NOTE Magnetic Resonance in Medicine 74:1698–1704 (2015)
Efficient Generation of T*2-Weighted Contrast by Interslice Echo-Shifting for Human Functional and Anatomical Imaging at 9.4 Tesla Philipp Ehses,1,2* Jonas Bause,2 G. Shajan,2 and Klaus Scheffler1,2 Purpose: Standard gradient-echo sequences are often prohibitively slow for T2-weighted imaging as long echo times prolong the repetition time of the sequence. Echo-shifting offers a way out of this dilemma by allowing an echo time that exceeds the repetition time. The purpose of this work is to present a gradient-echo sequence that is optimized for multislice T2 -weighted imaging applications by combining echoshifting with an interleaved slice excitation order. Theory and Methods: This combined approach offers two major advantages: First, it combines the advantages of both concepts, that is, echo time and pulse repetition time can be significantly increased without affecting scan time. Second, there is no echo-shifting related signal loss associated with this concept as only a single radiofrequency pulse is applied per pulse repetition time and slice. Results: A 9.4 Tesla high-resolution T2 -weighted anatomical brain scan of the proposed sequence is compared to a standard gradient-echo. Furthermore, results from 9.4 Tesla blood oxygen level dependent functional magnetic resonance imaging experiments with an in-plane resolution of 0.8 0.8 mm2 are presented. Conclusion: The proposed sequence allows for efficient generation of T2 -weighted contrast by combining echo-shifting with an interleaved slice excitation order. Magn Reson Med C 2014 Wiley Periodicals, Inc. 74:1698–1704, 2015. V Key words: echo-shifting; T*2 ; ultrahigh field; susceptibilityweighted images; fMRI
INTRODUCTION Long echo-time gradient-echo sequences with strong T2 weighting and high phase contrast have found widespread use in MRI. Examples for its use are blood oxygen level dependent functional magnetic resonance imaging (BOLD fMRI) (1), susceptibility weighted imaging (2), 1
€bingen, Department of Biomedical Magnetic Resonance, University of Tu €bingen, Germany. Tu 2 High-Field Magnetic Resonance Center, Max Planck Institute for Biological €bingen, Germany Cybernetics, Tu Grant sponsor: Helmholtz Alliance ICEMED - Imaging and Curing Environmental Metabolic Diseases, through the Initiative and Networking Fund of the Helmholtz Association. *Correspondence to: Philipp Ehses, High-Field Magnetic Resonance Center, Max Planck Institute for Biological Cybernetics, Spemannstr. 41, 72076 €bingen, Germany. E-mail: [email protected] Tu Part of this work was presented at the Annual Meeting of ISMRM in Melbourne, Australia, 2012. Received 13 August 2014; revised 21 October 2014; accepted 2 November 2014 DOI 10.1002/mrm.25570 Published online 2 December 2014 in Wiley Online Library (wileyonlinelibrary. com). C 2014 Wiley Periodicals, Inc. V
quantitative susceptibility mapping (3), and MRI thermometry (4). Single-shot echo-planar imaging (EPI) sequences offer a convenient way to generate T2 -weighted contrast and are routinely used in BOLD fMRI. However, static and dynamic magnetic field inhomogeneities can lead to strong distortion artifacts in EPI, resulting in a reduction in image quality. This problem becomes even more severe with increasing magnetic field and at higher resolutions. Conversely, conventional single-echo gradient-echo sequences, that do not suffer from major distortion artifacts, are often prohibitively slow for T2 -weighted imaging applications as long echo times prolong the repetition time of the sequence. Echo-shifted gradient-echo (ES-GRE) sequences (5–9) offer a way out of this dilemma by allowing an echo time that exceeds the repetition time. In multislice acquisitions, a combination of echo-shifting and an interleaved slice acquisition order allows to prolong echo time (TE) and repetition time (TR) without affecting scan time, as previously reported for a segmented EPI sequence (10). The prolonged TE helps to generate T2 -weighted contrast while the increased TR leads to higher signal-to-noise ratio (SNR) by reducing signal saturation. The aim of this work was to develop a multislice echo-shifted gradient-echo sequence with interleaved slice order, optimized for T2 -weighted imaging applications at ultrahigh field. 9.4 Tesla results from a highresolution T2 -weighted anatomical scan of the sequence are compared to that of a standard gradient-echo sequence. Furthermore, results from submillimeterresolution BOLD fMRI experiments obtained at 9.4 Tesla are presented.
THEORY Echo-Shifting Echo-shifting allows for an echo time that exceeds the repetition time of the sequence by delaying echo collection by one or more TR intervals (5). The excited magnetization is effectively stored on the transverse axis until spoiler gradients recall it in a later TR. For an echo shift of one TR interval and spoiling moment A, this can be done by applying gradient lobes of areas A and þ2A before and after each readout, respectively (Fig. 1, left). Another strategy is to apply a single gradient spoiler before each readout that alternates between þA and A with every TR interval (Fig. 1, right). This makes the spoiler gradients multi TR-periodic with a periodicity of two TR intervals (or n11 TR intervals for n echo-shifts).
1698
Interslice Echo-Shifting
1699
FIG. 1. Simplified sequence diagram for interslice ES-GRE with two slices and one echo shift. Left: TR-periodic spoiling. Right: multi TR-periodic spoiling. Gradients on the phase-encoding axis are fully balanced (not shown).
The necessary gradient moments for TR-periodic and multi-TR-periodic spoiling are compared in Table 1. At higher echo-shifting factors n, multi-TR-periodic spoiling becomes increasingly more efficient in terms of gradient switching: For n > 1, the maximum spoiler moment that is applied directly after the radiofrequency (RF) pulse is the same for both methods and is proportional to n. However, multi TR-periodic spoiling does not require a spoiler after the readout event, whereas TR-periodic spoiling requires one with an area proportional to n 1 1. Although total gradient switching (parenthesed values in Table 1) is for one echo-shift relatively similar for both methods, multi-TR-periodic spoiling still holds a small advantage as the larger part of the multi-TR-periodic spoiler gradient moment is required between the RF pulse and the readout window. This makes achieving longer echo times easier and more efficient. However, multi-TR-periodicity is not without drawbacks: Strong variations in uncompensated eddy currents between neighboring lines when using a na€ıve linear reordering strategy will in most cases result in strong ghosting artifacts. Switching to centric reordering (or a
similar suitable reordering scheme) can help to reduce the impact of these artifacts by distributing them over the full FoV, making them less pronounced and less visible. However, even in best case, the resulting artifacts will still appear as a slight noise enhancement. One general drawback of echo-shifting irrespective of the spoiling strategy is that it leads to a signal reduction as signal components are lost in higher-order pathways due to the application of multiple RF pulses per TR and slice. For an echo-shifting factor of n and flip angle a, the associated signal loss factor is given by cos 2n a2 (6). Interslice Echo-Shifting Combining a 2D interleaved multislice acquisition with echo-shifting has been previously proposed for a segmented EPI sequence with TR-periodic spoiling (10). This combined approach offers two major advantages: First, it combines the advantages of both concepts, that is, TE and TR can be significantly increased without affecting the acquisition time per volume (sometimes misleadingly called volume TR). Second, the fact that
Table 1 Spoiler Gradient Moments for GRE, as well as for ES-GRE with TR-Periodic and Multi-TR-Periodic Spoiling in Read Direction (Dephasing of 2p, Equivalent to One k-space)
Sequence type GRE TR-periodic ES-GRE (7,8) Multi-TR-periodic ES-GRE (5,6)
Echo shifts – 1 2 n 1 2 n
Read
Slice
Periodic use in TR k mod (n11)
pre-ADC
post-ADC
pre-ADC
post-ADC
– – – – 0 1 0 1...2 0 1...n
0 (0.5) 1 (0.5) 2 (1.5) n (n 0.5) 1 (0.5) 1 (1.5) 2 (1.5) 1 (1.5) n (n 0.5) 1 (1.5)
1 (0.5) 2 (1.5) 3 (2.5) n þ 1 (n þ 0.5) 0 (0.5)
0 (0.5) 0 (0.5) 0 (0.5) 0 (0.5) 0.5 (1)
– 0 (0.5) 0 (0.5) 0 (0.5) 1 (0.5)
0 (0.5)
0.5 (1)
0.75 (0.25)
0 (0.5)
0.5 (1)
nþ1 1 2n (2n)
Gradient moments are given relative to the moment of the readout and slice gradient, respectively. The numbers in parentheses are with dephasing from the readout and slice gradient accounted for (ignoring gradient ramps and assuming a symmetric readout and RF pulse). The total amount of gradient switching in the imaging sequence is better represented by these parenthesed values. As can be seen from the table, total gradient switching is much lower for multi-TR periodic spoiling than for TR-periodic spoiling, especially for high echo-shifting factors.
1700
only a single RF pulse is applied per TR and slice means that there is no echo-shifting related signal loss associated with this concept. As the delayed echo acquisition is interleaved with the slice acquisition, we propose to call this concept interslice echo-shifting (IS-ES) to better distinguish it from conventional echo-shifting. Simplified sequence diagrams for IS-ES with TR-periodic and multi TR-periodic spoiling are shown in Figure 1. Another advantage of interslice echo-shifting is that it allows the use of multi TR-periodic spoiler gradients without risking ghosting artifacts due to uncompensated eddy currents: if the number of slices is an integer multiple of n þ 1 (the periodicity of multi TR-periodic spoiling), all k-space lines of each respective slice are acquired using the same spoiler gradients. As a result, multi TR-periodic spoiling is in fact TR-periodic again, so that ghosting artifacts due to non-TR-periodic eddy currents can be avoided independent of the line reordering strategy. This removes the main disadvantage of multi-TR-periodic echo-shifting, making it the preferable spoiling strategy for interslice echo-shifting. Thus, we chose to use multi-TR periodic spoiling for all IS-ES experiments presented in this work. Note that for simplification, we will continue to refer to the two spoiling strategies by “TR-periodic” and “multi-TRperiodic,” although both strategies are in fact TR-periodic for each slice when used with interslice echo-shifting.
Ehses et al.
each lasting for the acquisition duration of five volumes (18.9 s). In total, 10 periods of each condition (100 time frames) were acquired in 6:20 min. A total of 12 interleaved slices were measured using the interslice echoshifted sequence with an echo-shifting factor of two. Other sequence parameters were as follows: nominal flip angle ¼ 25 , TR ¼ 84 ms (7 ms per slice), TE ¼ 18 ms, readout bandwidth ¼ 390 Hz/pixel, voxel size ¼ 0.8 0.8 1.0 mm3 (þ20% interslice gap), parallel imaging acceleration factor ¼ 3 (with 48 autocalibration lines acquired once at the beginning of the experiment), 6/8 partial Fourier, acquisition time per volume ¼ 3.78 s. Images were reconstructed offline from the raw data as described below. After motion correction using AFNI (12), functional analysis was performed with FEAT from the FSL package (13), using standard parameter settings and no spatial smoothing. The FEAT analysis is based on a model that is created from a convolution of the paradigm convolved with a standard hemodynamic response function. Activation maps are then obtained by fitting the acquired time series to the model. For temporal SNR comparison, one of the six volunteers was additionally scanned using a standard GRE sequence. Parameters were the same as in the IS-ES scan except that this sequence only allowed the acquisition of four slices in the given TR. Image Reconstruction
METHODS
To illustrate contrast and SNR similarities between the interslice echo-shifted sequence and a conventional gradient echo, high-resolution multislice data were acquired from one of the six volunteers. For better comparability, all contrast and SNR relevant sequence parameters were kept the same in both sequences: nominal flip angle ¼ 31 , TR ¼ 712.8 ms, TE ¼ 16 ms, readout bandwidth ¼ 290 Hz/pixel, voxel size ¼ 0.4 0.4 1.0 mm3 (þ20% interslice gap), acquisition time ¼ 5 min. One echo-shift was sufficient for achieving the chosen echo time without requiring any idle time in the sequence. The choice of a common TR and TE in both sequences allowed the acquisition of twice as many slices in the IS-ES scan (36 and 72 slices were acquired in the gradient-echo and IS-ES experiment, respectively).
Image reconstruction was performed in Matlab (The Mathworks). As a first step, the multichannel raw data was prewhitened (14) using noise data that was acquired prior to each scan (4096 noise samples). K-space lines that were skipped in the fMRI experiment due to parallel imaging and partial Fourier were reconstructed using GRAPPA (15) and a projection onto convex sets algorithm (16), respectively. The highresolution T2 -weighted data were fully sampled, so that a simple 2D fast Fourier transform was sufficient to convert the signal to the image domain. Coil data were then combined by estimating combination weights and coil sensitivity maps using the adaptive combine method (17). Coil data from the BOLD experiment were combined using Walsh’s original method, whereas the structural data were combined using a recently proposed modification (18), that improves phase reconstruction by enforcing its smoothness in a region around each voxel (a 11 11 square kernel was used in this work). The result is very similar to a homodyne filtered phase image. Signal level nonuniformities due to receiver coil sensitivity variations were then intensity normalized using the estimated coil sensitivity maps according to an approximation proposed by Griswold et al. (19). For better comparability, the two highresolution data sets were normalized using the same normalization factors (determined from the IS-ES scan).
BOLD fMRI Experiment
SNR and tSNR Analysis
BOLD fMRI data were acquired while each of the six subjects was performing a simple bilateral finger tapping task. Starting from a resting period, the paradigm alternated between a tasking period and a resting period,
Signal-to-noise ratio was calculated from the highresolution T2 -weighted datasets and the acquired noise data using the pseudo multiple replica method (20). To this end, a total of 100 pseudo replicas were generated
All experiments were performed on a 9.4 Tesla system (Siemens Healthcare, Erlangen, Germany) with a maximum gradient strength and slew rate of 70 mT/m and 200 mT/m/ms, respectively. A custom-built head coil (11) was used for RF transmission and reception (16 transmit/31 receive channels). Data were acquired from six healthy volunteers after obtaining informed consent and approval by the local ethics committee. High-Resolution T2 -Weighted Imaging Experiments
Interslice Echo-Shifting
1701
FIG. 2. Comparison of the high-resolution T2 -weighted data from the conventional gradient-echo scan and the interslice echo-shifted scan at an in-plane resolution of 0.4 0.4 mm2 (from left to right: magnitude image, phase image, and SNR map with histogram). Note that in the same scan time and using the same TR and TE, twice as many slices were covered with the interslice echo-shifted sequence than with the conventional gradient-echo. The SNR histogram only includes brain voxels from the 36 slices that were acquired in both scans.
by adding pseudorandom permutations of the noise data to the image. Furthermore, temporal SNR (tSNR) maps were calculated from the IS-ES and conventional gradient-echo BOLD data of one of the six volunteers. Susceptibility Weighted Imaging Reconstruction Susceptibility weighted images were reconstructed from the high-resolution T2 -weighted data according to Haacke et al. (2). A minimum intensity projection over five slices (6 mm) was then calculated from the susceptibility weighted imaging data. RESULTS Figure 2 shows magnitude, phase and the SNR map for the two high-resolution T2 -weighted datasets with an in-plane resolution of 0.4 0.4 mm2. The higher spatial coverage of the IS-ES data can be appreciated by looking at the sagittal
view. Contrast is very similar in the magnitude and phase images for both datasets (Fig. 2, left and middle). The similarity between the two is also evident in the SNR maps and histograms, shown on the right side of Figure 2. The minimum intensity projection of the susceptibility weighted imaging reconstruction, shown in Figure 3, is again comparable for the two datasets and shows very high detail. Three representative minimum intensity projections are shown, with the first slice missing from the conventional gradient-echo data due to the lower spatial coverage. The most inferior slice of the IS-ES GRE data exhibits reduced signal in some areas due to inhomogeneities of the transmit field. Figure 4 top shows the activation maps from the fMRI experiments overlayed over an example time frame for one exemplary slice for each volunteer. The results indicate that IS-ES GRE is well suited for high-resolution functional imaging at ultrahigh field. Strong activations
1702
Ehses et al.
FIG. 3. Minimum intensity projection of a susceptibility weighted imaging reconstruction of the high-resolution T2 weighted data. The projection thickness is 6 mm (five slices). Three representative slices are shown, with the first slice missing from the conventional scan’s data due to its lower spatial coverage. Contrast is very similar for both datasets (compare the zoomed insets).
were found in the primary motor cortex area of both hemispheres: maximum z-score was 15.7 for volunteer #3. Activations found in the somatosensory area were relatively weak, which may be explained by lower SNR due to a relatively low coil sensitivity and high g-factor loss in this area. The signal time course and the FEAT model for the voxel with maximum z-score is shown on the left below the activation maps. A very strong signal change (640%) can be observed here, explained by the fact that the voxel lies inside a vein. To show a time course that is more representative for gray matter, a median time course was calculated from all voxels with z-scores greater than four. The median signal change was approximately 64%, as shown at the bottom left corner of Figure 4. Temporal SNR (tSNR) maps for IS-ES GRE and a standard GRE BOLD acquisition are shown for one of the central slices of one volunteer in the bottom right corner of Figure 4. Median tSNR in the brain was 13% lower in IS-ES than in standard GRE (tSNRGRE =tSNRIS-ES ¼ 19:5=17:2). DISCUSSION AND CONCLUSION IS-ES GRE allows for efficient generation of T2 -weighted contrast by combining echo-shifting with an interleaved slice excitation order. To our knowledge, the combination of echo-shifting with an interleaved slice acquisition has first been proposed in 2006 for a segmented EPI sequence by Gibson et al. (10). However, the authors did not discuss one of the main advantages that this new concept holds over conventional echo-shifting, namely that there is no echo-shifting related signal loss due to higher coherence pathways as only a single RF pulse is applied per TE in each slice. Furthermore, our work
shows that multi-TR-periodic spoiling is preferable to the TR-periodic spoiling strategy used by Gibson et al. It is difficult to overstate the difference between the concept of echo-shifting and interslice echo-shifting: from the definition of echo-shifting follows that an interslice echo-shifted sequence is technically not echoshifted at all as the combination with an interleaved slice excitation order results in a TR that is always larger than TE. IS-ES GRE merely uses the same gradient spoiling technique as conventional echo-shifting to shift the echo from one sub-TR unit to another (as illustrated in Fig. 1). This insight illustrates why IS-ES does not suffer from any echo-shifting related signal loss. Furthermore, it explains why the multi TR-periodic spoiling strategy of echo-shifted sequences (5,6) is in fact TR-periodic for interslice echo-shifting. This makes multi-TR-periodic spoiling the preferable spoiling strategy for IS-ES GRE by reducing the required spoiler gradient moments, especially for higher echo-shifting factors (compare Table 1). The results from the high-resolution T2 -weighted data show that one echo-shift allows covering twice as many slices in the same acquisition time as with a standard GRE sequence while achieving similar magnitude and phase contrast as well as SNR (Fig. 2). Note that depending on the acquisition parameters and the imaged body region, IS-ES GRE may show a slightly different contrast than standard GRE due to stronger gradient switching and a higher number of RF pulses per unit time in IS-ES GRE: the former makes IS-ES GRE more sensitive to flow and diffusion while the latter leads to a stronger magnetization transfer induced signal attenuation, especially in white matter. Flow sensitivity and signal loss due to diffusion can be estimated from the
Interslice Echo-Shifting
1703
FIG. 4. Results from the finger tapping experiments with an in-plane resolution of 0.8 0.8 mm2: Top: Overlay of BOLD activation maps with an example time frame for a representative slice from each of the six volunteers. Bottom left: Time series (black) and fitted regressor (red) for volunteer #3 for the voxel with maximum z-score (¼ 15.7; voxel position indicated by white arrow), and for the median over all voxels with z-scores greater than 4 (yielding 5428 voxels). Bottom right: tSNR maps for IS-ES and standard GRE as calculated from the fMRI time-series for a central slice of one of the volunteers (#6). Median tSNR in the brain was 17.2 and 19.5 in the IS-ES and standard GRE, respectively (note that three times as many slices as in GRE were acquired in the IS-ES scan).
gradient table of the sequence: maximum b-values in the BOLD fMRI scan were 4:8 s=mm2 and 62:6 s=mm2 for the standard and IS-ES GRE, respectively. As only a single echo-shift was performed in the high-resolution IS-ES scan, the corresponding b-value was with 38:0s=mm2 (standard GRE: 4:6s=mm2 ) lower than in the BOLD experiment. Assuming a mean apparent diffusion coefficient of 0:8 103 mm2 =s in gray matter (21), this translates to a relatively modest expected signal loss in gray matter of 2.6% in the high-resolution scan and 4.5% in the fMRI scan (relative to standard GRE). Flow sensitivity in the echo-shifted scans, as measured by total first gradient moment m1, was 6–9 times higher than in standard GRE
(high-resolution experiments: m1;GRE =m1;IS-ES ¼ 119=708; ms2 ; BOLD: m mT ms2 ). ½mT m 1;GRE =m1;IS-ES ¼ 87=809; ½ m Magnetization transfer effects are expected to be small, as relatively low flip angles were used due to specific absorption rate (SAR) constraints. One of the main advantages that the proposed IS-ES GRE sequence holds over a single-shot EPI sequence is that it is relatively insensitive to B0 field inhomogeneities that are a common problem for EPI at high field strengths. Images from the IS-ES BOLD experiment and its activation maps are virtually free from distortions (see Fig. 4). Apart from this advantage which is particularly
1704
important at higher field, fMRI with IS-ES GRE also becomes more feasible with increasing field: As the echo time for optimal BOLD contrast decreases, less echoshifts are required to optimize IS-ES GRE for BOLD imaging. This reduces the amount of necessary gradient switching, resulting in less acoustic noise, a lower chance of peripheral nerve stimulation, and less diffusion related signal loss. Another advantage of the proposed sequence is that it reduces T2 -related blurring due to a shorter readout, meaning that the nominal resolution is closer to the actual resolution than in EPI. In contrast, the high nominal resolution that is achievable with EPI at ultrahigh-field is often partially offset by these blurring effects. However, these advantages come at the cost of a longer scan time (3.78 s for 12 slices compared to approximately 2 s for a typical submillimeter EPI with 20 or more slices) and a lower SNR efficiency compared to EPI. Furthermore, single-shot EPI is less sensitive to flow and other dynamic effects than IS-ES GRE. Magnetic field fluctuations, often caused by patient motion and breathing, are another problem for multishot sequences as they can cause signal fluctuations and image ghosting. Phase stabilization (22) (sometimes called frequency locking) can be used to alleviate this problem. This technique relies on the acquisition of an extra nonphase-encoded gradient-echo in each TR from which the field fluctuation can be determined. The phase of each acquired k-space line can then be adjusted to correct for the fluctuating field. While no phase stabilization was used in this work, it may be a useful tool to suppress signal fluctuations and to increase temporal SNR. However, as IS-ES GRE does not require idle time to achieve long echo times, adding an extra readout will somewhat reduce the time-efficiency of the sequence. Although image acquisition in the presented fMRI experiment was already accelerated using a GRAPPA acceleration factor of 3 and 6/8 partial Fourier, we believe that further speed improvements are possible. One promising approach would be to excite and acquire multiple slices simultaneously using the multislice CAIPIRINHA technique (23), which has also recently found application in EPI (24). Depending on coil geometry, CAIPIRINHA potentially allows higher parallel imaging acceleration factors by exploiting coil-sensitivity variations in both phaseencoding and slice-selection direction. Although multiple excitation volumes are absolutely necessary for IS-ES, a 3D acquisition in combination with 2D CAIPIRINHA (25) would also be possible by interleaving multiple 3D slabs. Finally, the efficiency of k-space sampling can be improved by moving to a segmented EPI acquisition as in Gibson et al. (10) or to a non-Cartesian trajectory such as spiral. In addition to the presented application of IS-ES GRE for high-resolution anatomical imaging and BOLD fMRI, this technique may also be useful for other applications were long echo times are essential such as quantitative susceptibility mapping (3) and proton resonance frequency shift thermometry (4). REFERENCES 1. Ogawa S, Menon RS, Tank DW, Kim SG, Merkle H, Ellermann JM, Ugurbil K. Functional brain mapping by blood oxygenation level-
Ehses et al. dependent contrast magnetic resonance imaging. A comparison of signal characteristics with a biophysical model. Biophys J 1993;64:803–812. 2. Haacke EM, Xu Y, Cheng YCN, Reichenbach JR. Susceptibility weighted imaging (SWI). Magn Reson Med 2004;52:612–618. 3. De Rochefort L, Brown R, Prince MR, Wang Y. Quantitative MR susceptibility mapping using piece-wise constant regularized inversion of the magnetic field. Magn Reson Med 2008;60:1003–1009. 4. Ishihara Y, Calderon A, Watanabe H, Okamoto K, Suzuki Y, Kuroda K. A precise and fast temperature mapping using water proton chemical shift. Magn Reson Med 1995;34:814–823. 5. Moonen CTW, Liu G, van Gelderen P, Sobering G. A fast gradientrecalled MRI technique with increased sensitivity to dynamic susceptibility effects. Magn Reson Med 1992;26:184–189. 6. Liu G, Sobering G, Olson AW, van Gelderen P, Moonen CT. Fast echo-shifted gradient-recalled MRI: combining a short repetition time with variable T2 weighting. Magn Reson Med 1993;30:68–75. 7. Moonen CT, Barrios Fa, Zigun JR, Gillen J, Liu G, Sobering G, Sexton R, Woo J, Frank J, Weinberger DR. Functional brain MR imaging based on bolus tracking with a fast T2 -sensitized gradient-echo method. Magn Reson Imaging 1994;12:379–385. 8. de Zwart JA, van Gelderen P, Kelly DJ, Moonen CT. Fast magneticresonance temperature imaging. J Magn Reson B 1996;112:86–90. 9. van Gelderen P, Duyn JH, Ramsey NF, Liu G, Moonen CTW. The PRESTO technique for fMRI. NeuroImage 2012;62:676–681. 10. Gibson A, Peters AM, Bowtell R. Echo-shifted multislice EPI for high-speed fMRI. Magn Reson Imaging 2006;24:433–442. 11. Shajan G, Kozlov M, Hoffmann J, Turner R, Scheffler K, Pohmann R. A 16-channel dual-row transmit array in combination with a 31element receive array for human brain imaging at 9.4 T. Magn Reson Med 2014;71:870–879. 12. Cox RW. AFNI: software for analysis and visualization of functional magnetic resonance neuroimages. Comput Biomed Res 1996;29:162–173. 13. Smith SM, Jenkinson M, Woolrich MW, et al. Advances in functional and structural MR image analysis and implementation as FSL. NeuroImage 2004;23(Suppl 1):S208–S219. 14. Pruessmann KP, Weiger M, B€ ornert P, Boesiger P. Advances in sensitivity encoding with arbitrary k-space trajectories. Magn Reson Med 2001;46:638–651. 15. Griswold MA, Jakob PM, Heidemann RM, Nittka M, Jellus V, Wang J, Kiefer B, Haase A. Generalized autocalibrating partially parallel acquisitions (GRAPPA). Magn Reson Med 2002;47:1202–1210. 16. Haacke E, Lindskogj E, Lin W. A fast, iterative, partial-fourier technique capable of local phase recovery. J Magn Reson 1991;92:126–145. 17. Walsh DO, Gmitro aF, Marcellin MW. Adaptive reconstruction of phased array MR imagery. Magn Reson Med 2000;43:682–690. 18. Inati SJ, Hansen MS, Kellmann P. A solution to the phase problem in adaptive coil combination. In Proceedings of the 21st Annual Meeting of the ISMRM, Salt Lake City, Utah, USA, 2013. p 2672. 19. Griswold M, Walsh D, Heidemann R, Haase A, Jakob P. The use of an adaptive reconstruction for array coil sensitivity mapping and intensity normalization. In Proceedings of the 10th Annual Meeting of the ISMRM, Honolulu, Hawaii, USA, 2002. p. 2410. 20. Robson PM, Grant AK, Madhuranthakam AJ, Lattanzi R, Sodickson DK, McKenzie Ca. Comprehensive quantification of signal-to-noise ratio and g-factor for image-based and k-space-based parallel imaging reconstructions. Magn Reson Med 2008;60:895–907. 21. Le Bihan D, Mangin JF, Poupon C, Clark Ca, Pappata S, Molko N, Chabriat H. Diffusion tensor imaging: concepts and applications. Journal of magnetic resonance imaging: JMRI 2001;13:534–546. 22. Henry PG, van de Moortele PF, Giacomini E, Nauerth A, Bloch G. Field-frequency locked in vivo proton MRS on a whole-body spectrometer. Magn Reson Med 1999;42:636–642. 23. Breuer FA, Blaimer M, Heidemann RM, Mueller MF, Griswold MA, Jakob PM. Controlled aliasing in parallel imaging results in higher acceleration (CAIPIRINHA) for multi-slice imaging. Magn Reson Med 2005;53:684–691. 24. Setsompop K, Gagoski Ba, Polimeni JR, Witzel T, Wedeen VJ, Wald LL. Blipped-controlled aliasing in parallel imaging for simultaneous multislice echo planar imaging with reduced g-factor penalty. Magn Reson Med 2012;67:1210–1224. 25. Breuer FA, Blaimer M, Mueller MF, Seiberlich N, Heidemann RM, Griswold Ma, Jakob PM. Controlled aliasing in volumetric parallel imaging (2D CAIPIRINHA). Magn Reson Med 2006;55:549–556.
Efficient Generation of T*2-Weighted Contrast by Interslice Echo-Shifting for Human Functional and Anatomical Imaging at 9.4 Tesla Philipp Ehses,1,2* Jonas Bause,2 G. Shajan,2 and Klaus Scheffler1,2 Purpose: Standard gradient-echo sequences are often prohibitively slow for T2-weighted imaging as long echo times prolong the repetition time of the sequence. Echo-shifting offers a way out of this dilemma by allowing an echo time that exceeds the repetition time. The purpose of this work is to present a gradient-echo sequence that is optimized for multislice T2 -weighted imaging applications by combining echoshifting with an interleaved slice excitation order. Theory and Methods: This combined approach offers two major advantages: First, it combines the advantages of both concepts, that is, echo time and pulse repetition time can be significantly increased without affecting scan time. Second, there is no echo-shifting related signal loss associated with this concept as only a single radiofrequency pulse is applied per pulse repetition time and slice. Results: A 9.4 Tesla high-resolution T2 -weighted anatomical brain scan of the proposed sequence is compared to a standard gradient-echo. Furthermore, results from 9.4 Tesla blood oxygen level dependent functional magnetic resonance imaging experiments with an in-plane resolution of 0.8 0.8 mm2 are presented. Conclusion: The proposed sequence allows for efficient generation of T2 -weighted contrast by combining echo-shifting with an interleaved slice excitation order. Magn Reson Med C 2014 Wiley Periodicals, Inc. 74:1698–1704, 2015. V Key words: echo-shifting; T*2 ; ultrahigh field; susceptibilityweighted images; fMRI
INTRODUCTION Long echo-time gradient-echo sequences with strong T2 weighting and high phase contrast have found widespread use in MRI. Examples for its use are blood oxygen level dependent functional magnetic resonance imaging (BOLD fMRI) (1), susceptibility weighted imaging (2), 1
€bingen, Department of Biomedical Magnetic Resonance, University of Tu €bingen, Germany. Tu 2 High-Field Magnetic Resonance Center, Max Planck Institute for Biological €bingen, Germany Cybernetics, Tu Grant sponsor: Helmholtz Alliance ICEMED - Imaging and Curing Environmental Metabolic Diseases, through the Initiative and Networking Fund of the Helmholtz Association. *Correspondence to: Philipp Ehses, High-Field Magnetic Resonance Center, Max Planck Institute for Biological Cybernetics, Spemannstr. 41, 72076 €bingen, Germany. E-mail: [email protected] Tu Part of this work was presented at the Annual Meeting of ISMRM in Melbourne, Australia, 2012. Received 13 August 2014; revised 21 October 2014; accepted 2 November 2014 DOI 10.1002/mrm.25570 Published online 2 December 2014 in Wiley Online Library (wileyonlinelibrary. com). C 2014 Wiley Periodicals, Inc. V
quantitative susceptibility mapping (3), and MRI thermometry (4). Single-shot echo-planar imaging (EPI) sequences offer a convenient way to generate T2 -weighted contrast and are routinely used in BOLD fMRI. However, static and dynamic magnetic field inhomogeneities can lead to strong distortion artifacts in EPI, resulting in a reduction in image quality. This problem becomes even more severe with increasing magnetic field and at higher resolutions. Conversely, conventional single-echo gradient-echo sequences, that do not suffer from major distortion artifacts, are often prohibitively slow for T2 -weighted imaging applications as long echo times prolong the repetition time of the sequence. Echo-shifted gradient-echo (ES-GRE) sequences (5–9) offer a way out of this dilemma by allowing an echo time that exceeds the repetition time. In multislice acquisitions, a combination of echo-shifting and an interleaved slice acquisition order allows to prolong echo time (TE) and repetition time (TR) without affecting scan time, as previously reported for a segmented EPI sequence (10). The prolonged TE helps to generate T2 -weighted contrast while the increased TR leads to higher signal-to-noise ratio (SNR) by reducing signal saturation. The aim of this work was to develop a multislice echo-shifted gradient-echo sequence with interleaved slice order, optimized for T2 -weighted imaging applications at ultrahigh field. 9.4 Tesla results from a highresolution T2 -weighted anatomical scan of the sequence are compared to that of a standard gradient-echo sequence. Furthermore, results from submillimeterresolution BOLD fMRI experiments obtained at 9.4 Tesla are presented.
THEORY Echo-Shifting Echo-shifting allows for an echo time that exceeds the repetition time of the sequence by delaying echo collection by one or more TR intervals (5). The excited magnetization is effectively stored on the transverse axis until spoiler gradients recall it in a later TR. For an echo shift of one TR interval and spoiling moment A, this can be done by applying gradient lobes of areas A and þ2A before and after each readout, respectively (Fig. 1, left). Another strategy is to apply a single gradient spoiler before each readout that alternates between þA and A with every TR interval (Fig. 1, right). This makes the spoiler gradients multi TR-periodic with a periodicity of two TR intervals (or n11 TR intervals for n echo-shifts).
1698
Interslice Echo-Shifting
1699
FIG. 1. Simplified sequence diagram for interslice ES-GRE with two slices and one echo shift. Left: TR-periodic spoiling. Right: multi TR-periodic spoiling. Gradients on the phase-encoding axis are fully balanced (not shown).
The necessary gradient moments for TR-periodic and multi-TR-periodic spoiling are compared in Table 1. At higher echo-shifting factors n, multi-TR-periodic spoiling becomes increasingly more efficient in terms of gradient switching: For n > 1, the maximum spoiler moment that is applied directly after the radiofrequency (RF) pulse is the same for both methods and is proportional to n. However, multi TR-periodic spoiling does not require a spoiler after the readout event, whereas TR-periodic spoiling requires one with an area proportional to n 1 1. Although total gradient switching (parenthesed values in Table 1) is for one echo-shift relatively similar for both methods, multi-TR-periodic spoiling still holds a small advantage as the larger part of the multi-TR-periodic spoiler gradient moment is required between the RF pulse and the readout window. This makes achieving longer echo times easier and more efficient. However, multi-TR-periodicity is not without drawbacks: Strong variations in uncompensated eddy currents between neighboring lines when using a na€ıve linear reordering strategy will in most cases result in strong ghosting artifacts. Switching to centric reordering (or a
similar suitable reordering scheme) can help to reduce the impact of these artifacts by distributing them over the full FoV, making them less pronounced and less visible. However, even in best case, the resulting artifacts will still appear as a slight noise enhancement. One general drawback of echo-shifting irrespective of the spoiling strategy is that it leads to a signal reduction as signal components are lost in higher-order pathways due to the application of multiple RF pulses per TR and slice. For an echo-shifting factor of n and flip angle a, the associated signal loss factor is given by cos 2n a2 (6). Interslice Echo-Shifting Combining a 2D interleaved multislice acquisition with echo-shifting has been previously proposed for a segmented EPI sequence with TR-periodic spoiling (10). This combined approach offers two major advantages: First, it combines the advantages of both concepts, that is, TE and TR can be significantly increased without affecting the acquisition time per volume (sometimes misleadingly called volume TR). Second, the fact that
Table 1 Spoiler Gradient Moments for GRE, as well as for ES-GRE with TR-Periodic and Multi-TR-Periodic Spoiling in Read Direction (Dephasing of 2p, Equivalent to One k-space)
Sequence type GRE TR-periodic ES-GRE (7,8) Multi-TR-periodic ES-GRE (5,6)
Echo shifts – 1 2 n 1 2 n
Read
Slice
Periodic use in TR k mod (n11)
pre-ADC
post-ADC
pre-ADC
post-ADC
– – – – 0 1 0 1...2 0 1...n
0 (0.5) 1 (0.5) 2 (1.5) n (n 0.5) 1 (0.5) 1 (1.5) 2 (1.5) 1 (1.5) n (n 0.5) 1 (1.5)
1 (0.5) 2 (1.5) 3 (2.5) n þ 1 (n þ 0.5) 0 (0.5)
0 (0.5) 0 (0.5) 0 (0.5) 0 (0.5) 0.5 (1)
– 0 (0.5) 0 (0.5) 0 (0.5) 1 (0.5)
0 (0.5)
0.5 (1)
0.75 (0.25)
0 (0.5)
0.5 (1)
nþ1 1 2n (2n)
Gradient moments are given relative to the moment of the readout and slice gradient, respectively. The numbers in parentheses are with dephasing from the readout and slice gradient accounted for (ignoring gradient ramps and assuming a symmetric readout and RF pulse). The total amount of gradient switching in the imaging sequence is better represented by these parenthesed values. As can be seen from the table, total gradient switching is much lower for multi-TR periodic spoiling than for TR-periodic spoiling, especially for high echo-shifting factors.
1700
only a single RF pulse is applied per TR and slice means that there is no echo-shifting related signal loss associated with this concept. As the delayed echo acquisition is interleaved with the slice acquisition, we propose to call this concept interslice echo-shifting (IS-ES) to better distinguish it from conventional echo-shifting. Simplified sequence diagrams for IS-ES with TR-periodic and multi TR-periodic spoiling are shown in Figure 1. Another advantage of interslice echo-shifting is that it allows the use of multi TR-periodic spoiler gradients without risking ghosting artifacts due to uncompensated eddy currents: if the number of slices is an integer multiple of n þ 1 (the periodicity of multi TR-periodic spoiling), all k-space lines of each respective slice are acquired using the same spoiler gradients. As a result, multi TR-periodic spoiling is in fact TR-periodic again, so that ghosting artifacts due to non-TR-periodic eddy currents can be avoided independent of the line reordering strategy. This removes the main disadvantage of multi-TR-periodic echo-shifting, making it the preferable spoiling strategy for interslice echo-shifting. Thus, we chose to use multi-TR periodic spoiling for all IS-ES experiments presented in this work. Note that for simplification, we will continue to refer to the two spoiling strategies by “TR-periodic” and “multi-TRperiodic,” although both strategies are in fact TR-periodic for each slice when used with interslice echo-shifting.
Ehses et al.
each lasting for the acquisition duration of five volumes (18.9 s). In total, 10 periods of each condition (100 time frames) were acquired in 6:20 min. A total of 12 interleaved slices were measured using the interslice echoshifted sequence with an echo-shifting factor of two. Other sequence parameters were as follows: nominal flip angle ¼ 25 , TR ¼ 84 ms (7 ms per slice), TE ¼ 18 ms, readout bandwidth ¼ 390 Hz/pixel, voxel size ¼ 0.8 0.8 1.0 mm3 (þ20% interslice gap), parallel imaging acceleration factor ¼ 3 (with 48 autocalibration lines acquired once at the beginning of the experiment), 6/8 partial Fourier, acquisition time per volume ¼ 3.78 s. Images were reconstructed offline from the raw data as described below. After motion correction using AFNI (12), functional analysis was performed with FEAT from the FSL package (13), using standard parameter settings and no spatial smoothing. The FEAT analysis is based on a model that is created from a convolution of the paradigm convolved with a standard hemodynamic response function. Activation maps are then obtained by fitting the acquired time series to the model. For temporal SNR comparison, one of the six volunteers was additionally scanned using a standard GRE sequence. Parameters were the same as in the IS-ES scan except that this sequence only allowed the acquisition of four slices in the given TR. Image Reconstruction
METHODS
To illustrate contrast and SNR similarities between the interslice echo-shifted sequence and a conventional gradient echo, high-resolution multislice data were acquired from one of the six volunteers. For better comparability, all contrast and SNR relevant sequence parameters were kept the same in both sequences: nominal flip angle ¼ 31 , TR ¼ 712.8 ms, TE ¼ 16 ms, readout bandwidth ¼ 290 Hz/pixel, voxel size ¼ 0.4 0.4 1.0 mm3 (þ20% interslice gap), acquisition time ¼ 5 min. One echo-shift was sufficient for achieving the chosen echo time without requiring any idle time in the sequence. The choice of a common TR and TE in both sequences allowed the acquisition of twice as many slices in the IS-ES scan (36 and 72 slices were acquired in the gradient-echo and IS-ES experiment, respectively).
Image reconstruction was performed in Matlab (The Mathworks). As a first step, the multichannel raw data was prewhitened (14) using noise data that was acquired prior to each scan (4096 noise samples). K-space lines that were skipped in the fMRI experiment due to parallel imaging and partial Fourier were reconstructed using GRAPPA (15) and a projection onto convex sets algorithm (16), respectively. The highresolution T2 -weighted data were fully sampled, so that a simple 2D fast Fourier transform was sufficient to convert the signal to the image domain. Coil data were then combined by estimating combination weights and coil sensitivity maps using the adaptive combine method (17). Coil data from the BOLD experiment were combined using Walsh’s original method, whereas the structural data were combined using a recently proposed modification (18), that improves phase reconstruction by enforcing its smoothness in a region around each voxel (a 11 11 square kernel was used in this work). The result is very similar to a homodyne filtered phase image. Signal level nonuniformities due to receiver coil sensitivity variations were then intensity normalized using the estimated coil sensitivity maps according to an approximation proposed by Griswold et al. (19). For better comparability, the two highresolution data sets were normalized using the same normalization factors (determined from the IS-ES scan).
BOLD fMRI Experiment
SNR and tSNR Analysis
BOLD fMRI data were acquired while each of the six subjects was performing a simple bilateral finger tapping task. Starting from a resting period, the paradigm alternated between a tasking period and a resting period,
Signal-to-noise ratio was calculated from the highresolution T2 -weighted datasets and the acquired noise data using the pseudo multiple replica method (20). To this end, a total of 100 pseudo replicas were generated
All experiments were performed on a 9.4 Tesla system (Siemens Healthcare, Erlangen, Germany) with a maximum gradient strength and slew rate of 70 mT/m and 200 mT/m/ms, respectively. A custom-built head coil (11) was used for RF transmission and reception (16 transmit/31 receive channels). Data were acquired from six healthy volunteers after obtaining informed consent and approval by the local ethics committee. High-Resolution T2 -Weighted Imaging Experiments
Interslice Echo-Shifting
1701
FIG. 2. Comparison of the high-resolution T2 -weighted data from the conventional gradient-echo scan and the interslice echo-shifted scan at an in-plane resolution of 0.4 0.4 mm2 (from left to right: magnitude image, phase image, and SNR map with histogram). Note that in the same scan time and using the same TR and TE, twice as many slices were covered with the interslice echo-shifted sequence than with the conventional gradient-echo. The SNR histogram only includes brain voxels from the 36 slices that were acquired in both scans.
by adding pseudorandom permutations of the noise data to the image. Furthermore, temporal SNR (tSNR) maps were calculated from the IS-ES and conventional gradient-echo BOLD data of one of the six volunteers. Susceptibility Weighted Imaging Reconstruction Susceptibility weighted images were reconstructed from the high-resolution T2 -weighted data according to Haacke et al. (2). A minimum intensity projection over five slices (6 mm) was then calculated from the susceptibility weighted imaging data. RESULTS Figure 2 shows magnitude, phase and the SNR map for the two high-resolution T2 -weighted datasets with an in-plane resolution of 0.4 0.4 mm2. The higher spatial coverage of the IS-ES data can be appreciated by looking at the sagittal
view. Contrast is very similar in the magnitude and phase images for both datasets (Fig. 2, left and middle). The similarity between the two is also evident in the SNR maps and histograms, shown on the right side of Figure 2. The minimum intensity projection of the susceptibility weighted imaging reconstruction, shown in Figure 3, is again comparable for the two datasets and shows very high detail. Three representative minimum intensity projections are shown, with the first slice missing from the conventional gradient-echo data due to the lower spatial coverage. The most inferior slice of the IS-ES GRE data exhibits reduced signal in some areas due to inhomogeneities of the transmit field. Figure 4 top shows the activation maps from the fMRI experiments overlayed over an example time frame for one exemplary slice for each volunteer. The results indicate that IS-ES GRE is well suited for high-resolution functional imaging at ultrahigh field. Strong activations
1702
Ehses et al.
FIG. 3. Minimum intensity projection of a susceptibility weighted imaging reconstruction of the high-resolution T2 weighted data. The projection thickness is 6 mm (five slices). Three representative slices are shown, with the first slice missing from the conventional scan’s data due to its lower spatial coverage. Contrast is very similar for both datasets (compare the zoomed insets).
were found in the primary motor cortex area of both hemispheres: maximum z-score was 15.7 for volunteer #3. Activations found in the somatosensory area were relatively weak, which may be explained by lower SNR due to a relatively low coil sensitivity and high g-factor loss in this area. The signal time course and the FEAT model for the voxel with maximum z-score is shown on the left below the activation maps. A very strong signal change (640%) can be observed here, explained by the fact that the voxel lies inside a vein. To show a time course that is more representative for gray matter, a median time course was calculated from all voxels with z-scores greater than four. The median signal change was approximately 64%, as shown at the bottom left corner of Figure 4. Temporal SNR (tSNR) maps for IS-ES GRE and a standard GRE BOLD acquisition are shown for one of the central slices of one volunteer in the bottom right corner of Figure 4. Median tSNR in the brain was 13% lower in IS-ES than in standard GRE (tSNRGRE =tSNRIS-ES ¼ 19:5=17:2). DISCUSSION AND CONCLUSION IS-ES GRE allows for efficient generation of T2 -weighted contrast by combining echo-shifting with an interleaved slice excitation order. To our knowledge, the combination of echo-shifting with an interleaved slice acquisition has first been proposed in 2006 for a segmented EPI sequence by Gibson et al. (10). However, the authors did not discuss one of the main advantages that this new concept holds over conventional echo-shifting, namely that there is no echo-shifting related signal loss due to higher coherence pathways as only a single RF pulse is applied per TE in each slice. Furthermore, our work
shows that multi-TR-periodic spoiling is preferable to the TR-periodic spoiling strategy used by Gibson et al. It is difficult to overstate the difference between the concept of echo-shifting and interslice echo-shifting: from the definition of echo-shifting follows that an interslice echo-shifted sequence is technically not echoshifted at all as the combination with an interleaved slice excitation order results in a TR that is always larger than TE. IS-ES GRE merely uses the same gradient spoiling technique as conventional echo-shifting to shift the echo from one sub-TR unit to another (as illustrated in Fig. 1). This insight illustrates why IS-ES does not suffer from any echo-shifting related signal loss. Furthermore, it explains why the multi TR-periodic spoiling strategy of echo-shifted sequences (5,6) is in fact TR-periodic for interslice echo-shifting. This makes multi-TR-periodic spoiling the preferable spoiling strategy for IS-ES GRE by reducing the required spoiler gradient moments, especially for higher echo-shifting factors (compare Table 1). The results from the high-resolution T2 -weighted data show that one echo-shift allows covering twice as many slices in the same acquisition time as with a standard GRE sequence while achieving similar magnitude and phase contrast as well as SNR (Fig. 2). Note that depending on the acquisition parameters and the imaged body region, IS-ES GRE may show a slightly different contrast than standard GRE due to stronger gradient switching and a higher number of RF pulses per unit time in IS-ES GRE: the former makes IS-ES GRE more sensitive to flow and diffusion while the latter leads to a stronger magnetization transfer induced signal attenuation, especially in white matter. Flow sensitivity and signal loss due to diffusion can be estimated from the
Interslice Echo-Shifting
1703
FIG. 4. Results from the finger tapping experiments with an in-plane resolution of 0.8 0.8 mm2: Top: Overlay of BOLD activation maps with an example time frame for a representative slice from each of the six volunteers. Bottom left: Time series (black) and fitted regressor (red) for volunteer #3 for the voxel with maximum z-score (¼ 15.7; voxel position indicated by white arrow), and for the median over all voxels with z-scores greater than 4 (yielding 5428 voxels). Bottom right: tSNR maps for IS-ES and standard GRE as calculated from the fMRI time-series for a central slice of one of the volunteers (#6). Median tSNR in the brain was 17.2 and 19.5 in the IS-ES and standard GRE, respectively (note that three times as many slices as in GRE were acquired in the IS-ES scan).
gradient table of the sequence: maximum b-values in the BOLD fMRI scan were 4:8 s=mm2 and 62:6 s=mm2 for the standard and IS-ES GRE, respectively. As only a single echo-shift was performed in the high-resolution IS-ES scan, the corresponding b-value was with 38:0s=mm2 (standard GRE: 4:6s=mm2 ) lower than in the BOLD experiment. Assuming a mean apparent diffusion coefficient of 0:8 103 mm2 =s in gray matter (21), this translates to a relatively modest expected signal loss in gray matter of 2.6% in the high-resolution scan and 4.5% in the fMRI scan (relative to standard GRE). Flow sensitivity in the echo-shifted scans, as measured by total first gradient moment m1, was 6–9 times higher than in standard GRE
(high-resolution experiments: m1;GRE =m1;IS-ES ¼ 119=708; ms2 ; BOLD: m mT ms2 ). ½mT m 1;GRE =m1;IS-ES ¼ 87=809; ½ m Magnetization transfer effects are expected to be small, as relatively low flip angles were used due to specific absorption rate (SAR) constraints. One of the main advantages that the proposed IS-ES GRE sequence holds over a single-shot EPI sequence is that it is relatively insensitive to B0 field inhomogeneities that are a common problem for EPI at high field strengths. Images from the IS-ES BOLD experiment and its activation maps are virtually free from distortions (see Fig. 4). Apart from this advantage which is particularly
1704
important at higher field, fMRI with IS-ES GRE also becomes more feasible with increasing field: As the echo time for optimal BOLD contrast decreases, less echoshifts are required to optimize IS-ES GRE for BOLD imaging. This reduces the amount of necessary gradient switching, resulting in less acoustic noise, a lower chance of peripheral nerve stimulation, and less diffusion related signal loss. Another advantage of the proposed sequence is that it reduces T2 -related blurring due to a shorter readout, meaning that the nominal resolution is closer to the actual resolution than in EPI. In contrast, the high nominal resolution that is achievable with EPI at ultrahigh-field is often partially offset by these blurring effects. However, these advantages come at the cost of a longer scan time (3.78 s for 12 slices compared to approximately 2 s for a typical submillimeter EPI with 20 or more slices) and a lower SNR efficiency compared to EPI. Furthermore, single-shot EPI is less sensitive to flow and other dynamic effects than IS-ES GRE. Magnetic field fluctuations, often caused by patient motion and breathing, are another problem for multishot sequences as they can cause signal fluctuations and image ghosting. Phase stabilization (22) (sometimes called frequency locking) can be used to alleviate this problem. This technique relies on the acquisition of an extra nonphase-encoded gradient-echo in each TR from which the field fluctuation can be determined. The phase of each acquired k-space line can then be adjusted to correct for the fluctuating field. While no phase stabilization was used in this work, it may be a useful tool to suppress signal fluctuations and to increase temporal SNR. However, as IS-ES GRE does not require idle time to achieve long echo times, adding an extra readout will somewhat reduce the time-efficiency of the sequence. Although image acquisition in the presented fMRI experiment was already accelerated using a GRAPPA acceleration factor of 3 and 6/8 partial Fourier, we believe that further speed improvements are possible. One promising approach would be to excite and acquire multiple slices simultaneously using the multislice CAIPIRINHA technique (23), which has also recently found application in EPI (24). Depending on coil geometry, CAIPIRINHA potentially allows higher parallel imaging acceleration factors by exploiting coil-sensitivity variations in both phaseencoding and slice-selection direction. Although multiple excitation volumes are absolutely necessary for IS-ES, a 3D acquisition in combination with 2D CAIPIRINHA (25) would also be possible by interleaving multiple 3D slabs. Finally, the efficiency of k-space sampling can be improved by moving to a segmented EPI acquisition as in Gibson et al. (10) or to a non-Cartesian trajectory such as spiral. In addition to the presented application of IS-ES GRE for high-resolution anatomical imaging and BOLD fMRI, this technique may also be useful for other applications were long echo times are essential such as quantitative susceptibility mapping (3) and proton resonance frequency shift thermometry (4). REFERENCES 1. Ogawa S, Menon RS, Tank DW, Kim SG, Merkle H, Ellermann JM, Ugurbil K. Functional brain mapping by blood oxygenation level-
Ehses et al. dependent contrast magnetic resonance imaging. A comparison of signal characteristics with a biophysical model. Biophys J 1993;64:803–812. 2. Haacke EM, Xu Y, Cheng YCN, Reichenbach JR. Susceptibility weighted imaging (SWI). Magn Reson Med 2004;52:612–618. 3. De Rochefort L, Brown R, Prince MR, Wang Y. Quantitative MR susceptibility mapping using piece-wise constant regularized inversion of the magnetic field. Magn Reson Med 2008;60:1003–1009. 4. Ishihara Y, Calderon A, Watanabe H, Okamoto K, Suzuki Y, Kuroda K. A precise and fast temperature mapping using water proton chemical shift. Magn Reson Med 1995;34:814–823. 5. Moonen CTW, Liu G, van Gelderen P, Sobering G. A fast gradientrecalled MRI technique with increased sensitivity to dynamic susceptibility effects. Magn Reson Med 1992;26:184–189. 6. Liu G, Sobering G, Olson AW, van Gelderen P, Moonen CT. Fast echo-shifted gradient-recalled MRI: combining a short repetition time with variable T2 weighting. Magn Reson Med 1993;30:68–75. 7. Moonen CT, Barrios Fa, Zigun JR, Gillen J, Liu G, Sobering G, Sexton R, Woo J, Frank J, Weinberger DR. Functional brain MR imaging based on bolus tracking with a fast T2 -sensitized gradient-echo method. Magn Reson Imaging 1994;12:379–385. 8. de Zwart JA, van Gelderen P, Kelly DJ, Moonen CT. Fast magneticresonance temperature imaging. J Magn Reson B 1996;112:86–90. 9. van Gelderen P, Duyn JH, Ramsey NF, Liu G, Moonen CTW. The PRESTO technique for fMRI. NeuroImage 2012;62:676–681. 10. Gibson A, Peters AM, Bowtell R. Echo-shifted multislice EPI for high-speed fMRI. Magn Reson Imaging 2006;24:433–442. 11. Shajan G, Kozlov M, Hoffmann J, Turner R, Scheffler K, Pohmann R. A 16-channel dual-row transmit array in combination with a 31element receive array for human brain imaging at 9.4 T. Magn Reson Med 2014;71:870–879. 12. Cox RW. AFNI: software for analysis and visualization of functional magnetic resonance neuroimages. Comput Biomed Res 1996;29:162–173. 13. Smith SM, Jenkinson M, Woolrich MW, et al. Advances in functional and structural MR image analysis and implementation as FSL. NeuroImage 2004;23(Suppl 1):S208–S219. 14. Pruessmann KP, Weiger M, B€ ornert P, Boesiger P. Advances in sensitivity encoding with arbitrary k-space trajectories. Magn Reson Med 2001;46:638–651. 15. Griswold MA, Jakob PM, Heidemann RM, Nittka M, Jellus V, Wang J, Kiefer B, Haase A. Generalized autocalibrating partially parallel acquisitions (GRAPPA). Magn Reson Med 2002;47:1202–1210. 16. Haacke E, Lindskogj E, Lin W. A fast, iterative, partial-fourier technique capable of local phase recovery. J Magn Reson 1991;92:126–145. 17. Walsh DO, Gmitro aF, Marcellin MW. Adaptive reconstruction of phased array MR imagery. Magn Reson Med 2000;43:682–690. 18. Inati SJ, Hansen MS, Kellmann P. A solution to the phase problem in adaptive coil combination. In Proceedings of the 21st Annual Meeting of the ISMRM, Salt Lake City, Utah, USA, 2013. p 2672. 19. Griswold M, Walsh D, Heidemann R, Haase A, Jakob P. The use of an adaptive reconstruction for array coil sensitivity mapping and intensity normalization. In Proceedings of the 10th Annual Meeting of the ISMRM, Honolulu, Hawaii, USA, 2002. p. 2410. 20. Robson PM, Grant AK, Madhuranthakam AJ, Lattanzi R, Sodickson DK, McKenzie Ca. Comprehensive quantification of signal-to-noise ratio and g-factor for image-based and k-space-based parallel imaging reconstructions. Magn Reson Med 2008;60:895–907. 21. Le Bihan D, Mangin JF, Poupon C, Clark Ca, Pappata S, Molko N, Chabriat H. Diffusion tensor imaging: concepts and applications. Journal of magnetic resonance imaging: JMRI 2001;13:534–546. 22. Henry PG, van de Moortele PF, Giacomini E, Nauerth A, Bloch G. Field-frequency locked in vivo proton MRS on a whole-body spectrometer. Magn Reson Med 1999;42:636–642. 23. Breuer FA, Blaimer M, Heidemann RM, Mueller MF, Griswold MA, Jakob PM. Controlled aliasing in parallel imaging results in higher acceleration (CAIPIRINHA) for multi-slice imaging. Magn Reson Med 2005;53:684–691. 24. Setsompop K, Gagoski Ba, Polimeni JR, Witzel T, Wedeen VJ, Wald LL. Blipped-controlled aliasing in parallel imaging for simultaneous multislice echo planar imaging with reduced g-factor penalty. Magn Reson Med 2012;67:1210–1224. 25. Breuer FA, Blaimer M, Mueller MF, Seiberlich N, Heidemann RM, Griswold Ma, Jakob PM. Controlled aliasing in volumetric parallel imaging (2D CAIPIRINHA). Magn Reson Med 2006;55:549–556.
