Multiple Echo and Inversion Time MPRAGE with Inner Loop

Multiple Echo and Inversion Time MPRAGE with Inner Loop GRAPPA Acceleration and Prospective Motion Correction for
Minimally Distorted Multispectral Brain Morphometry
Andre J. W. van der Kouwe1, M. Dylan Tisdall1, Himanshu Bhat2, Bruce Fischl1, and Jonathan R. Polimeni1
Radiology, Massachusetts General Hospital, Charlestown, MA, United States, 2Siemens Healthcare USA, Charlestown, MA, United States
Target audience: Clinicians and researchers interested in efficient high quality brain imaging, multispectral morphometry and tissue parameter mapping
with a single modified MPRAGE acquisition.
Purpose: Distortions and blurring in MPRAGE affect morphometric measurements (e.g. cortical thickness) that are signatures of brain disease1.
Remaining B0-inhomogeneities after shimming result in accumulated phase errors across
each readout and therefore erroneous spatial encoding (susceptibility distortions). In addition,
T2*-decay across the readout results in broadening of the point spread function (PSF).
Replacing the single, low-bandwidth gradient echo readout with multiple shorter, highbandwidth readouts results in reduced susceptibility distortion and a narrower PSF in the
readout direction. SNR is recovered by combining the readouts in image reconstruction. T2*
may be estimated by fitting signal decay across the echoes.
T1-recovery during partition (k-space “slice”) encoding after each inversion (MP) pulse in
3D MPRAGE similarly results in a broadened PSF in this phase encoding direction. Partition
encoding time can be reduced by GRAPPA acceleration in this inner phase encoding loop
(“inner-loop GRAPPA”, ILG), thus reducing distortion but also reducing SNR. Replacing each
partition encoding block with multiple ILG-accelerated blocks experiencing different inversion
times recovers SNR when the partitions are combined in image reconstruction. T1 may then
be estimated by fitting the signal predicted by Bloch simulation across inversion times2.
Images may be blurred further by subject motion during the long high resolution
acquisitions, which may require several minutes even with outer loop acceleration.
Methods: We implemented a generalized MPRAGE sequence2 that allows ≤12 gradient
echoes, ≤8 inversion times, ILG acceleration, and incorporates vNavs for real-time
prospective motion correction and reacquisition of damaged TR intervals3. The excitation
scheme is shown in Figure 1 alongside that of a standard MPRAGE protocol recommended
for brain morphometry4. Motion-corrected volumes for each TE and TI are reconstructed
immediately on the scanner using improved “IcePat” routines for GRAPPA acceleration.
The method was evaluated in two volunteers on a 3 T Skyra (Siemens Healthcare, Figure 1: Single steady-state TR (between inversions) of
Erlangen) with a 32-channel head coil. The protocol was selected to match the timing of the (left) MEMPxRAGE-ILG and (right) conventional
conventional 1×1×1 mm3 3D MPRAGE (no acceleration), with optimal gray matter/white MPRAGE sequence. (Top) Excitation scheme (blue),
matter/CSF contrast for morphometry4. MEMPxRAGE parameters: 2 gradient echoes, 3 ADC events (red). (Bottom) Signal evolution for white
inversion times, TR 2.53 s, TE 1.69/3.55 ms, BW 650 Hz/px, TI 700/1400/2100 ms, flip angle matter at 3 T, sampled at times of ADC reads.
7/7/7°, 3× ILG acceleration (40 ref. lines), 176 sagittal slices, 256×256 matrix, 1×1×1 mm3
resolution, Tacq 11:21 min:s. vNav parameters: 3D EPI, TR 11 ms, TE 5.1 ms, 32×32 matrix, 32 sagittal slices,
6/8 partition partial Fourier, 8×8×8 mm3 voxels, BW 4464 Hz/px, Tacq/nav 275 ms.
A forward model for signal generation based on Bloch
equations was implemented in Matlab (MathWorks, Natick,
MA) and used to predict signals for a given T1 relaxation
time2 (Figure 1). Figure 2 shows corresponding PSFs for
each sequence. Note the broader PSF for conventional
MPRAGE. The proton density (PD) and T1 time that best
predicted the observed signal evolution was found at each
Figure 2: PSF for white matter at 3 T in
voxel across the range of inversion times. For comparison,
inner-phase encoding direction for (left)
two multiecho FLASH (MEF) volumes were acquired on the
MEMPxRAGE-ILG TI 1 (blue), 2 (green)
same subjects: TR 20 ms, TE 1.91+1.9n, n=0,..,7, BW 650
and 3 (red) and (right) MPRAGE. Profiles
Hz/px, flip angles 5° and 30° (separate acquisitions), Tacq
scaled by mean modeled signal intensity.
5:08 min:s, 4× GRAPPA acceleration, geometry matches
MEMPxRAGE-ILG. Standard methods were used to estimate PD and T1 from the MEF scans.
Results: Figure 3 shows T1 and PD estimates from MEMPxRAGE-ILG using Bloch simulation, and
estimates for the same subject derived by fitting the steady-state equation for the MEF scans5,6. Figure
Figure 3: Estimates of T1 (top) and PD (bottom), 4 shows a FreeSurfer surface model/parcellation from the combined MEMPxRAGE-ILG volumes.
acquisition/Bloch Discussion: GRAPPA successfully recovers volumes at all TIs and TEs with minimal residual aliasing.
simulation (left) and multiecho FLASH
The more uniform signal over the partition at each
acquisition/model equation fit (right).
inversion time (due to acceleration) together with
the availability of more than two inversion times better constrains the Bloch simulation and may
provide better T1 fits. Results are comparable to T1 fitting with MEF, although the latter may be
more affected by dielectric resonance effects. The effect of the embedded vNavs is subtle and can
be modeled (Figure 1). As in traditional FLASH acquisitions and with appropriate timing, the
multiple gradient echoes from the MEMPxRAGE-ILG sequence can be used to estimate the B0 field
map, separate fat from water, and estimate T2* decay at each voxel using traditional techniques7,8.
Figure 4: Surface reconstruction (left) and surface
Bloch simulation can also be used to fit T2*, however more than two echoes would be preferable.
parcellation (right) from MEMPxRAGE-ILG.
Conclusion: MEMPxRAGE with ILG and vNavs is an efficient sequence for high quality, low
distortion and low blur acquisitions, useful for quantitative tissue parameter mapping and high resolution brain morphometry in a single acquisition.
Acknowledgements: This work was supported by: NIH R21MH096559, R01HD071664, R21EB008547, K01EB011498, K99HD074649, P41RR014075,
and the Ellison Medical Foundation. Thanks to Drs. T. Kober and G. Krueger for access to image sorting code.
References: [1] Bakkour et al., Neurology 72(12):1048-55, 2009. [2] Marques et al., Neuroimage 49(2):1271-81, 2010. [3] Tisdall et al., MRM 68(2):38999, 2012. [4] Van der Kouwe et al., Neuroimage 40(2):559-69, 2008. [5] Deoni et al., MRM 53(1):237-41, 2005. [6] Fischl et al., Neuroimage 23 suppl
1:S69-84, 2004. [7] Glover, JMRI 1(5):521-30. [8] Yu et al., JMRI 26(4):1153-61.
Proc. Intl. Soc. Mag. Reson. Med. 22 (2014)