High-resolution segmented-accelerated EPI using Variable Flip Angle FLEET with tailored slice profiles
Avery JL Berman 1,2, Thomas Witzel1,2, William A Grissom 3,4, Daniel Park 1, Kawin Setsompop1,2,5, and Jonathan R. Polimeni1,2,5

1Athinoula A. Martinos Center for Biomedical Imaging, Massachusetts General Hospital, Boston, MA, United States, 2Department of Radiology, Harvard Medical School, Boston, MA, United States, 3Vanderbilt University Institute of Imaging Science, Nashville, TN, United States, 4Department of Biomedical Engineering, Vanderbilt University, Nashville, TN, United States, 5Harvard-MIT Division of Health Sciences and Technology, Massachusetts Institute of Technology, Cambridge, MA, United States


New evidence suggests that fMRI has spatial specificity at scales far below current voxel sizes, but encoding limits preclude single-shot EPI at sufficient spatial resolution. Segmented EPI can help overcome these limits, but is well-known to be temporally unstable. Here we propose a reordering of the EPI segments, known as FLEET, combined with variable progression of flip angles to maximize the image signal level and a tailored RF pulse design to maintain compatible slice profiles. We demonstrate that this approach provides stable segmented EPI acquisitions with negligible ghosting, and when combined with acceleration can provide submillimeter fMRI acquisitions at 3T.


There is growing evidence that hemodynamic responses to neuronal activation are far more spatially specific than previously believed,1 motivating$$$\;$$$fMRI$$$\;$$$acquisitions$$$\;$$$with$$$\;$$$higher spatial resolutions.$$$\;$$$However, we are limited by the spatial encoding capabilities$$$\;$$$of$$$\;$$$MRI to reach these higher resolutions$$$\;$$$while$$$\;$$$maintaining whole-brain coverage.2 Increasing the acceleration factor results in $$$\sqrt(R)$$$ and $$$g$$$-factor signal-to-noise ratio (SNR) penalties, and increasing the readout duration to sample further in k-space results$$$\;$$$in increased $$$T_2^*$$$-induced spatial blurring and B0 distortions. Segmented EPI is a possible solution to this problem, but it suffers from spurious ghosting resulting from inter-segment phase differences arising from motion- and respiration-induced B0 changes. It was recently shown that the temporal SNR (tSNR) of BOLD time-series based on accelerated single-shot EPI can be dramatically increased by swapping the slice-segment ordering of the multi-shot-EPI-based autocalibration scan using the fast low-angle excitation echo-planar technique (FLEET) to minimize the inter-segment delay.3,4 This method used constant, low flip angles with added dummy scans to achieve equal magnetization across segments, which led to an overall loss in tSNR—which is tolerable for ACS data acquisition. Here, we extend this approach to acquire reordered segmented, multi-shot EPI for the fMRI acquisition itself, using a variable-flip-angle (VFA) scheme (VFA-FLEET) to maximize the SNR of the fMRI data and Shinnar-Le Roux (SLR) pulses to produce consistent magnetization between segments.


By accounting for the pseudo-steady-state of VFA-FLEET, target flip angles are determined recursively:3$$ \alpha_{i-1}=\tan^{-1}(\sin(\alpha_i)).$$ To maximize magnetization, the final excitation is set to 90°, giving $$$\alpha_i$$$={45°,90°} or {35°,45°,90°} for 2- or 3-shots, respectively (with no dummies). Owing to the non-square slice profiles, using Hann-windowed sinc RF pulses results in non-uniform slice profiles from shot-to-shot;5-7 figure 1 shows how tailored SLR pulses overcome this.


Acquisition: All experiments were conducted at 3 T using a 32-channel receive coil. Four subjects (3F, 29±4-years) were scanned with conventional-segmented EPI, and VFA-FLEET EPI with sinc (VFA-FLEET-sinc) and SLR (VFA-FLEET-SLR) pulses, using 2 or 3 segments (Nseg), unaccelerated, matrix=96x96, 2.1-mm isotropic resolution, 30/33 slices (Nseg=2/3), 20% slice gap, TE=30ms, TR=Nseg×2.4s, 62 repetitions. The flip angle of VFA-FLEET that determines the image signal level is the first flip angle; to dissociate the role of segment reordering from flip angle, the conventional-segmented was repeated for each Nseg with α=90° and 45°(Nseg=2) or 35°(Nseg=3). Combined segmented-accelerated was tested on two subjects (1F, 32±5-years) using matrix=128x128, 1.5-mm isotropic resolution, 33/31 slices (Nseg=2/3), no slice gap, TE=30ms, TR=Nseg×2.2s, and all combinations of Nseg=2/3 and R=3/4 resulting in an effective acceleration of Nseg×R for each segment. Images were reconstructed offline using navigator-based ghost-correction within each segment then navigator-based ghost-correction between segments. Optionally, to account for differences in shot-to-shot signal, a scaling factor that minimized the mean-square error between navigators was applied to the segments. GRAPPA reconstruction with ACS-FLEET4 was applied to the combined accelerated segments. Analysis: For each time-series, the first two volumes were discarded, and the remaining volumes were motion-corrected and linear-drift-corrected. tSNR$$$\;$$$and$$$\;$$$skew—the$$$\;$$$deviation$$$\;$$$of$$$\;$$$a$$$\;$$$voxel’s temporal intensity distribution from normality—were used to characterize the various acquisitions.


Figure 2 illustrates how spurious ghosts in the conventional-segmented images are effectively eliminated in both VFA-FLEET sequences and stable ghosts in the VFA-FLEET-sinc images are reduced in the VFA-FLEET-SLR images. In Figure 3, the impact of spurious ghosting is reflected in the tSNR and temporal skew maps. In Figure 4, the group-averaged whole-brain tSNR was highest for the conventional-segmented acquisitions and lowest for VFA-FLEET-SLR although differences in flip angle and slice profile explain some of these differences. The intersegment-normalization did remove stable ghosts in both VFA-FLEET acquisitions, but at the cost of decreased tSNR. Figure 5 shows the feasibility of acquiring sub-mm3 resolutions with just R=4 VFA-FLEET-SLR.


While conventional-segmented had the highest tSNR, it was marred by nonuniformity that would make its sensitivity to activation spatially heterogeneous. Although VFA-FLEET-SLR had reduced group tSNR, its homogeneity and its ability to faithfully excite the same slice profile make it the more attractive option since it inherently reduces ghosting. As we translate this to 7T, where B0-fluctuations are longer-ranging, spurious ghosting will be exacerbated, further motivating this technique. Care will need to be taken to account for the increased B1-inhomogeneity at ultra-high fields. Finally, segmentation decreases the temporal resolution by Nseg; however, VFA-FLEET is compatible with simultaneous multi-slice imaging, and the RF phase can be designed to eliminate the CAIPI blips.8


VFA-FLEET offers a solution to the encoding limits on single-shot EPI and to spurious ghosting of conventional-segmented EPI. While stable ghosts remained in the VFA-FLEET-sinc acquisition, using tailored-SLR RF pulses to improve the signal uniformity between segments nearly completely eliminated them. With improved spatial homogeneity of tSNR relative to conventional-segmented, VFA-FLEET, combined with acceleration, may provide more reliable detection of brain activity at ultra-high-resolution while maintaining whole-brain coverage.



1 Uludag, K. & Blinder, P. Linking brain vascular physiology to hemodynamic response in ultra-high field MRI. Neuroimage 168, 279-295, doi:10.1016/j.neuroimage.2017.02.063 (2018).

2 Polimeni, J. R. & Wald, L. L. Magnetic Resonance Imaging technology-bridging the gap between noninvasive human imaging and optical microscopy. Curr Opin Neurobiol 50, 250-260, doi:10.1016/j.conb.2018.04.026 (2018).

3 Mansfield, P. Spatial mapping of the chemical shift in NMR. Magn Reson Med 1, 370-386 (1984). 4 Polimeni, J. R. et al. Reducing sensitivity losses due to respiration and motion in accelerated echo planar imaging by reordering the autocalibration data acquisition. Magn Reson Med 75, 665-679, doi:10.1002/mrm.25628 (2016).

5 Pauly, J., Leroux, P., Nishimura, D. & Macovski, A. Parameter Relations for the Shinnar-Leroux Selective Excitation Pulse Design Algorithm. IEEE transactions on medical imaging 10, 53-65, doi:Doi 10.1109/42.75611 (1991).

6 Kim, S. G., Hu, X., Adriany, G. & Ugurbil, K. Fast interleaved echo-planar imaging with navigator: high resolution anatomic and functional images at 4 Tesla. Magn Reson Med 35, 895-902 (1996).

7 Kang, D. H., Chung, J. Y., Kim, D. E., Kim, Y. B. & Cho, Z. H. in 20th International Society of Magnetic Resonance in Medicine Annual Meeting. 4175.

8 Kang, D. H. et al. in 19th International Society of Magnetic Resonance in Medicine Annual Meeting. 4575.

9 Polimeni, J. R. et al. in 20th International Society of Magnetic Resonance in Medicine Annual Meeting. 2222.


Figure 1: Sinc vs. SLR RF pulses and slice profiles for 2-shots and 3-shots. S1 corresponds to the RF/profile for shot 1, S2 to shot 2, and so on. Due to $$$M_z$$$ being between shots, the sinc pulses generate non-uniform slices profiles in magnitude (column 2), phase (column 4), and integrated $$$M_{xy}$$$ (column 5). The SLR pulses properly account for this. The S2 and S3 SLR pulses contain more RF energy near their end, reflecting the fact that they must produce more high-frequency excitation to achieve the same out-of-slice to in-slice transitions as $$$M_z$$$ is attenuated between shots.

Figure 2: Animated GIF of segmented EPI acquisitions including conventional, VFA-FLEET with standard sinc excitation pulses and VFA-FLEET with the proposed tailored SLR pulses. The top row shows the reconstructed data across ~4 minutes. In each frame, the image is windowed both at a standard level and with an increased maximum intensity to visualize the background ghosts caused by misalignment across segments. The bottom row displays the estimated head displacement parameters (estimated during post-processing motion correction) with the orange dot indicating the time point corresponding to the time-series movies above. Red arrows indicate sporadic ghosts; yellow indicate stable ghosts.

Figure 3: tSNR and temporal skew maps from a subject who showed excessive motion (top) and another subject with normal motion (<0.5mm). In both cases, the tSNR in the conventional segmented acquisition shows spatial heterogeneity arising from spurious ghosting. Deviations from 0 in the skew maps indicate where ghosting is prominent. The overall tSNR of the VFA-FLEET acquisitions are lower than the conventional-segmented due to their decreased effective flip angle (45°/35° for 2/3-shot vs. 90° for conventional); however, the tSNR of the conventional-segmented acquisitions suffer from spatial heterogeneity that is highlighted in the skew maps.

Figure 4: Mean whole-brain tSNR across subjects. For the unaccelerated case, conventional-segmented was repeated with the same effective flip angle as the VFA-FLEET acquisitions ( “Conv.-Eff. α”, α = 45° or 35°). The different colour bars represent reconstruction with inter-segment phase correction (blue) or with phase correction + inter-segment normalization (orange) to account for shot-to-shot signal non-uniformities like slice profile effects. Normalization reduced the ghosting (not shown) but also reduced the tSNR (although, in the accelerated cases, its effect was marginal). Overall, the tSNR of VFA-FLEET is lower than conventional segmented, however, this does not show the spatial heterogeneity in Conventional-segmented.

Figure 5: The “ultra”-high-resolution BOLD images acquired in two subjects at 3 T and averaged across 10 repetitions using VFA-FLEET-SLR (top) and conventional-segmented (bottom). This proof-of-principle demonstrates the ability of segmented-accelerated to acquire sub-mm3 voxels.

Proc. Intl. Soc. Mag. Reson. Med. 27 (2019)