tissue and blood T₁ values and increased contamination from BOLD effects. In addition, compared to 3T, 7T MRI suffers from increased geometrical distortions, e.g. when using echo-planar-imaging (EPI), and from increased power deposition, the latter especially problematic for the spin-echo-train sequences commonly used for VASO-MRI. Third, non-steady-state blood spin effects become substantial at 7T when only a head coil is available for radiofrequency transmit. In this study, the magnetization-transfer-enhanced-VASO (MT-VASO) approach was applied to maximize tissue-blood signal difference, which boosted SNR by 149 ± 13% (n=7) compared to VASO. Second, a 3D fast gradient-echo sequence with low flip-angle (7°) and short echo-time (1.8ms) was employed to minimize the BOLD effect and to reduce image distortion and power deposition. Finally, a magnetization-reset technique was combined with a motion-sensitized-driven-equilibrium (MSDE) approach to suppress three types of non-steady-state spins. Our initial fMRI results in normal human brains at 7T with this optimized VASO sequence showed better SNR than at 3T.

Introduction

Vascular-space-occupancy (VASO) MRI (1,2) measures cerebral blood volume (CBV) changes through extravascular tissue signal changes. Based on the difference between blood and tissue T₁ relaxation times, inversion recovery is employed to minimize blood signal while keeping substantial tissue signal for detection. During neuronal activation, a negative signal change is found, which is attributed to vasodilation causing a reduction of tissue signal in the voxel. Several variations of VASO sequences have been developed in recent years to improve its MRI sensitivity and contrast specificity (3–21). While originally proposed as a functional MRI (fMRI) methodology in normal brain (22–27), VASO MRI has started to make inroads into clinical applications, such as aging (28), stroke (29) and brain cancer (30,31).

Recently, ultra-high magnetic field (7 Tesla (7T)) human scanners have become available. In view of the approximate linearity of signal-to-noise ratio (SNR) in physiological samples with field strength, 7T is expected to have an increase in SNR by a little over a factor of two compared to the commonly used 3 Tesla (3T). However, there are several technical challenges for proper implementation of VASO MRI at 7T. First, the difference between the longitudinal relaxation times (T₁) of grey matter (GM) and blood reduces with field strength (about 21% at 7T (32), 45% at 3T (33,34)), so that residual GM signal at the blood nulling point is reduced, which significantly negates the sensitivity gain at higher field. Second, the imaging sequences commonly used for VASO MRI at 3T, such as echo-planar-imaging (EPI), turbo spin echo (TSE) imaging (6) and gradient spin echo (GRASE) imaging (13,15), suffer from geometrical distortion (EPI and GRASE) and high power deposition (TSE and GRASE) at 7T. In addition, the blood-oxygenation-level-dependent (BOLD) effect is much greater at higher field. A gradient echo (GE) EPI sequence with the shortest possible echo time (TE) may still produce a substantial positive signal change during neural activity (due to the BOLD effect) at 7T, which counteracts the small (1–3%) negative VASO signal change and results in an underestimate of CBV changes in the brain. Thirdly, non-steady-state inflowing blood spin effects, which would result in an overestimate of CBV changes (4,11,35), become more substantial at 7T because (i) it takes longer to reach steady state for a longer blood T₁; (ii) the head coils currently used for radiofrequency (RF) transmit have limited coverage for inversion and a body coil is not yet available for 7T.

Here, we use the magnetization transfer (MT) enhanced VASO (MT-VASO) technique (12) to magnify the detectable tissue signal (thus SNR) at the time of blood nulling. This can be done with the specific absorption rate (SAR) well below the Food and Drug Administration (FDA) limit. The relative enhancement effect between MT-VASO and VASO is much greater at 7T than 3T. Second, we show that a three dimensional (3D) fast gradient echo (fast GRE, also known as turbo field echo, TFE, or TurboFLASH) imaging sequence can be employed to reduce geometrical distortion in images (vs. EPI), lower SAR (vs. TSE), and decrease TE markedly, thus minimizing the BOLD contamination for VASO MRI at 7T. Thirdly, we demonstrate that the effect from non-steady-state inflowing blood spins can be suppressed by combining the previously proposed “magnetization reset” technique (7,35) and a motion-sensitized driven equilibrium (MSDE) preparation (36–38). We show that these improvements allow 3D single shot MT-VASO fMRI (2mm isotropic voxel, 21 slices, temporal resolution 4s) at 7T during a visual task in normal subjects. The SNR of this optimized VASO sequence at 7T and 3T is compared.

Materials and Methods

VASO pulse sequence optimization

Fig. 1 illustrates the final VASO sequence implemented at 7T. In addition to the standard VASO spatially nonselective inversion and an image readout module at the blood nulling inversion time (TI), we added and optimized several parts of the sequence to improve its performance:

Figure 1.

MT-VASO pulse sequence implemented at 7T (time scale not in proportion). Shaded and open rectangles with thick dark borders represent radiofrequency (RF) and gradient (GR) pulses, respectively. Two consecutive scans (TR) are shown. Three time periods (I, II, III) are marked, corresponding to the time when the three types of non-steady-state blood spins flow into the transmit coil, respectively (see Methods). Note that the three periods are defined with respect to the second TR shown in the figure, which is considered the current TR.

MT-VASO

Application of an MT pulse before the VASO inversion pulse can be used to prepare a smaller tissue magnetization before inversion and accelerate the recovery process after inversion, thereby obtaining a higher tissue signal at the same blood nulling time, and consequently boost the SNR and contrast-to-noise ratio (CNR) for VASO MRI (12). This is based on the fact that tissue signal is strongly affected by the MT pulse while blood is not when using medium irradiation power (≤3µT) and duration (≤500ms), and/or a frequency offset far away from water resonance (≥40ppm) (12,39,40). Here, a 400ms 2.5µT block-shaped MT saturation pulse at a frequency offset of −40ppm was applied immediately before the VASO inversion pulse. The choice of the parameters was made based on the criteria of minimal MT effect in blood, substantial MT effect in tissue, and SAR well below FDA limit. Note that the MT effect is asymmetric around the water resonance in tissue (41) but not in blood (12,39,40). Therefore, the MT effect in tissue is slightly larger at a negative frequency offset (−40ppm) than the corresponding positive offset (40ppm).

Inversion pulse

A hyperbolic secant adiabatic pulse was employed for spatially nonselective inversion in VASO MRI. Based on the results reported in a previous study (42) in which maximum B₁ of 20µT was used, the pulse was empirically optimized on a head-shaped phantom (H₂O/NaCl/CuSO₄) and a human subject to achieve sufficient (>95%) inversion efficiency on our 7T scanner with a maximum B₁ of 15µT. The resulting parameters for the adiabatic pulse (as defined in (43)) are: duration=20ms, β=525rad/sec, µ=8.4, peak B₁=15µT, bandwidth=1050Hz. Simulations show that the pulse can maintain >95% inversion at half of the maximum B₁ and about 80% inversion at a third of maximum B₁. This tolerance is important at 7T where a substantial spatial variation in B₁ is commonly seen.

Imaging sequence

A 3D spoiled fast GRE (also known as T₁-enhanced TFE or TurboFLASH) readout, which comprises of a train of spoiled gradient echoes with low flip angle (FA) and short repetition times (TR) between each echo, was employed. This sequence is expected to have less geometrical distortion than EPI, lower SAR than TSE due to lower FAs, and permits very short TE to minimize the BOLD contamination in VASO MRI. Imaging parameters: voxel size 2mm isotropic, 21 slices, TR_GRE (this is the TR between two echoes during the fast GRE readout)/TE=3.9/1.8ms, FA=7°, TR (TR between two consecutive scans)=4s, turbo direction=radial, parallel imaging acceleration (SENSE factor)=3×2(AP×FH). A low-high (also known as “centric”) phase encoding was used so that the center of k-space (kz=ky=0 for 3D), which determines the gross signal intensity in the image, was acquired at the first echo. This is important for minimizing the time delay, thus the T₁ relaxation during the delay, between the T₂ preparation module (discussed next) and the acquisition of center of k-space. One confounding factor when using fast GRE sequences with short TR values is that image acquisition takes place when the longitudinal magnetization approaches a steady state (44,45) (this is the steady state during the readout pulse train, not to be confused with the one in the inflow effect, which is discussed next). While the signal intensity in MRI images is dictated by the center of k-space, the evolution of magnetizations during the acquisition of outer k-space lines would lead to a distorted point spread function (PSF), which causes blurring artifacts in the images (44). Epstein et al have demonstrated that when a low FA (<20°) is used, such artifacts are minimal even without a preparatory phase during which dummy echoes are acquired and/or FA sweep is implemented (44). To confirm this for our sequence parameters, we performed numerical simulations (parameters described later in this section) to calculate the magnetization for each echo in the fast GRE readout, and discrete Fourier transform to obtain the PSF. The resulting PSF was only slightly distorted with maximum amplitude of the side lobes less than 5% of the main peak, which indicates that such blurring artifacts are minimal in our sequence with a FA of 7°.

Suppression of non-steady-state blood spin effects

The choice of TI for blood nulling in VASO MRI is based on the assumption that all blood spins within the imaging volume have reached the inversion steady state of approximately nulled signal before image acquisition. However, this assumption may not hold if TR is short relative to blood T₁ and/or the inversion volume (determined by transmit coil coverage) is small so that some fast flowing blood spins are not inverted sufficiently often before image acquisition to reach the steady state. It has been shown that such non-steady-state blood spin effects are negligible at 3T when a body coil is used for RF transmit and TR is sufficiently long (>2s) (11). However, this effect may become substantial at 7T due to the longer blood T₁ and smaller transmit coil (only a head coil is available currently). The non-steady-state inflowing blood spins can be categorized into three types based on the time when they flow into the transmit coil (periods I, II and III, respectively, in Fig. 1; note that the three periods are defined with respect to the second TR shown in Fig. 1): I) spins flowing in before the end of readout of the previous TR; II) spins flowing in between the readout of previous TR and the inversion pulse of current TR; III) spins flowing in after the inversion pulse of current TR. Type I non-steady-state blood spins can be eliminated by a spatially nonselective saturation module (90° RF pulse followed by spoiler gradients) applied immediately after each readout (7,35). This “magnetization reset” technique (35) establishes the steady state for all spins within the transmit coil for the next scan. However, it is not effective for spins flowing into the coil after the saturation module (type II and III spins). The difference between type II and III spins is that they have experienced only one and zero inversion pulse before image acquisition, respectively. Those spins, provided they flow fast enough to reach the imaging volume before next acquisition, will still have non-steady-state effects on the overall VASO signal. We therefore incorporated a motion-sensitized driven equilibrium (MSDE) technique (36–38) into the VASO sequence in order to eliminate contaminations from these type II and III non-steady-state spins. The MSDE technique has been used previously in angiography (MRA) (36–38) and arterial spin labeling (ASL) (46). A spatially nonselective Carr–Purcell–Meiboom–Gill (CPMG) based T₂ preparation module with inserted motion-sensitized crushing gradients (36–38) in the z-direction (Fig. 1, two refocusing pulses, effective TE=15ms, inter-echo spacing (τ_CPMG)=7.5ms, velocity encoding (V_enc)=3cm/s) was applied immediately before the readout. The signals from flowing blood spins, including type II and III non-steady-state spins, will be crushed by the motion-sensitized gradients. Two, instead of one, refocusing pulses were applied to better compensate for the transmit (B₁) field inhomogeneity and an adjusted gradient scheme was used to reduce eddy currents (37). The effective TE is minimized to reduce signal loss from T₂ decay and to suppress the T₂-weighted BOLD effect during functional activation. The potential residual BOLD contamination from the T₂ preparation module was investigated in subsequent experiments. The MSDE module is followed by the 3D fast GRE readout (separated by a spoiler gradient of 6ms, during which negligible T₁ relaxation can occur), so that the vascular crushing effects are well preserved when the first echo (center of k-space, determines the gross signal intensity in the image, see Imaging sequence) is acquired. The RF pulses used in magnetization reset and MSDE were adopted from a previous work by Visser et al (42), in which they were optimized specifically for 7T.

Determination of steady state blood nulling time (TI)

The inversion time (TI) is defined as the interval between the inversion pulse and the first echo in 3D fast GRE readout when the center of k-space is acquired (Fig. 1). Numerical simulations (described next) were used to determine the optimal TI for nulling steady state blood signal at TI while accounting for the signal evolution during the T₂ preparation module and 3D fast GRE readout. The 3D fast GRE sequence has a relatively substantial effect on the effective TI as it has multiple excitation pulses with short TR. Note that the number of excitation pulses (echoes) can change with imaging parameters such as spatial resolution, number of slices and parallel imaging acceleration. On the other hand, the T₂ preparation module has a minor effect because of its short duration (15ms) and nearly nulled steady state blood signal at the beginning of the module (so little T₂ decay can occur during MSDE). For this reason, the optimal TI is not sensitive to the blood T₂ value used in simulation. To confirm this, we used blood T₂ of 67ms (100% O₂) and 19ms (60% O₂) (unpublished data at 7T) in the simulation, and the difference between the resulting optimal TIs is less than 1ms (TR=4s, TI=903ms for the imaging parameters described in Imaging sequence).

Simulations

Bloch simulations were performed to determine optimal TI and estimate the residual tissue (GM, white matter, WM) and cerebrospinal fluid (CSF) signal in the 7T VASO sequence described above with and without MT enhancement. All residual transverse magnetizations were spoiled and therefore not included in the simulations. The MT saturation in MT-VASO was experimentally measured in each subject (see Experiments), and the average values were used in the simulations. Typical T₁ and T₂ values in healthy human brains were used: 7T: T_1,blood=2587ms, T_1,GM=2132ms, T_1,WM=1220ms, T_1,CSF=4425ms (32), T_2,blood (80% O₂)=43ms (unpublished data), T_2,GM=50ms (47); 3T: T_1,blood=1624ms (34), T_1,GM=1122ms (33), T_2,blood (80% O₂)=55ms (48), T_2,GM=80ms (33). The simulation was repeated for a few TRs until steady state is reached. Only the steady state signal (proportional to the magnitude of longitudinal magnetization) was reported. All computing programs were coded in Matlab R2009b (Mathworks, Natick, MA, USA).

Experiments

Seven healthy human subjects (4 female, 3 male), who gave written informed consent before participating in this Johns Hopkins Institutional Review Board (IRB) approved, Health Insurance Portability and Accountability Act (HIPAA)-compliant study, were scanned on a 7T Philips MRI scanner (Philips Medical Systems, Best, The Netherlands). A 32-channel phased-array head coil (Nova Medical, Wilmington, MA) was used for RF reception and a head-only quadrature coil for transmit. Two rectangular pads (23×10×2mm) filled with high dielectric constant materials (49) were placed between the lateral sides of the subjects’ head and the coil, in order to improve field homogeneity (50). fMRI sessions were performed using visual stimulation with blue/yellow flashing checkerboard (40s off/24s on, 4 repetitions, 1 extra off period in the end) delivered using a projector from the back of magnet. Each fMRI run took 4 minutes and 56 seconds during which 74 image volumes (TR=4s) were acquired. Two sets of experiments were carried out on each subject on separate days.

Experiment set I is intended to test the MT enhancement in MT-VASO and compare GE EPI and 3D fast GRE sequences for MT-VASO MRI at 7T. Four pseudo-randomized fMRI experiments were performed on each subject: (a) single slice MT-VASO with GE EPI readout (single shot, TR/TI=4s/1294ms, 3 echoes, TE/echo spacing (ES)=15/20ms, FA=70°, SENSE factor=4, partial Fourier fraction=0.7, fat suppression). The same inversion and MT pulses, and magnetization reset module as described earlier were used. Bipolar crushing gradients (51,52) (V_enc=3cm/s), which had the same vascular crushing effect as the MSDE module used with 3D fast GRE readout, were applied in the GE EPI sequence. (b) VASO (no MT pulse) with 3D fast GRE readout, TR/TI=4s/903ms, magnetization reset and MSDE applied; (c) MT-VASO with 3D fast GRE readout, TR/TI=4s/903ms, magnetization reset and MSDE applied; (d) only the 3D fast GRE readout, TR=4s, no inversion and MT pulses, no magnetization reset and MSDE. In addition, three scans without functional stimulation were carried out on each subject: (e) to measure MT effects, the same 3D fast GRE readout with and without MT saturation, no delay between MT and readout, TR=4s, 15 averages; (f) to estimate the equilibrium MR signal (S₀), the same 3D fast GRE readout with a longer TR=20s, no inversion and MT pulses, magnetization reset and MSDE applied, 10 averages; (g) for anatomical reference, 3D Magnetization Prepared RApid Gradient Echo (MPRAGE) (voxel=1mm isotropic, TR/TE/TI=4.0/1.8/1291ms, SENSE=2×2). Common parameters in (a–f): field of view (FOV)=210×210mm², voxel=2mm isotropic. The single slice in (a) was aligned with the center of the 3D volume in (b×f). The highest specific absorption rate (SAR) shown on the scanner (scan (c)) was less than 2.1W/kg, which is approximately 66% of the maximum SAR approved by FDA.

Experiment set II is intended to test the magnetization reset and MSDE modules, and the potential T₂-weighted BOLD effects from the T₂ preparation pulses in MSDE. Four pseudo-randomized fMRI experiments were performed on each subject: (a) MT-VASO with 3D fast GRE readout, TR/TI=4s/892ms, no magnetization reset and MSDE; (b) MT-VASO with 3D fast GRE readout, TR/TI=4s/892ms, with magnetization reset but no MSDE; (c) MT-VASO with 3D fast GRE readout, TR/TI=4s/903ms, with magnetization reset and the T₂ preparation pulses in MSDE, but no crushing gradients; (d) MT-VASO with 3D fast GRE readout, TR/TI=4s/903ms, with magnetization reset and MSDE as described in Methods. Common parameters: FOV=210×210mm², voxel=2mm isotropic. Note that the TIs in (a,b) are slightly shorter due to the absence of T₂ preparation pulses.

3T experiments: To compare SNR between 3T and 7T, the same MT-VASO sequence (with magnetization reset and MSDE) was implemented on a 3T human MRI scanner (Philips Medical Systems, Best, The Netherlands). Two of the seven subjects (1 female, 1 male) were scanned with a body coil (approximately 650 mm in axial length) used for RF transmit and a 32-channel phased-array head coil (In Vivo, Corporation, Florida) for reception. All parameters were identical to Experiment set I scan (c) on 7T (same readout bandwidth), except for TI = 764ms. The imaging volume was carefully placed to match the location in the 7T scans for each subject. No fMRI paradigm was played out in these 3T scans.

Data analysis

The Statistical Parametric Mapping (SPM8, University College London, UK) software package and several in-house Matlab R2009b (Mathworks, Natick, MA, USA) routines were used for data analysis. For each participant, all 3D fast GRE images were re-aligned and co-registered with the anatomical reference (MPRAGE images), from which masks of GM, WM and CSF were generated (all done with SPM8 routines). A general linear model is employed to detect functional activation in VASO fMRI scans (P=0.05, t-score<−2, cluster size≥4, and SNR>5). The fractional signal in each voxel was computed by normalizing to the average baseline signal. The relative signal change (ΔS/S) in each voxel was calculated by subtracting the average fractional signal in that voxel during the baseline period from that during activation. The average values over all activated voxels and subjects were reported in Results. The data acquired during the first 20s (5 time points) and 8s (2 time points) in the resting and activation periods, respectively, were excluded in the calculation so that the underlying hemodynamic response is allowed to reach its steady state. SNR of the MR images was calculated with two baseline images in the middle of the paradigm, using a method published by Kruger et al (53). For each ROI (GM, WM, CSF), the signal was defined as the mean intensity of VASO images (S), while the noise level (N) was estimated with the standard deviation of a difference map from the two baseline images within the ROI. The CNR for each voxel was defined as $\mathit{\text{SNR}}\times |\mathrm{\Delta }S/S|\times \sqrt{N_a}$ (N_a = number of images).

Results

Table 1 summarizes the measured MT effects, simulated and measured residual signal (S/S₀) at blood nulling TI, measured SNR and CNR in GM, WM and CSF for VASO and MT-VASO at 7T (Experiment set I). Masks generated from the MPRAGE scans were used. The measured residual signals (S/S₀) were all consistent (P<0.05) with the simulated values. The mean S/S₀ of GM in VASO was slightly larger than the simulated value. However, due to the large standard deviation, which may result from low signal and heterogeneous T₁ values in GM, they were still statistically comparable (P<0.05). For the same reason, the measured signal enhancement in MT-VASO compared to conventional VASO was lower than simulated. The SNR enhancement in GM was approximately 150% for MT-VASO, in proportion to the measured residual signal, as expected. No significant difference (P>0.1) was found in the functional signal changes (ΔS/S) for VASO (−2.02±0.36%) and MT-VASO (−1.98±0.32%) (n=7), in line with previous results at 3T (12). Therefore, the CNR enhancement in GM during functional activation was also proportional to that in SNR and residual signal. Although the MT effects in WM were much greater than GM, the relative enhancement (40–45%) of residual signal and SNR in MT-VASO was much smaller for WM, mainly because of the already large WM signal in conventional VASO. No significant MT effect was detected in CSF. As a result, the CSF signals and SNR in VASO and MT-VASO were comparable (P>0.1), in line with the simulations.

Table 1.

Comparison of MT effects, simulated and measured residual signal at blood nulling TI, SNR and CNR in GM, WM and CSF for VASO and MT-VASO at 7T. ^*

		GM	WM	CSF
Measured MT effect (MTR) 1 (%)		9.1 ± 5.7	17.5 ± 3.3	−0.1 ± 1.5
Simulated relative signal	VASO S/S0 (%) 2	2.3	15.0	3.83
	MT-VASO S/S0 (%)	8.3	21.9	3.83
	Enhancement4 (%)	261	46	0
Measured relative signal	VASO S/S0 (%)	3.2 ± 2.5	15.6 ± 1.9	5.1 ± 1.5
	MT-VASO S/S0 (%)	8.2 ± 4.4	21.6 ± 1.9	5.0 ± 1.7
	Enhancement (%)	152 ± 18	41 ± 6	1.± 12.3
Measured SNR	VASO	7.6 ± 3.4	36.1 ± 3.0	8.9 ± 2.6
	MT-VASO	18.9 ± 8.9	52.4 ± 3.1	8.7 ± 2.9
	Enhancement (%)	149 ± 13	45 ± 5	−1.9 ± 13.2
Measured CNR5	VASO	1.5 ± 0.4	N/A	N/A
	MT-VASO	3.8 ± 0.9	N/A	N/A
	Enhancement (%)	160 ± 16	N/A	N/A

^*Mean values ± standard deviation over all subjects (n=7).

¹MT effect is defined as the percentage of signal drop after an MT saturation pulse as compared to the signal without the saturation, which is commonly defined as the MT ratio (MTR). The MT effects here are from the off-resonance MT pulse in front of the VASO inversion pulse, not including the on-resonance MT effect from the inversion pulse itself, which should be comparable at 3T and 7T.

²S represents the residual signal at the blood nulling TI. S₀ is the equilibrium MR signal of the corresponding tissue (i.e. water density C is included).

³S/S₀ is negative for CSF. The absolute values are reported here to be compared with modulus signals acquired in MR images.

⁴Enhancement is defined as the percentage of signal increase in MT-VASO as compared to that in the conventional VASO.

⁵CNR is averaged over activated GM voxels only.

Fig. 2 illustrates representative 3D fast GRE MT-VASO images (a), activation map (b) and average time course (c) from voxels meeting activation criteria (Experiment set I, scan (c)). The activated voxels were well localized in the visual cortex, with an average negative signal change of −1.98±0.32% (n=7), the magnitude of which is in line with theoretical predictions at 7T (54).

Figure 2.

Representative results from one subject for VASO MRI at 7T. (a) Typical MT-VASO images (3D fast GRE readout, 21 slices acquired, 16 slices shown). (b) Activation maps for MT-VASO fMRI (glass brain overlay). A: anterior; P: posterior; L: left; R: right. (c) Average time course from voxels meeting activation criteria in (b). The dark horizontal bars under the time course indicate the visual stimulation periods. Blue and red dots were used for calculating signals during baseline and stimulation, respectively. The error bars represent inter-voxel standard deviations.

The results for comparing GE EPI and 3D fast GRE MT-VASO sequences (Experiment set I, scans (a) and (c), respectively) are shown in Fig. 3 and Table 2. For slices close to the nasal cavity, the geometrical distortion and signal void artifact in the frontal area in 3D fast GRE VASO (Fig. 3b) was much reduced compared to GE EPI VASO (Fig. 3a). To investigate the residual BOLD effects, three echoes in GE EPI VASO scans were used to extrapolate to effective TE=0. In the extrapolated images (Fig. 3d), slightly more activated voxels (P<0.1, Table 2, Figs. 3c,d) were detected compared to the original GE EPI VASO images with shortest TE (15ms). The average signal changes in voxels that were activated in both images (TE=0, 15ms) were also larger (P<0.01, Table 2, Fig. 3f) in the extrapolated data (TE=0). When comparing the TE=0 GE EPI VASO data to the matching slice in 3D fast GRE VASO, both gave comparable (P>0.1, Table 2, Figs. 3d,e) number and pattern of activated voxels, relative signal changes, and time courses (Fig. 3f). Fig. 3f also shows that when only the 3D fast GRE readout (without inversion, Experiment set I, scan (d)) was played out during the same visual stimulation paradigm, no significant positive or negative signal changes (−0.02±0.39%, n=7, P>0.5) were detected in the same activated voxels from scan (c, 3D fast GRE VASO).

Figure 3.

Comparison of GE EPI and 3D fast GRE MT-VASO. (a,b) Image from a single slice GE EPI MT-VASO scan and the matching slice in a 3D fast GRE MT-VASO scan, respectively. The location of the images was close to the nasal cavity. Geometrical distortion and signal void in the frontal area are obvious in GE EPI but barely noticeable in 3D fast GRE. (c–e) Images from one subject with activated voxels overlaid with their t-scores (color scale). (c) GE EPI MT-VASO, first echo, TE=15ms; (d) GE EPI MT-VASO extrapolated to TE=0, only brain region displayed; (e) 3D fast GRE MT-VASO with magnetization reset and MSDE. (f) Time courses of relative signal changes. Error bars represent inter-subject variations (n=7). Two vertical dash lines indicate the start and end of visual stimulation. Four blocks of baseline and stimulation periods are averaged to one block. Time is with respect to the start of visual stimulation.

Table 2.

Summary of results from Experiment set I (scans (a) and (c)).^*

	GE EPI MT-VASO 1st Echo (TE=15ms) scan (a)	GE EPI MT-VASO After extrapolation (TE=0) scan (a)	3D fast GRE MT-VASO1 scan (c)
Number of activated voxels	112±32	154±38	152±44
ΔS/S (%) 2	−1.13±0.31	−1.95±0.31	−1.98±0.32

^*Mean values ± standard deviation over all subjects (n=7).

¹Only the middle slice in scan (c, 3D fast GRE MT-VASO) was selected to match the single slice in scan (a, GE EPI MT-VASO).

²ΔS/S=(S_act-S_base)/S_base. S_act and S_base are signals during stimulation and baseline, respectively.

Fig. 4 and Table 3 summarize the results from Experiment set II. All activated voxels were well localized in the visual cortex and the activation pattern was similar among all four scans. The number of activated voxels (all slices included) averaged over all subjects (n=7, Table 3) was statistically comparable (P>0.1) for the four scans. Even though scan (d) (MT-VASO with magnetization reset and MSDE) showed slightly less activated voxels, possibly due to signals from fast flowing spins in large vessels crushed by the motion-sensitized gradients, the standard deviation was large and the difference was not statistically significant. To compare the time courses and signal changes across scans (a–d), only voxels that were activated in all four scans were selected. The general trend of these four time courses shown in Fig. 4e was similar, consistent with the VASO time courses in the literature. The relative signal changes (ΔS/S) during visual stimulation (Table 3) were the largest (P<0.01) in scan (a), comparable (P>0.1) between scans (b) and (c), and smallest (P<0.01) in scan (d).

Figure 4.

Results from Experiment set II. (a–d) Middle slices from one subject with activated voxels overlaid with their t-scores (color scale). (a) MT-VASO; (b) MT-VASO with magnetization reset (m.r.); (c) MT-VASO with magnetization reset and T₂ preparation pulses but no crushing gradients in between; (d) MT-VASO with magnetization reset and MSDE. (e) Time courses averaged over voxels (all slices) that were activated in all four scans (a–d) in each subject, and then averaged over subjects (n=7). Error bars represent inter-subject variations. Two vertical dash lines indicate the start and end of visual stimulation. Four blocks of baseline and stimulation periods are averaged to one block. Time is with respect to the start of visual stimulation.

Table 3.

Summary of results from Experiment set II (scans (a–d)).^*

	MT-VASO scan (a)	MT-VASO +m.r.1 scan (b)	MT-VASO +m.r.+T2prep2 scan (c)	MT-VASO +m.r.+MSDE scan (d)
Number of activated voxels	942 ±382	1129 ± 518	1147 ± 544	819 ± 359
ΔS/S (%) 3	−3.98 ± 0.75	−2.94 ± 0.61	−3.16 ± 0.45	−2.04 ± 0.19

^*Mean values ± standard deviation over all subjects (n=7). All slices were included.

¹Magnetization reset.

²T₂ preparation pulses without motion-sensitized crushing gradients in between.

³ΔS/S=(S_act-S_base)/S_base. S_act and S_base are signals during stimulation and baseline, respectively. This is averaged over voxels that were activated in all four scans. All slices are included.

To compare the SNR of the optimized MT-VASO sequence at 3T and 7T, we first (Table 4) simulated the GM signals at blood nulling TI and the SNR ratios (7T/3T) assuming that SNR is proportional to MR signal and field strength. Note that the effects are TR dependent. For conventional VASO, the residual GM signals increase with TR at both fields. For MT-VASO, the residual GM signals are less TR dependent, with higher signals at short (2.5s) and long (10s) TRs. The MT enhancement effects, which decrease with TR at both fields, are much greater at 7T within each TR. The residual GM signals are lower at 7T for all TRs with both VASO and MT-VASO. The SNR for conventional VASO at 7T is at best comparable to 3T at long TR (10s). However, the SNR for MT-VASO at 7T increases 7–30% compared to 3T, with greater improvement at short and long TRs. It is not trivial to experimentally compare SNR across scanners since many aspects of the hardware may be different. The identical 3D MT-VASO scans (same readout bandwidth, blood nulling TI adjusted with field strength) performed on our 3T system yielded GM SNR of 7.4±3.5 (n=2). GM SNR at 7T (Table 1) was 156±21% better than 3T (P<0.01).

Table 4.

Comparison of residual GM signals at blood nulling TI and SNR at 3T and 7T. ^*

TR (s)		2.5	4	5	8	10
Residual GM signal, 3T (S/S01, %)	VASO	3.0	10.4	14.2	19.4	20.3
	MT-VASO	17.2	16.4	18.4	22.2	23.0
	Enhancement2 (%)	484.1	57.2	29.0	14.5	13.4
Residual GM signal, 7T (S/S0, %)	VASO	0.5	2.3	3.7	6.9	8.0
	MT-VASO	9.6	8.3	8.6	10.2	11.1
	Enhancement (%)	1786.3	261.0	132.0	48.6	37.6
Residual GM signal ratio, (7T/3T, %)	VASO	17.3	22.2	25.9	35.4	39.7
	MT-VASO	55.8	50.9	46.6	46.0	48.1
SNR ratio, (7T/3T, %)3	VASO	40.3	51.7	60.5	82.7	92.6
	MT-VASO	130.3	118.7	108.8	107.3	112.3

^*Simulations based on the same 3D fast GRE VASO/MT-VASO sequence described in Methods. Assume that the same MT effect (9.1%, Table 1) can be induced at both fields.

¹S represents the residual signal at the blood nulling TI. S₀ is the equilibrium MR signal of the corresponding tissue (i.e. water density C is included).

²Enhancement defined as the percentage of signal increase in MT-VASO as compared to that in the conventional VASO.

³Assuming that the intrinsic SNR of MR images increases linearly with field strength.

Discussion

We implemented three technical improvements for VASO MRI in order to optimize the sequence for use at 7T. First, to compensate for the tissue signal loss at blood nulling TI due to the diminishing T₁ difference between blood and tissue, we adopted the MT-VASO approach, which was previously developed at 3T, and has been demonstrated to be able to improve SNR and CNR in GM for approximately 40% for VASO MRI (12). This enhancement was much greater at 7T, giving SNR and CNR improvements in GM of more than 150% compared to conventional VASO.

The imaging sequences commonly used for VASO MRI in humans at 3T include EPI, TSE (6) and GRASE (13,15). At 7T, EPI suffers from more severe geometrical distortion and signal void artifacts, especially in tissue regions close to large air cavities (caused by different magnetic susceptibility between tissue and air, Fig. 3a). While the TSE sequence is much less sensitive to such artifacts (6), its long RF refocusing pulse train with high FAs (normally>100°) produces high power deposition (SAR), becoming a major constraint at 7T. For the resolution in this study (2mm isotropic), a 3D TSE sequence at 7T with TR=4s, TE=7ms (shortest), partial Fourier fraction=0.7×0.7 and refocusing angle=100°, can only accommodate approximately 6 slices within the FDA SAR limit. The GRASE sequence combines the advantages of EPI and TSE, and choice of appropriate parameters can balance the effects of geometrical distortion and power deposition. In several recent studies at 3T (13,15,26), GRASE has become the method of choice for VASO MRI. Unfortunately, our attempt to use GRASE at 7T for the current resolution and coverage (2mm isotropic voxel, 21 slices) while keeping SAR under the FDA limit did not produce images of satisfactory quality and SNR. The 3D fast GRE sequence employed in this study, which is also used in high resolution anatomical scans (MP-Rage), was able to furnish images with much less geometrical distortion (vs. EPI, as shown in Fig. 3b) and reasonable SNR. Since low flip angles (FA=7° here) are typically used in fast GRE, its power deposition is small compared to TSE and GRASE, and the highest SAR in our 3D fast GRE VASO sequence with MT and MSDE was smaller than 2.1W/kg, about 66% of the FDA limit. A 3D sequence rather than a multi-slice readout was chosen in this study, mainly because lower gradient strength is required for 3D slab selection and crosstalk between adjacent slices due to imperfect slice selection is reduced in 3D sequences (55), and TI is the same for all slices in 3D (ky=kz=0) but different for each slice in a multi-slice readout.

The BOLD effect increases at 7T due to greater magnetic susceptibility effects at higher field. The positive BOLD signal change during neuronal activation contaminates the small negative VASO signal change, which may lead to a substantial underestimate of CBV change. Our data (Table 2 and Figs. 3c,d,f) show that more activated voxels and larger signal changes were measured after extrapolating the original GE-EPI images to effective TE=0, at which the T₂^* weighting is removed and BOLD effects should be minimized. This indicates that at 7T, even with a relatively short TE (15ms, shortest possible with current resolution and imaging parameters), GE EPI VASO still suffers from substantial BOLD contamination. The 3D fast GRE sequence allows using very short TE (1.8ms), which minimizes such BOLD contamination. Our data (Table 2, Figs. 3d,e,f) show that short-TE 3D fast GRE VASO produced comparable signal changes and a comparable number of activated voxels to the extrapolated (TE=0) GE EPI data, implying negligible BOLD contamination. To further confirm this, functional scans were performed using only the 3D fast GRE readout (no inversion and others). No significant positive or negative signal changes (Fig. 3f) were detected.

The non-steady-state blood spin effects in VASO MRI become substantial at 7T when a head coil is used for RF transmit. Previous studies (11,35) have demonstrated that type I and II non-steady-state blood spins would furnish a larger (more negative) VASO signal change during vasodilation, while type III spins have the opposite effect. Our data in Fig. 4 and Table 3 show that with magnetization reset, VASO signal changes dropped from −3.98±0.75% to −2.94±0.61%, indicating that type I non-steady-state spins were suppressed. This result is consistent with a previous study (35) at 3T. After adding the MSDE module in this study, the VASO signal changes further decreased to −2.04±0.19%, which indicates that effects from type II non-steady-state spins were also suppressed. Note that only spins reaching the imaging volume before acquisition contribute to the measurable MR signal. Therefore, type III spins are expected to have much less impact than the other two types, as they enter the transmit coil at a later time and thus are less likely to arrive at the imaging volume before acquisition. If there were some type III spins flowing fast enough to contribute, their velocities should be greater than those of type II spins, and thus should be suppressed by the crushing gradients in MSDE. The VASO signal change measured with magnetization reset and MSDE agrees reasonably with the calculations in a recent theoretical study (54), which predicted about −1.5% VASO signal change at 7T assuming a 25% CBV increase during activation.

The T₂ preparation pulses in MSDE may introduce some unwanted signal decay and BOLD effects due to additional T₂ weighting. With a short effective TE (i.e. duration of MSDE, 15ms here), such signal decay was barely noticeable, as shown in Figs. 4a–d. The potential T₂-weighted BOLD effects can be evaluated by comparing results from scans (b, no T₂ preparation) and (c, with T₂ preparations pulses but no crushing gradients in between) in Experiment set II. The number and pattern of activated voxels, as well as the signal changes and time courses from common voxels (Table 3, Fig. 4) in these two scans were all comparable (P>0.1). This implies that there was little BOLD effect resulting from the T₂ preparation module. A reasonable explanation is that when a short effective TE (15ms, closer to venous blood T₂, but much shorter than tissue T₂ at 7T) is used, the T₂-weighted BOLD effect is mainly in the intravascular compartment (48,56), which is nulled in VASO MRI.

The intrinsic SNR of MRI at 7T is expected to be approximately 2.3 times of the SNR at 3T (proportional to field strength). Unfortunately, for VASO MRI, the diminishing T₁ difference between blood and tissue at 7T reduces the residual tissue signal (thus SNR) at blood nulling time. The simulations in Table 4 show that the SNR improvement for VASO MRI at 7T is TR dependent. With the MT-VASO technique, 7T still furnishes 7–30% higher SNR compared to 3T. Our experimental results with identical 3D fast GRE MT-VASO scans on both fields showed much greater SNR improvement at 7T (156%). This is possibly due to the following factors: 1) the 32-channel receiving coils used on the two scanners are from different manufacturers (3T: In Vivo; 7T: Nova Medical) and the coil elements are closer to the head at 7T; 2) a body coil and a quadrature head coil were used for RF transmit on 3T and 7T scanners, respectively; 3) the gradient systems are also slightly different; 4) RF wavelength reduces at higher field, leading to enhanced RF propagation, reflection and interference effects. Such wave behavior translates into more distinct coil sensitivities, and thus yields lower (favorable) geometry factors (g factors) for parallel imaging. Therefore, the same reduction factor (SENSE factor) results in less SNR loss at higher field (57–60). These factors could by themselves furnish some difference in SNR.

Conclusions

We demonstrated the feasibility of performing VASO MRI in human brain at 7T. The MT-VASO technique, originally proposed at 3T, magnifies the difference between tissue and blood signals during inversion recovery, thereby boosted SNR and CNR by more than 150% compared to conventional VASO at 7T. A 3D fast GRE sequence was used to mitigate geometrical distortion, power deposition, and the positive BOLD contamination in VASO MRI. The non-steady-state blood spin effects, which becomes substantial at 7T where only head coil is available for RF transmit, were suppressed using magnetization reset and MSDE approaches. The implemented 3D fast GRE MT-VASO sequence can be used for non-invasive high resolution (2mm isotropic) CBV based fMRI in human brain at 7T. Although these technical improvements were originally driven by the particular challenges at higher field, they can be used at other field strength as well. Both simulations and experiments with the implemented sequence on 3T and 7T showed the benefit of increased SNR for VASO MRI at higher field.