FULL PAPER Magnetic Resonance in Medicine 00:00–00 (2015)
Quantitative and Functional Pulsed Arterial Spin Labeling in the Human Brain at 9.4 T Jonas Bause,1,2 Philipp Ehses,1,3 Christian Mirkes,1,3 G. Shajan,1 Klaus Scheffler,1,3 and Rolf Pohmann1* by a number of studies performed at 7 T with diverse ASL preparations and signal sampling strategies (5–8). Some of these studies used the pulsed arterial spin labeling (PASL) technique known as flow-sensitive alternating inversion recovery (FAIR) (9,10), which has comparatively low power deposition and is virtually unsusceptible to magnetization transfer effects. The feasibility of quantitative FAIR-ASL at ultrahigh fields was confirmed by a comparative study performed at 3 T and 7 T, where an increase in temporal SNR of 62% and a 78% higher image SNR was found for the data obtained at 7 T (11). However, the implementation of PASL at field strengths above 3 T is not straightforward. The FAIR technique uses a nonselective inversion pulse to create the label and a selective inversion around the image plane for the control condition. The commonly used signal models for quantification of perfusion assume that a spatially well-defined bolus is created during the labeling phase of the sequence (12). On ultrahigh-field scanners, however, the labeling is typically performed with a head coil that produces a spatially limited transmit field (Bþ 1 ) that weakens toward the inferior end of the coil. A label created with such a coil will result in an underestimation of perfusion if not considered in the model. This source of error can be avoided by spatially confining the labeling slab to a region with sufficient Bþ 1 (13), which can also help to reduce spatial variations of the inversion efficiency caused by static magnetic field (B0) inhomogeneities. In addition, the inversion efficiency as well as the slab profile of the labeling and the control pulses may suffer due to relaxation during the inversion pulse, which can lead to systematic differences between the two conditions (14). The resulting offset of static tissue signal can be reduced by an in-plane signal saturation before or after the inversion pulse (15,16). The stronger B0 offresonances that are associated with high field strengths affect the PASL preparation, but also the ASL image acquisition itself where it may cause distortions and signal dropouts. This is especially the case for the most commonly used gradient echo (GRE) echo planar imaging (EPI) method. Above all, the increased specific absorption rate (SAR) at ultrahigh fields can constrain the execution of ASL sequences due to safety limitations. Because all these restrictions scale with field strength, PASL at 9.4 T can be expected to be even more challenging than at 7 T. In this study, the first results of pulsed ASL with FAIR at 9.4 T are presented. For the implementation, an adiabatic inversion pulse and a multipulse inplane presaturation scheme were optimized. Both were
Purpose: The feasibility of multislice pulsed arterial spin labeling (PASL) of the human brain at 9.4 T was investigated. To demonstrate the potential of arterial spin labeling (ASL) at this field strength, quantitative, functional, and high-resolution (1.05 1.05 2 mm3) ASL experiments were performed. Methods: PASL was implemented using a numerically optimized adiabatic inversion pulse and presaturation scheme. Quantitative measurements were performed at 3 T and 9.4 T and evaluated on a voxel-by-voxel basis. In a functional experiment, activation maps obtained with a conventional blood-oxygen-level-dependent (BOLD)-weighted sequence were compared with a functional ASL (fASL) measurement. Results: Quantitative measurements revealed a 23% lower perfusion in gray matter and 17% lower perfusion in white matter at 9.4 T compared with 3 T. Furthermore almost identical transit delays and bolus durations were found at both field strengths whereas the calculated voxel volume corrected signal-to-noise ratio was 1.9 times higher at 9.4 T. This result was confirmed by the high-resolution experiment. The functional experiment yielded comparable activation maps for the fASL and BOLD measurements. Conclusion: Although PASL at ultrahigh field strengths is limited by high specific absorption rate, functional and quantitative perfusion-weighted images showing a high degree of detail can be obtained. Magn Reson Med 000:000–000, C 2015 Wiley Periodicals, Inc. 2015. V Key words: ASL; perfusion; ultrahigh field; quantification; fMRI; 9.4 T
INTRODUCTION Arterial spin labeling (ASL) can benefit twofold from ultrahigh-field MRI (1). First, the higher field strength leads to an increase in the intrinsic signal-to-noise ratio (SNR) (2,3), and second, a higher and longer-lasting perfusion-related signal change can be measured due to the longer longitudinal relaxation times (4). The great potential of ultrahigh-field ASL in humans was demonstrated 1 Max Planck Institute for Biological Cybernetics, High-Field Magnetic Reso€bingen, Germany. nance Center, Tu 2 Graduate Training Center of Neuronal Sciences, International Max Planck €bingen, Tu €bingen, Germany. Research School, University of Tu 3 €bingen, Department of Biomedical Magnetic Resonance, University of Tu €bingen, Germany Tu
*Correspondence to: Rolf Pohmann, Ph.D., Max Planck Institute for Biological Cybernetics, High-Field Magnetic Resonance Center, Spemannstr. 41, €bingen, Germany. E-mail: [email protected] 72076 Tu Received 13 October 2014; revised 4 February 2015; accepted 5 February 2015 DOI 10.1002/mrm.25671 Published online 00 Month 2015 in Wiley Online Library (wileyonlinelibrary. com). C 2015 Wiley Periodicals, Inc. V
1
2
evaluated by measuring their spatial selectivity and efficiency. In a quantitative experiment, multislice perfusion-weighted images (PWI) were acquired at several inversion times (TIs) and at two different field strengths (3 T and 9.4 T) to analyze the characteristics of the ASL signal. One possible application of ASL at 9.4 T is the detailed investigation of physiological mechanisms related to neuronal activity (9). Furthermore, functional ASL (fASL) offers a higher spatial accuracy than GREbased blood-oxygen-level-dependent (BOLD) imaging, since it is not affected by draining vein signals (17). For this reason, the feasibility of fASL with FAIR at 9.4 T was investigated. The obtained activation patterns were compared with those of a standard BOLD-weighted GREEPI. Finally, high-resolution PWIs were acquired at 3 T and 9.4 T to demonstrate the current capabilities of PASL at 9.4 T. METHODS Six healthy volunteers (mean age, 32 6 7 years; female, n ¼ 2; male, n ¼ 4) were measured on a 9.4 T MRI scanner (Siemens Healthcare, Erlangen, Germany) using a custom-built, 16-channel transmit array combined with a 31-channel receive helmet (18). The transmit coil elements had a coverage of 182 mm in z-direction and were driven in circular polarized mode in all experiments. Additionally, four of the subjects were scanned with the same protocol on a clinical 3 T system (Siemens Healthcare, Erlangen, Germany) that was running the same software version as the 9.4 T scanner. On this system, the body coil was used for transmission, and a 32-channel receive array was used for signal reception. The study was approved by the local ethics committee, and all subjects gave informed, written consent before being scanned. All obtained raw data were reconstructed offline with self written routines in MATLAB (MathWorks, Natick, Massachusetts, USA) in order to have the same post-processing and filtering procedures for all measurements. Inversion Pulse and In-Plane Presaturation Optimization A time-resampled frequency offset corrected inversion (TR-FOCI) pulse with a duration of 10 ms and a design thickness of 50 mm (full width, half maximum) was optimized using a genetic algorithm (19). The target shape of the slab profile had a transition band (jMzj < 0.99) of 2 mm. During the optimization, the inversion profile of each parameter set was evaluated at four different Bþ 1 values (3.50, 5.75, 7, and 20 mT) in order to take the stronger Bþ 1 inhomogeneities into account. SAR was not included in the optimization process because the required pulse voltage mainly depended on the inversion efficiency in regions with low transmit field strength for a given pulse duration. Furthermore, influences of B0 off-resonances and relaxation processes during the pulse were neglected. To determine the effects of static field inhomogeneities and relaxation, simulations with the minimum required transmit field strength for inversion (Bþ 1 ¼ 5.4 mT) and for Bþ 1 ¼ 20 mT were performed. In these simulations, a B0 off-resonance offset of 200 Hz and a gray matter trans-
Bause et al.
verse relaxation time of 36 ms were assumed based on relaxivity measurements in rats at 9.4 T (20). Longitudinal relaxation during the inversion was neglected, because the influence on the slab profile is small at ultrahigh fields when comparatively short pulses are used. The actual inversion profile and efficiency of the TR-FOCI pulse was assessed via an inversion recovery measurement in a spherical phantom (diameter ¼ 150 mm) with typical relaxation times of brain tissue at 9.4 T (T1 ¼ 2100 ms, T2 ¼ 40 ms). Image acquisition was performed using a GRE-EPI sequence with the following parameters: echo time (TE) ¼ 12 ms; repetition time (TR) ¼ 10 s; nominal flip angle ¼ 60 ; 15 repetitions with different TIs from 50 to 6000 ms; generalized autocalibrating partially parallel acquisitions (GRAPPA) ¼ 4; partial Fourier factor ¼ 6/8; voxel size ¼ 1.24 1.24 4 mm3; number of slices ¼ 1. The inversion efficiency (a) was then obtained by fitting the image data to the inversion recovery signal equation ! TI ; [1] MðTIÞ ¼ M0 1 2a exp T1 where M0 denotes the equilibrium magnetization and T1 the longitudinal relaxation time. In addition, the performance of the selective and nonselective pulse was measured in a single subject with a sequence having the same TIs but the following parameters: TE ¼ 9.7 ms; TR ¼ 11 s; nominal flip angle ¼ 90 ; GRAPPA ¼ 3; partial Fourier factor ¼ 6/8; voxel size ¼ 2 2 3 mm3; number of slices ¼ 4 (gap ¼ 21 mm). For the profile measurements, the inversion slab was oriented axially, while the readout was performed in the sagittal direction with minimum temporal spacing between the acquisitions of adjacent slices. B0 shimming was performed for the region of the selective inversion slab. The same peak pulse voltage was chosen for the TR-FOCI pulse as that used in the ASL experiments. Moreover, the transmit field strength was measured in the phantom and in vivo using the actual flip angle imaging technique (21,22). For the in-plane presaturation, a water suppression enhanced through T1 effects (WET) scheme with four blocks (16,23) was implemented right before the inversion pulse. Each block had a duration of 15 ms and consisted of a radiofrequency (RF) pulse followed by a spoiler gradient. The flip angles for each block were numerically optimized with MATLAB by minimizing the residual longitudinal magnetization at the end of the last spoiler gradient for Bþ variations of 6 50% (step 1 width ¼ 5%) and a T1 range of 800–2800 ms (step width ¼ 100 ms). In the simulations, complete spoiling of transverse magnetization was assumed after each block and relaxation during the RF pulses was neglected. The profile of a 50-mm-thick WET saturation slab was measured in the spherical phantom described previously by acquiring two images with and without presaturation. As in the ASL experiments, hamming-filtered, sincshaped RF pulses (duration ¼ 11 ms; time bandwidth product ¼ 12) were used to achieve narrow transition bands while keeping the SAR contribution limited. In the slab profile mapping sequence, individual segments of a spoiled GRE were acquired immediately after signal
Quantitative and Functional Pulsed ASL at 9.4 T
3
preparation. Further parameters were: total number of segments ¼ 6 (lines per segment ¼ 8); TE ¼ 3 ms; TR ¼ 7 ms; nominal flip angle ¼ 10 ; voxel size ¼ 1.28 0.64 4 mm3; delay after each segment ¼ 10 s. The saturation efficiency was calculated as Saturation efficiency ¼
ðSIno sat SIsat Þ ; SIno sat
[2]
where SIno sat and SIsat are signal intensities for the image acquired without and with presaturation, respectively. The in vivo performance of a 40-mm saturation band was measured using the same sequence as for the inversion efficiency assessment but with minimal delay between the end of the presaturation and the beginning of the readout. Quantitative ASL Three subjects (mean age, 33 years) were measured at 3 T and 9.4 T to investigate the comparability of quantitative perfusion measurements at high and ultrahigh field strengths. The ASL protocol consisted of perfusion measurements acquired at seven different TIs (TI ¼ 400– 2800 ms; step size ¼ 400 ms) using a FAIR-prepared sequence with WET in-plane presaturation and GRE-EPI readout. To minimize systematic perfusion errors caused by subject motion or system drift, the order of the TIs was randomized for all experiments. For each TI, 118 volumes (9 slices, thickness ¼ 3 mm, gap ¼ 1 mm) were acquired with a TR of 4 s. A further TR reduction was not possible at 9.4 T due to our SAR restrictions, which are more conservative than the International Electrotechnical Commission regulations for the “normal” operation mode by a factor of 2. Approximately 75% of the total power deposition was caused by the inversion pulse and 8% was caused by the in-plane presaturation. M0 was measured prior to the first inversion at the beginning of each sequence after a relaxation delay of 12 s. Further imaging parameters at 3 T were: TE ¼ 18 ms; flip angle ¼ 90 ; in-plane resolution ¼ 3 mm2; partial Fourier factor ¼ 6/8; GRAPPA ¼ 2. Further imaging parameters at 9.4 T were: TE ¼ 17 ms; nominal flip angle ¼ 80 ; inplane resolution ¼ 2 mm2; partial Fourier factor ¼ 6/8; GRAPPA ¼ 3. The higher in-plane resolution at 9.4 T was necessary to reduce intravoxel dephasing and image blurring caused by strong susceptibility effects at ultrahigh fields. A bipolar gradient (Venc ¼ 40 mm/s) between excitation and the start of the readout served to attenuate signal from fast-flowing intravascular blood (24). The optimized inversion pulse with a design thickness of 50 mm was used for the FAIR control condition, resulting in a nominal gap of 7.5 mm between the imaging region and the edge of the selective inversion region. The thickness of the labeling slab was limited to 200 mm to achieve the same bolus duration on both systems. For the determination of T1, an inversion recovery dataset was acquired using the same sequence as for the inversion efficiency determination but with the imaging parameters of the ASL measurement. After reconstruction, motion correction was performed for the tag and control images separately and the T1 of
tissue was obtained from a fit of the inversion recovery data to equation [1]. Each M0 scan and the T1 map were realigned to the M0 scan acquired in the first measurement, and the obtained transformation matrices were also applied to the other frames of the ASL measurement. Both motion correction and realignment were performed using adapted algorithms of SPM8 (University College London, London, UK). Thereafter, PWIs were generated by calculating the difference of the mean control from the mean label images. The two outermost slices were clipped by the motion correction algorithm in most measurements and were therefore excluded from further analysis. The remaining slices were fitted voxel-wise to the general kinetic model for pulsed ASL (12) with the fit variables blood flow (f), transit delay (Dt), and bolus duration (t). For both field strengths, an inversion efficiency of 0.95 and a blood/brain coefficient of 0.95 mL/g (25) was assumed. The longitudinal relaxation time of arterial blood (T1 blood) was assumed to be 1.6 s at 3 T (26) and 2.2 s at 9.4 T. The 9.4 T value was determined by assuming a cubic root increase of T1 with field strength (27). T1 of tissue was extracted from the previously calculated T1 maps. The obtained blood flow and transit delay were averaged for the T1 values of gray and white matter (T1 GM ¼ 1.2–1.7 s at 3 T and 1.9–2.4 s at 9.4 T; T1 WM ¼ 0.8– 1.0 s at 3 T and 1.2–1.7 s at 9.4 T). Voxels that showed a fitted perfusion of more than 150 mL/100 g/min were assumed to contain uncrushed signal from vessels and were excluded from the masks. Temporal and image SNR of the perfusion-weighted data was calculated for the combined gray and white matter regions of the PWIs obtained at a TI of 1.6 s using the methods described by Feinberg et al. (28) and was averaged over the subjects for each field strength. fASL Three subjects were measured with FAIR-ASL and a standard BOLD-weighted GRE-EPI sequence. Duration, spatial coverage, image resolution, and number of measurements were identical for the two sequences (number of slices ¼ 9; number of volumes ¼ 112; isotropic resolution ¼ 2 mm; slice gap ¼ 0.3 mm; GRAPPA ¼ 3; partial Fourier factor ¼ 6/8; nominal flip angle ¼ 100 ). The minimum achievable echo time (TE ¼ 11 ms) and no flow crusher were used for the ASL sequence. Signal preparation was performed as in the quantitative experiment (TI ¼ 1700 ms). The BOLD-weighted EPI protocol had the same image acquisition settings but a TE of 21 ms in order to achieve a stronger T2 weighting. During the experiments, the subjects were asked to perform seven repetitions of a bilateral finger-tapping task (32/32 s of resting/tapping). A high-resolution (0.8 mm isotropic) GRE measurement was performed for coregistration. No motion correction was performed for the ASL data since the algorithm detected artificial motion in the label images in two of the three subjects evoked by the strong paradigm related signal change. The corresponding control measurements, however, showed a maximum displacement of smaller than half a voxel during the sequence in all three subjects. Functional analysis was
4
Bause et al.
performed with the FMRIB Expert Analysis Tool (FEAT) and coregistered to the anatomical reference with the FMRIB Linear Image Registration Tool (FLIRT) of the FSL package (29). High-Resolution ASL To assess the potential of high-resolution ASL at 9.4 T, a FAIR prepared GRE-EPI (TI ¼ 2 s) with a spatial resolution of 1.05 1.05 2 mm3 was acquired in two subjects. For each subject, 162 volumes consisting of seven slices (gap ¼ 2 mm) were measured in 11 min. Further imaging parameters were: TE ¼ 21 ms; nominal flip angle ¼ 90 ; GRAPPA ¼ 4; partial Fourier factor ¼ 6/8; no flow crusher. For comparison, the same measurement was performed at 3 T. Due to the lower T2 -related signal losses and reduced distortions at 3 T, we were able to decrease the parallel imaging factor to GRAPPA ¼ 3, which increased the TE to 23 ms. Furthermore, it would have been possible to shorten the TR at 3 T due to lower SAR. For this reason, a TR of 3 s was assumed for the 3 T measurements, resulting in a scan time decrease from 11 min to 8 min. In order to allow a fair comparison between the two field strengths, the number of averages that were used to calculate the 9.4 T PWIs was reduced accordingly (from 80 to 60). As an anatomical reference, GRE images with a resolution of 0.6 0.6 1 mm3 were acquired from the same volume at 9.4 T.
FIG. 1. a: Effect of B0 off-resonances on the inversion slab profile for two different transmit field strengths. The deformation of the profile is stronger for lower Bþ 1 levels. b: Simulated influence of transverse relaxation during the pulse. A decrease in inversion efficiency and selectivity can be observed, which becomes even more prominent at higher transmit field strengths.
saturation efficiency can also be observed outside of the saturation slab and close to the edge of the phantom due to strong B0 inhomogeneities. The Bþ 1 field measured in vivo at four sagittal positions is depicted in Figure 3a. Because an inversion pulse voltage of approximately 500 V was used for all for in vivo experiments, complete inversion is expected in large parts of the brain except for the inferior part of the cerebellum (white arrow). This is in agreement with the fitted efficiency of the nonselective inversion pulse (Fig. 3b, top row). The bottom row of Figure 3b shows the efficiency profile of the selective pulse. In these images, no off-resonance–related distortions of the slab profile can be observed. Furthermore, all inversion efficiency maps obtained from the in vivo experiment show again a T2like contrast. Even though the transmit field distribution in the brain was less homogenous than in the phantom, a high degree of signal suppression was achieved with the optimized presaturation (Fig. 3c).
RESULTS Inversion Pulse and In-Plane Presaturation Optimization The design parameters of the optimized TR-FOCI pulse are listed in Table 1. According to the simulation, its theoretical inversion efficiency was a > 0.99 in 80% of the nominal slice thickness for a transmit field strength greater than 5.4 mT when the effects of B0 and relaxation were neglected. Figure 1a shows the simulated influence of static magnetic field inhomogeneities on the slab profile. A degradation of the inversion profile is visible for the simulation with Bþ 1 ¼ 5.4 mT, but not for the higher transmit field strength of 20 mT. In contrast, the effect of T2 on the profile declined for the lower Bþ 1 level (Fig. 1b). The fitted efficiency in the phantom confirmed the robustness of the pulse against Bþ 1 inhomogeneities but also the strong effect of transverse relaxation on the slab profile (Fig. 2b). The measured transition band (0.09 < a < 0.9) had a thickness of 14 mm, which reduced the actual inversion slab thickness (a > 0.9) to 36 mm (design thickness 50 mm). Figure 2c shows the profile of the 50-mm-thick presaturation slab in the phantom using the optimized flip angles of 180 , 89.5 , 69.9 , and 116.1 . A high degree of signal suppression was achieved for the entire saturation band despite the inhomogeneous transmit field. However, an increase of
Quantitative ASL PWIs at different TIs (five out of nine measured slices) are shown in Figure 4 (top). Although a flow crusher of 40 mm/s was used in this experiment, high signal can still be observed in some areas of the images acquired at short TIs. This indicates that not all signal of fast flowing
Table 1 Parameters Defining the Shape of the Optimized TR-FOCI Pulse Amax
w
r1
r2
r3
r4
r5
m
b
t1
t2
31.52
0.28
0.49
0.13
0.32
0.83
1.00
0.80
5.34
0.46
0.32
For a detailed description of the individual parameters, see Hurley et al. (19).
Quantitative and Functional Pulsed ASL at 9.4 T
5
23% lower in gray matter and 17% lower in white matter, respectively. The fitted transit delay of the same subject is shown in the bottom row of Figure 4. For both field strengths, the highest transit delay was observed in the occipital lobe and in deep white matter. However, unrealistically low transit delay values were obtained for the most inferior and superior slices, probably due to remaining static tissue signal which was higher in the data obtained at 9.4 T than in the 3 T measurements. Figure 5 shows the increase in mean gray matter transit delay for the different slices. The bolus durations at 3 T were 1.27 s for gray matter and 1.73 s for white matter. The values obtained at 9.4 T were slightly higher for gray matter (1.44 s) but lower for white matter (1.52 s). The low value in white matter is due in part to the poor fit in the most inferior slices, as shown by the fact that the bolus duration increased to 1.65 s if the most inferior and most superior evaluated slices were neglected. The SNR of the PWIs at a TI of 1.6 s of all three subjects in gray and white matter (slices 2–8) was relatively similar (5.26 6 0.5 at 3 T; 4.47 6 0.11 at 9.4 T), despite the large difference in voxel volume (factor of 2.3). For the temporal SNR, the obtained values were almost identical (0.69 6 0.05 at 3 T; 0.63 6 0.05 at 9.4 T). fASL Figure 6a shows the mean PWI of an exemplary slice of two subjects during the rest and stimulus periods. A strong increase in perfusion signal during the fingertapping task was seen in the motor cortex of both hemispheres and the supplementary motor area. The FIG. 2. a: Measured transmit-field distribution in the phantom. b: Profile along the gray line (left) of the fitted efficiency confirms the predicted loss in selectivity and efficiency caused by relaxation processes during the pulse (right). The blue box indicates the transition band where 0.09 < a < 0.9. c: The optimized in-plane presaturation scheme has a good selectivity and shows a high degree of signal suppression despite the strong transmit field inhomogeneities (left). Some signal suppression can also be observed toward the edges of the phantom (right) due to stronger B0 variations.
spins was spoiled by the bipolar gradient, which may be due to different vessel orientations. The higher in-plane resolution of the 9.4 T measurements performed leads to a better perfusion contrast due to reduced partial voluming when compared with the 3 T measurements. On the other hand, the signal change over TI is more obvious in the 3 T data due to the shorter T1 of tissue at the lower field strength. In addition, an increase in signal arising from draining veins at TIs 1.6 s can be observed in the images acquired at the lower field strength. The average gray matter perfusion in all three subjects and all evaluated seven slices was 55.8 6 4.9 mL/ 100 g/min for the measurements performed at 3 T and 42.8 6 3.3 mL/100 g/min for the 9.4 T datasets. For white matter perfusion, the same trend was observed, with 29.7 6 3.4 mL/100 g/min at 3 T and 24.8 6 0.6 mL/ 100 g/min at 9.4 T. Thus, the mean fitted perfusion values obtained at the higher field strength were approximately
FIG. 3. a: Measured transmit field (four sagittal slices of the 3D volume are shown) at 9.4 T. The white arrow indicates the area with very low transmit field strength in the cerebellum. b: Fitted efficiency of the selective and nonselective inversion in the brain. c: Measured slab profile of the WET saturation with optimized flip angles.
6
Bause et al.
FIG. 4. PWIs acquired at multiple TIs at 9.4 T (top left) and 3 T (top right) of one subject. The resolution of the images is 2 2 3 mm3 and 3 3 3 mm3 for the 9.4 T and 3 T data, respectively. Only five out of nine acquired slices are shown. The middle and bottom rows show perfusion maps and transit delay distribution obtained from the fit of the multiple TI data to the general kinetic model.
activation maps are overlaid on the corresponding anatomical reference in Figure 6b, together with the corresponding z-score maps of the reference BOLD dataset. Figure 6c shows the time course of the median perfusion (orange) and BOLD signal (blue) in the 300 voxels with the highest z-score in each of the datasets for subject 1. A reduced temporal signal stability and poststimulus undershoots can be observed in the time course of the ASL dataset. High-Resolution ASL Figure 7 depicts five of seven slices of the anatomical reference images as well as the high-resolution PWIs of
both subjects obtained by averaging all acquired volumes. A great amount of detail in the cortical structure is visible in the ASL measurements due to the high resolution and the consequently reduced partial volume effects. A clear difference between gray matter sulci and the intermediate surrounding CSF was observed, especially in the occipital lobe. However, a higher noise level was visible in the medial part of the PWIs, which is probably due to the high GRAPPA acceleration factor and the declining perfusion signal at the long TI. An example slice of the PWIs acquired at 3 T is shown in Figure 8 (left). The image obtained at the higher field strength shows a better contrast and more details (Fig. 8, right) even though it was calculated with a reduced
Quantitative and Functional Pulsed ASL at 9.4 T
FIG. 5. Transit delay of the bolus in gray matter (mean and standard deviation over all subjects) at 3 T (gray) and 9.4 T (black).
number of averages to take the possible reduction in repetition time at 3 T into account. DISCUSSION The in vivo and phantom evaluation of the developed inversion pulse and the presaturation scheme confirmed their theoretical stability against transmit field variations. However, the measurements also showed that the efficiency and selectivity of the inversion pulse was heavily affected by transverse relaxation. Although the influence of T2 can be diminished by reducing the pulse power, this in turn increases the impact of offresonances. The slab profile can then only be improved significantly at the cost of higher SAR. Often, longer RF
FIG. 6. a: Mean perfusion signal (TI ¼ 1700 ms) in an exemplary slice during rest and activation measured in two of the three subjects. The perfusion change during the finger-tapping paradigm is visible in the motor cortex of both subjects. b: Overlay of the fASL and BOLD activation maps on the anatomical reference. Note the different scaling of the color bars. c: Median signal time course of the 300 voxels that showed the highest z-score in the functional ASL (orange) and the BOLD measurement (blue) of subject 1. The gray boxes symbolize the stimulation periods.
7
pulses are used to reduce the required pulse voltage and thus SAR; however, this can increase the susceptibility to T2 relaxation or static field inhomogeneities. At worst, the FAIR condition, which requires that the selective and nonselective pulse have the same inversion efficiency in the image region, is violated (14). In the quantitative experiment, high perfusion signal was observed in regions with draining veins at TIs 1.6 s caused by the rather thin control slab, which also led to a labeling of blood superior to the imaging region. Although this effect occurred at both field strengths, it was more pronounced in the images acquired at 3 T (Fig. 4, top) which can be explained by the very short T2 of venous blood at 9.4 T (30). For this reason, a perfusion signal in the sagittal sinus is also visible in the 9.4 T PWIs of the functional experiment, which were acquired with a shorter TE. There may be multiple reasons for perfusion differences that were observed between the quantitative measurements at 3 T and 9.4 T. First, a single compartment model was used for fitting the multiple TI data. This model does not consider T2 relaxation during the GREEPI readout, which can result in an underestimation of perfusion, especially at high field strengths (31). Second, the higher spatial resolution of the 9.4 T measurements reduces partial volume effects in the T1 maps and the perfusion images. Almost identical gray matter transit delays were found in slices 3–7 for the two field strengths (Fig. 5). Typically, a nominal distance of 1– 2 cm is used between the labeling and imaging slab to
8
Bause et al.
FIG. 7. Mean difference signal (voxel size ¼ 1.05 1.05 2 mm3; TI ¼ 2000 ms) measured at 9.4 T in two different subjects in
Quantitative and Functional Pulsed Arterial Spin Labeling in the Human Brain at 9.4 T Jonas Bause,1,2 Philipp Ehses,1,3 Christian Mirkes,1,3 G. Shajan,1 Klaus Scheffler,1,3 and Rolf Pohmann1* by a number of studies performed at 7 T with diverse ASL preparations and signal sampling strategies (5–8). Some of these studies used the pulsed arterial spin labeling (PASL) technique known as flow-sensitive alternating inversion recovery (FAIR) (9,10), which has comparatively low power deposition and is virtually unsusceptible to magnetization transfer effects. The feasibility of quantitative FAIR-ASL at ultrahigh fields was confirmed by a comparative study performed at 3 T and 7 T, where an increase in temporal SNR of 62% and a 78% higher image SNR was found for the data obtained at 7 T (11). However, the implementation of PASL at field strengths above 3 T is not straightforward. The FAIR technique uses a nonselective inversion pulse to create the label and a selective inversion around the image plane for the control condition. The commonly used signal models for quantification of perfusion assume that a spatially well-defined bolus is created during the labeling phase of the sequence (12). On ultrahigh-field scanners, however, the labeling is typically performed with a head coil that produces a spatially limited transmit field (Bþ 1 ) that weakens toward the inferior end of the coil. A label created with such a coil will result in an underestimation of perfusion if not considered in the model. This source of error can be avoided by spatially confining the labeling slab to a region with sufficient Bþ 1 (13), which can also help to reduce spatial variations of the inversion efficiency caused by static magnetic field (B0) inhomogeneities. In addition, the inversion efficiency as well as the slab profile of the labeling and the control pulses may suffer due to relaxation during the inversion pulse, which can lead to systematic differences between the two conditions (14). The resulting offset of static tissue signal can be reduced by an in-plane signal saturation before or after the inversion pulse (15,16). The stronger B0 offresonances that are associated with high field strengths affect the PASL preparation, but also the ASL image acquisition itself where it may cause distortions and signal dropouts. This is especially the case for the most commonly used gradient echo (GRE) echo planar imaging (EPI) method. Above all, the increased specific absorption rate (SAR) at ultrahigh fields can constrain the execution of ASL sequences due to safety limitations. Because all these restrictions scale with field strength, PASL at 9.4 T can be expected to be even more challenging than at 7 T. In this study, the first results of pulsed ASL with FAIR at 9.4 T are presented. For the implementation, an adiabatic inversion pulse and a multipulse inplane presaturation scheme were optimized. Both were
Purpose: The feasibility of multislice pulsed arterial spin labeling (PASL) of the human brain at 9.4 T was investigated. To demonstrate the potential of arterial spin labeling (ASL) at this field strength, quantitative, functional, and high-resolution (1.05 1.05 2 mm3) ASL experiments were performed. Methods: PASL was implemented using a numerically optimized adiabatic inversion pulse and presaturation scheme. Quantitative measurements were performed at 3 T and 9.4 T and evaluated on a voxel-by-voxel basis. In a functional experiment, activation maps obtained with a conventional blood-oxygen-level-dependent (BOLD)-weighted sequence were compared with a functional ASL (fASL) measurement. Results: Quantitative measurements revealed a 23% lower perfusion in gray matter and 17% lower perfusion in white matter at 9.4 T compared with 3 T. Furthermore almost identical transit delays and bolus durations were found at both field strengths whereas the calculated voxel volume corrected signal-to-noise ratio was 1.9 times higher at 9.4 T. This result was confirmed by the high-resolution experiment. The functional experiment yielded comparable activation maps for the fASL and BOLD measurements. Conclusion: Although PASL at ultrahigh field strengths is limited by high specific absorption rate, functional and quantitative perfusion-weighted images showing a high degree of detail can be obtained. Magn Reson Med 000:000–000, C 2015 Wiley Periodicals, Inc. 2015. V Key words: ASL; perfusion; ultrahigh field; quantification; fMRI; 9.4 T
INTRODUCTION Arterial spin labeling (ASL) can benefit twofold from ultrahigh-field MRI (1). First, the higher field strength leads to an increase in the intrinsic signal-to-noise ratio (SNR) (2,3), and second, a higher and longer-lasting perfusion-related signal change can be measured due to the longer longitudinal relaxation times (4). The great potential of ultrahigh-field ASL in humans was demonstrated 1 Max Planck Institute for Biological Cybernetics, High-Field Magnetic Reso€bingen, Germany. nance Center, Tu 2 Graduate Training Center of Neuronal Sciences, International Max Planck €bingen, Tu €bingen, Germany. Research School, University of Tu 3 €bingen, Department of Biomedical Magnetic Resonance, University of Tu €bingen, Germany Tu
*Correspondence to: Rolf Pohmann, Ph.D., Max Planck Institute for Biological Cybernetics, High-Field Magnetic Resonance Center, Spemannstr. 41, €bingen, Germany. E-mail: [email protected] 72076 Tu Received 13 October 2014; revised 4 February 2015; accepted 5 February 2015 DOI 10.1002/mrm.25671 Published online 00 Month 2015 in Wiley Online Library (wileyonlinelibrary. com). C 2015 Wiley Periodicals, Inc. V
1
2
evaluated by measuring their spatial selectivity and efficiency. In a quantitative experiment, multislice perfusion-weighted images (PWI) were acquired at several inversion times (TIs) and at two different field strengths (3 T and 9.4 T) to analyze the characteristics of the ASL signal. One possible application of ASL at 9.4 T is the detailed investigation of physiological mechanisms related to neuronal activity (9). Furthermore, functional ASL (fASL) offers a higher spatial accuracy than GREbased blood-oxygen-level-dependent (BOLD) imaging, since it is not affected by draining vein signals (17). For this reason, the feasibility of fASL with FAIR at 9.4 T was investigated. The obtained activation patterns were compared with those of a standard BOLD-weighted GREEPI. Finally, high-resolution PWIs were acquired at 3 T and 9.4 T to demonstrate the current capabilities of PASL at 9.4 T. METHODS Six healthy volunteers (mean age, 32 6 7 years; female, n ¼ 2; male, n ¼ 4) were measured on a 9.4 T MRI scanner (Siemens Healthcare, Erlangen, Germany) using a custom-built, 16-channel transmit array combined with a 31-channel receive helmet (18). The transmit coil elements had a coverage of 182 mm in z-direction and were driven in circular polarized mode in all experiments. Additionally, four of the subjects were scanned with the same protocol on a clinical 3 T system (Siemens Healthcare, Erlangen, Germany) that was running the same software version as the 9.4 T scanner. On this system, the body coil was used for transmission, and a 32-channel receive array was used for signal reception. The study was approved by the local ethics committee, and all subjects gave informed, written consent before being scanned. All obtained raw data were reconstructed offline with self written routines in MATLAB (MathWorks, Natick, Massachusetts, USA) in order to have the same post-processing and filtering procedures for all measurements. Inversion Pulse and In-Plane Presaturation Optimization A time-resampled frequency offset corrected inversion (TR-FOCI) pulse with a duration of 10 ms and a design thickness of 50 mm (full width, half maximum) was optimized using a genetic algorithm (19). The target shape of the slab profile had a transition band (jMzj < 0.99) of 2 mm. During the optimization, the inversion profile of each parameter set was evaluated at four different Bþ 1 values (3.50, 5.75, 7, and 20 mT) in order to take the stronger Bþ 1 inhomogeneities into account. SAR was not included in the optimization process because the required pulse voltage mainly depended on the inversion efficiency in regions with low transmit field strength for a given pulse duration. Furthermore, influences of B0 off-resonances and relaxation processes during the pulse were neglected. To determine the effects of static field inhomogeneities and relaxation, simulations with the minimum required transmit field strength for inversion (Bþ 1 ¼ 5.4 mT) and for Bþ 1 ¼ 20 mT were performed. In these simulations, a B0 off-resonance offset of 200 Hz and a gray matter trans-
Bause et al.
verse relaxation time of 36 ms were assumed based on relaxivity measurements in rats at 9.4 T (20). Longitudinal relaxation during the inversion was neglected, because the influence on the slab profile is small at ultrahigh fields when comparatively short pulses are used. The actual inversion profile and efficiency of the TR-FOCI pulse was assessed via an inversion recovery measurement in a spherical phantom (diameter ¼ 150 mm) with typical relaxation times of brain tissue at 9.4 T (T1 ¼ 2100 ms, T2 ¼ 40 ms). Image acquisition was performed using a GRE-EPI sequence with the following parameters: echo time (TE) ¼ 12 ms; repetition time (TR) ¼ 10 s; nominal flip angle ¼ 60 ; 15 repetitions with different TIs from 50 to 6000 ms; generalized autocalibrating partially parallel acquisitions (GRAPPA) ¼ 4; partial Fourier factor ¼ 6/8; voxel size ¼ 1.24 1.24 4 mm3; number of slices ¼ 1. The inversion efficiency (a) was then obtained by fitting the image data to the inversion recovery signal equation ! TI ; [1] MðTIÞ ¼ M0 1 2a exp T1 where M0 denotes the equilibrium magnetization and T1 the longitudinal relaxation time. In addition, the performance of the selective and nonselective pulse was measured in a single subject with a sequence having the same TIs but the following parameters: TE ¼ 9.7 ms; TR ¼ 11 s; nominal flip angle ¼ 90 ; GRAPPA ¼ 3; partial Fourier factor ¼ 6/8; voxel size ¼ 2 2 3 mm3; number of slices ¼ 4 (gap ¼ 21 mm). For the profile measurements, the inversion slab was oriented axially, while the readout was performed in the sagittal direction with minimum temporal spacing between the acquisitions of adjacent slices. B0 shimming was performed for the region of the selective inversion slab. The same peak pulse voltage was chosen for the TR-FOCI pulse as that used in the ASL experiments. Moreover, the transmit field strength was measured in the phantom and in vivo using the actual flip angle imaging technique (21,22). For the in-plane presaturation, a water suppression enhanced through T1 effects (WET) scheme with four blocks (16,23) was implemented right before the inversion pulse. Each block had a duration of 15 ms and consisted of a radiofrequency (RF) pulse followed by a spoiler gradient. The flip angles for each block were numerically optimized with MATLAB by minimizing the residual longitudinal magnetization at the end of the last spoiler gradient for Bþ variations of 6 50% (step 1 width ¼ 5%) and a T1 range of 800–2800 ms (step width ¼ 100 ms). In the simulations, complete spoiling of transverse magnetization was assumed after each block and relaxation during the RF pulses was neglected. The profile of a 50-mm-thick WET saturation slab was measured in the spherical phantom described previously by acquiring two images with and without presaturation. As in the ASL experiments, hamming-filtered, sincshaped RF pulses (duration ¼ 11 ms; time bandwidth product ¼ 12) were used to achieve narrow transition bands while keeping the SAR contribution limited. In the slab profile mapping sequence, individual segments of a spoiled GRE were acquired immediately after signal
Quantitative and Functional Pulsed ASL at 9.4 T
3
preparation. Further parameters were: total number of segments ¼ 6 (lines per segment ¼ 8); TE ¼ 3 ms; TR ¼ 7 ms; nominal flip angle ¼ 10 ; voxel size ¼ 1.28 0.64 4 mm3; delay after each segment ¼ 10 s. The saturation efficiency was calculated as Saturation efficiency ¼
ðSIno sat SIsat Þ ; SIno sat
[2]
where SIno sat and SIsat are signal intensities for the image acquired without and with presaturation, respectively. The in vivo performance of a 40-mm saturation band was measured using the same sequence as for the inversion efficiency assessment but with minimal delay between the end of the presaturation and the beginning of the readout. Quantitative ASL Three subjects (mean age, 33 years) were measured at 3 T and 9.4 T to investigate the comparability of quantitative perfusion measurements at high and ultrahigh field strengths. The ASL protocol consisted of perfusion measurements acquired at seven different TIs (TI ¼ 400– 2800 ms; step size ¼ 400 ms) using a FAIR-prepared sequence with WET in-plane presaturation and GRE-EPI readout. To minimize systematic perfusion errors caused by subject motion or system drift, the order of the TIs was randomized for all experiments. For each TI, 118 volumes (9 slices, thickness ¼ 3 mm, gap ¼ 1 mm) were acquired with a TR of 4 s. A further TR reduction was not possible at 9.4 T due to our SAR restrictions, which are more conservative than the International Electrotechnical Commission regulations for the “normal” operation mode by a factor of 2. Approximately 75% of the total power deposition was caused by the inversion pulse and 8% was caused by the in-plane presaturation. M0 was measured prior to the first inversion at the beginning of each sequence after a relaxation delay of 12 s. Further imaging parameters at 3 T were: TE ¼ 18 ms; flip angle ¼ 90 ; in-plane resolution ¼ 3 mm2; partial Fourier factor ¼ 6/8; GRAPPA ¼ 2. Further imaging parameters at 9.4 T were: TE ¼ 17 ms; nominal flip angle ¼ 80 ; inplane resolution ¼ 2 mm2; partial Fourier factor ¼ 6/8; GRAPPA ¼ 3. The higher in-plane resolution at 9.4 T was necessary to reduce intravoxel dephasing and image blurring caused by strong susceptibility effects at ultrahigh fields. A bipolar gradient (Venc ¼ 40 mm/s) between excitation and the start of the readout served to attenuate signal from fast-flowing intravascular blood (24). The optimized inversion pulse with a design thickness of 50 mm was used for the FAIR control condition, resulting in a nominal gap of 7.5 mm between the imaging region and the edge of the selective inversion region. The thickness of the labeling slab was limited to 200 mm to achieve the same bolus duration on both systems. For the determination of T1, an inversion recovery dataset was acquired using the same sequence as for the inversion efficiency determination but with the imaging parameters of the ASL measurement. After reconstruction, motion correction was performed for the tag and control images separately and the T1 of
tissue was obtained from a fit of the inversion recovery data to equation [1]. Each M0 scan and the T1 map were realigned to the M0 scan acquired in the first measurement, and the obtained transformation matrices were also applied to the other frames of the ASL measurement. Both motion correction and realignment were performed using adapted algorithms of SPM8 (University College London, London, UK). Thereafter, PWIs were generated by calculating the difference of the mean control from the mean label images. The two outermost slices were clipped by the motion correction algorithm in most measurements and were therefore excluded from further analysis. The remaining slices were fitted voxel-wise to the general kinetic model for pulsed ASL (12) with the fit variables blood flow (f), transit delay (Dt), and bolus duration (t). For both field strengths, an inversion efficiency of 0.95 and a blood/brain coefficient of 0.95 mL/g (25) was assumed. The longitudinal relaxation time of arterial blood (T1 blood) was assumed to be 1.6 s at 3 T (26) and 2.2 s at 9.4 T. The 9.4 T value was determined by assuming a cubic root increase of T1 with field strength (27). T1 of tissue was extracted from the previously calculated T1 maps. The obtained blood flow and transit delay were averaged for the T1 values of gray and white matter (T1 GM ¼ 1.2–1.7 s at 3 T and 1.9–2.4 s at 9.4 T; T1 WM ¼ 0.8– 1.0 s at 3 T and 1.2–1.7 s at 9.4 T). Voxels that showed a fitted perfusion of more than 150 mL/100 g/min were assumed to contain uncrushed signal from vessels and were excluded from the masks. Temporal and image SNR of the perfusion-weighted data was calculated for the combined gray and white matter regions of the PWIs obtained at a TI of 1.6 s using the methods described by Feinberg et al. (28) and was averaged over the subjects for each field strength. fASL Three subjects were measured with FAIR-ASL and a standard BOLD-weighted GRE-EPI sequence. Duration, spatial coverage, image resolution, and number of measurements were identical for the two sequences (number of slices ¼ 9; number of volumes ¼ 112; isotropic resolution ¼ 2 mm; slice gap ¼ 0.3 mm; GRAPPA ¼ 3; partial Fourier factor ¼ 6/8; nominal flip angle ¼ 100 ). The minimum achievable echo time (TE ¼ 11 ms) and no flow crusher were used for the ASL sequence. Signal preparation was performed as in the quantitative experiment (TI ¼ 1700 ms). The BOLD-weighted EPI protocol had the same image acquisition settings but a TE of 21 ms in order to achieve a stronger T2 weighting. During the experiments, the subjects were asked to perform seven repetitions of a bilateral finger-tapping task (32/32 s of resting/tapping). A high-resolution (0.8 mm isotropic) GRE measurement was performed for coregistration. No motion correction was performed for the ASL data since the algorithm detected artificial motion in the label images in two of the three subjects evoked by the strong paradigm related signal change. The corresponding control measurements, however, showed a maximum displacement of smaller than half a voxel during the sequence in all three subjects. Functional analysis was
4
Bause et al.
performed with the FMRIB Expert Analysis Tool (FEAT) and coregistered to the anatomical reference with the FMRIB Linear Image Registration Tool (FLIRT) of the FSL package (29). High-Resolution ASL To assess the potential of high-resolution ASL at 9.4 T, a FAIR prepared GRE-EPI (TI ¼ 2 s) with a spatial resolution of 1.05 1.05 2 mm3 was acquired in two subjects. For each subject, 162 volumes consisting of seven slices (gap ¼ 2 mm) were measured in 11 min. Further imaging parameters were: TE ¼ 21 ms; nominal flip angle ¼ 90 ; GRAPPA ¼ 4; partial Fourier factor ¼ 6/8; no flow crusher. For comparison, the same measurement was performed at 3 T. Due to the lower T2 -related signal losses and reduced distortions at 3 T, we were able to decrease the parallel imaging factor to GRAPPA ¼ 3, which increased the TE to 23 ms. Furthermore, it would have been possible to shorten the TR at 3 T due to lower SAR. For this reason, a TR of 3 s was assumed for the 3 T measurements, resulting in a scan time decrease from 11 min to 8 min. In order to allow a fair comparison between the two field strengths, the number of averages that were used to calculate the 9.4 T PWIs was reduced accordingly (from 80 to 60). As an anatomical reference, GRE images with a resolution of 0.6 0.6 1 mm3 were acquired from the same volume at 9.4 T.
FIG. 1. a: Effect of B0 off-resonances on the inversion slab profile for two different transmit field strengths. The deformation of the profile is stronger for lower Bþ 1 levels. b: Simulated influence of transverse relaxation during the pulse. A decrease in inversion efficiency and selectivity can be observed, which becomes even more prominent at higher transmit field strengths.
saturation efficiency can also be observed outside of the saturation slab and close to the edge of the phantom due to strong B0 inhomogeneities. The Bþ 1 field measured in vivo at four sagittal positions is depicted in Figure 3a. Because an inversion pulse voltage of approximately 500 V was used for all for in vivo experiments, complete inversion is expected in large parts of the brain except for the inferior part of the cerebellum (white arrow). This is in agreement with the fitted efficiency of the nonselective inversion pulse (Fig. 3b, top row). The bottom row of Figure 3b shows the efficiency profile of the selective pulse. In these images, no off-resonance–related distortions of the slab profile can be observed. Furthermore, all inversion efficiency maps obtained from the in vivo experiment show again a T2like contrast. Even though the transmit field distribution in the brain was less homogenous than in the phantom, a high degree of signal suppression was achieved with the optimized presaturation (Fig. 3c).
RESULTS Inversion Pulse and In-Plane Presaturation Optimization The design parameters of the optimized TR-FOCI pulse are listed in Table 1. According to the simulation, its theoretical inversion efficiency was a > 0.99 in 80% of the nominal slice thickness for a transmit field strength greater than 5.4 mT when the effects of B0 and relaxation were neglected. Figure 1a shows the simulated influence of static magnetic field inhomogeneities on the slab profile. A degradation of the inversion profile is visible for the simulation with Bþ 1 ¼ 5.4 mT, but not for the higher transmit field strength of 20 mT. In contrast, the effect of T2 on the profile declined for the lower Bþ 1 level (Fig. 1b). The fitted efficiency in the phantom confirmed the robustness of the pulse against Bþ 1 inhomogeneities but also the strong effect of transverse relaxation on the slab profile (Fig. 2b). The measured transition band (0.09 < a < 0.9) had a thickness of 14 mm, which reduced the actual inversion slab thickness (a > 0.9) to 36 mm (design thickness 50 mm). Figure 2c shows the profile of the 50-mm-thick presaturation slab in the phantom using the optimized flip angles of 180 , 89.5 , 69.9 , and 116.1 . A high degree of signal suppression was achieved for the entire saturation band despite the inhomogeneous transmit field. However, an increase of
Quantitative ASL PWIs at different TIs (five out of nine measured slices) are shown in Figure 4 (top). Although a flow crusher of 40 mm/s was used in this experiment, high signal can still be observed in some areas of the images acquired at short TIs. This indicates that not all signal of fast flowing
Table 1 Parameters Defining the Shape of the Optimized TR-FOCI Pulse Amax
w
r1
r2
r3
r4
r5
m
b
t1
t2
31.52
0.28
0.49
0.13
0.32
0.83
1.00
0.80
5.34
0.46
0.32
For a detailed description of the individual parameters, see Hurley et al. (19).
Quantitative and Functional Pulsed ASL at 9.4 T
5
23% lower in gray matter and 17% lower in white matter, respectively. The fitted transit delay of the same subject is shown in the bottom row of Figure 4. For both field strengths, the highest transit delay was observed in the occipital lobe and in deep white matter. However, unrealistically low transit delay values were obtained for the most inferior and superior slices, probably due to remaining static tissue signal which was higher in the data obtained at 9.4 T than in the 3 T measurements. Figure 5 shows the increase in mean gray matter transit delay for the different slices. The bolus durations at 3 T were 1.27 s for gray matter and 1.73 s for white matter. The values obtained at 9.4 T were slightly higher for gray matter (1.44 s) but lower for white matter (1.52 s). The low value in white matter is due in part to the poor fit in the most inferior slices, as shown by the fact that the bolus duration increased to 1.65 s if the most inferior and most superior evaluated slices were neglected. The SNR of the PWIs at a TI of 1.6 s of all three subjects in gray and white matter (slices 2–8) was relatively similar (5.26 6 0.5 at 3 T; 4.47 6 0.11 at 9.4 T), despite the large difference in voxel volume (factor of 2.3). For the temporal SNR, the obtained values were almost identical (0.69 6 0.05 at 3 T; 0.63 6 0.05 at 9.4 T). fASL Figure 6a shows the mean PWI of an exemplary slice of two subjects during the rest and stimulus periods. A strong increase in perfusion signal during the fingertapping task was seen in the motor cortex of both hemispheres and the supplementary motor area. The FIG. 2. a: Measured transmit-field distribution in the phantom. b: Profile along the gray line (left) of the fitted efficiency confirms the predicted loss in selectivity and efficiency caused by relaxation processes during the pulse (right). The blue box indicates the transition band where 0.09 < a < 0.9. c: The optimized in-plane presaturation scheme has a good selectivity and shows a high degree of signal suppression despite the strong transmit field inhomogeneities (left). Some signal suppression can also be observed toward the edges of the phantom (right) due to stronger B0 variations.
spins was spoiled by the bipolar gradient, which may be due to different vessel orientations. The higher in-plane resolution of the 9.4 T measurements performed leads to a better perfusion contrast due to reduced partial voluming when compared with the 3 T measurements. On the other hand, the signal change over TI is more obvious in the 3 T data due to the shorter T1 of tissue at the lower field strength. In addition, an increase in signal arising from draining veins at TIs 1.6 s can be observed in the images acquired at the lower field strength. The average gray matter perfusion in all three subjects and all evaluated seven slices was 55.8 6 4.9 mL/ 100 g/min for the measurements performed at 3 T and 42.8 6 3.3 mL/100 g/min for the 9.4 T datasets. For white matter perfusion, the same trend was observed, with 29.7 6 3.4 mL/100 g/min at 3 T and 24.8 6 0.6 mL/ 100 g/min at 9.4 T. Thus, the mean fitted perfusion values obtained at the higher field strength were approximately
FIG. 3. a: Measured transmit field (four sagittal slices of the 3D volume are shown) at 9.4 T. The white arrow indicates the area with very low transmit field strength in the cerebellum. b: Fitted efficiency of the selective and nonselective inversion in the brain. c: Measured slab profile of the WET saturation with optimized flip angles.
6
Bause et al.
FIG. 4. PWIs acquired at multiple TIs at 9.4 T (top left) and 3 T (top right) of one subject. The resolution of the images is 2 2 3 mm3 and 3 3 3 mm3 for the 9.4 T and 3 T data, respectively. Only five out of nine acquired slices are shown. The middle and bottom rows show perfusion maps and transit delay distribution obtained from the fit of the multiple TI data to the general kinetic model.
activation maps are overlaid on the corresponding anatomical reference in Figure 6b, together with the corresponding z-score maps of the reference BOLD dataset. Figure 6c shows the time course of the median perfusion (orange) and BOLD signal (blue) in the 300 voxels with the highest z-score in each of the datasets for subject 1. A reduced temporal signal stability and poststimulus undershoots can be observed in the time course of the ASL dataset. High-Resolution ASL Figure 7 depicts five of seven slices of the anatomical reference images as well as the high-resolution PWIs of
both subjects obtained by averaging all acquired volumes. A great amount of detail in the cortical structure is visible in the ASL measurements due to the high resolution and the consequently reduced partial volume effects. A clear difference between gray matter sulci and the intermediate surrounding CSF was observed, especially in the occipital lobe. However, a higher noise level was visible in the medial part of the PWIs, which is probably due to the high GRAPPA acceleration factor and the declining perfusion signal at the long TI. An example slice of the PWIs acquired at 3 T is shown in Figure 8 (left). The image obtained at the higher field strength shows a better contrast and more details (Fig. 8, right) even though it was calculated with a reduced
Quantitative and Functional Pulsed ASL at 9.4 T
FIG. 5. Transit delay of the bolus in gray matter (mean and standard deviation over all subjects) at 3 T (gray) and 9.4 T (black).
number of averages to take the possible reduction in repetition time at 3 T into account. DISCUSSION The in vivo and phantom evaluation of the developed inversion pulse and the presaturation scheme confirmed their theoretical stability against transmit field variations. However, the measurements also showed that the efficiency and selectivity of the inversion pulse was heavily affected by transverse relaxation. Although the influence of T2 can be diminished by reducing the pulse power, this in turn increases the impact of offresonances. The slab profile can then only be improved significantly at the cost of higher SAR. Often, longer RF
FIG. 6. a: Mean perfusion signal (TI ¼ 1700 ms) in an exemplary slice during rest and activation measured in two of the three subjects. The perfusion change during the finger-tapping paradigm is visible in the motor cortex of both subjects. b: Overlay of the fASL and BOLD activation maps on the anatomical reference. Note the different scaling of the color bars. c: Median signal time course of the 300 voxels that showed the highest z-score in the functional ASL (orange) and the BOLD measurement (blue) of subject 1. The gray boxes symbolize the stimulation periods.
7
pulses are used to reduce the required pulse voltage and thus SAR; however, this can increase the susceptibility to T2 relaxation or static field inhomogeneities. At worst, the FAIR condition, which requires that the selective and nonselective pulse have the same inversion efficiency in the image region, is violated (14). In the quantitative experiment, high perfusion signal was observed in regions with draining veins at TIs 1.6 s caused by the rather thin control slab, which also led to a labeling of blood superior to the imaging region. Although this effect occurred at both field strengths, it was more pronounced in the images acquired at 3 T (Fig. 4, top) which can be explained by the very short T2 of venous blood at 9.4 T (30). For this reason, a perfusion signal in the sagittal sinus is also visible in the 9.4 T PWIs of the functional experiment, which were acquired with a shorter TE. There may be multiple reasons for perfusion differences that were observed between the quantitative measurements at 3 T and 9.4 T. First, a single compartment model was used for fitting the multiple TI data. This model does not consider T2 relaxation during the GREEPI readout, which can result in an underestimation of perfusion, especially at high field strengths (31). Second, the higher spatial resolution of the 9.4 T measurements reduces partial volume effects in the T1 maps and the perfusion images. Almost identical gray matter transit delays were found in slices 3–7 for the two field strengths (Fig. 5). Typically, a nominal distance of 1– 2 cm is used between the labeling and imaging slab to
8
Bause et al.
FIG. 7. Mean difference signal (voxel size ¼ 1.05 1.05 2 mm3; TI ¼ 2000 ms) measured at 9.4 T in two different subjects in
