Introduction

Gradient-echo echo-planar imaging (GRE-EPI) has become the method of choice for functional MRI (fMRI) due to its ultrafast acquisition and T₂^* sensitivity. However, EPI is also noted for its low effective bandwidth along the phase encoding (PE) direction¹. One consequence of this low bandwidth is a large apparent spatial shift of the lipids due to the chemical shift between the lipids and water, resulting in artifacts in the image. The shift artifact is commonly removed by prepending an RF pulse to suppress the lipid signal. In this study, we examine the potential to significantly reduce the SAR by circumventing the fat-suppression pulse at 7T. To compensate for the removal of the fat-suppression pulse we implement a reconstruction based on parallel imaging to separate the lipid and water images. By reducing SAR, an EPI scan without fat suppression can offer fMRI studies greater flexibility to shorten the repetition time and/or increase the volume coverage.

Methods

Separate lipid and water images can be recovered by considering the acquisition as a simultaneous multi slice (SMS) problem with a built in CAIPIRINHA shift – due to the chemical shift – as was demonstrated in Ref. 2. The substantial lipid-water spatial shift at 7T, for commonly used scan parameters, allows for a reliable separation. In this study we implemented a reconstruction method for three parallel imaging aspects: (i) in-plane PE acceleration, (ii) SMS acceleration, and (iii) lipid-water separation. We used BART³ with L1 norm to reconstruct the final images. The scans in this study were performed on a 7T MAGNETOM Terra (Siemens Healthcare, Erlangen, Germany) using a commercial 1Tx/32Rx head coil (Nova Medical, Wilmington, MA), and a head phantom which includes both a brain-like compartment and a neck + outer layer (skin, lipids, and muscle) compartment⁴. A set of simulations was performed to examine the effect of the lipid artifact. The simulations were based on brain GRE water and lipid images, shifting the lipid image to match the EPI shift. These images and matching sensitivity maps (generated from the images) were used to reconstruct and simulate an image without Fat Suppression, as well as a water-only image using the fat separation technique. The lipid image was scaled to match the different T₂^* decay (at the EPI TE) of the lipid and the gray matter (estimated as 12.5±2.5 msec and 24.8±1.4 msec, respectively). An additional analysis on phantom was performed to examine – as a function of the Fat Suppression flip angle – both the lipid artifact strength and the energy contribution of the Fat Suppression pulse. Finally, resting-state fMRI and an fMRI motor-task experiment were repeated to compare both the tSNR and the t-test between Fat Suppression and Fat Separation.

Results

Figure 1 shows schematically the equivalence of lipid-water separation (and in analogy also of SMS) to in-plane parallel imaging (e.g. SENSE).
Figure 2 shows the dependence of the lipid artifact strength on the phase of the lipid signal in simulations. Without fat suppression the artifact reaches a 23% error. Using “fat separation” a root-mean-square-error (rmse) of less than 1.7% is reached. The simulation also test the effect of shifting the sensitivity maps by one pixel to simulate motion during an fMRI experiment. This increased the lipid error without Fat Suppression to 32%, but the rmse of the water image (after fat separation) only increased to 3.3%.
Figure 3 shows the (relative) strength of the lipid signal as a function of the Fat Suppression flip angle. A lipid signal close to zero was found at ~125°, compared to the sequence’s default flip angle of 110° which contributes ~66% of the total energy (product GRE-EPI). SAR can be reduced by lowering the flip angle. For example, 80° will introduce only ~7% lipid artifact, but will still contribute 60% of the total energy, reducing SAR by only 15%.
Figure 4 gives the g-factor maps for different acceleration factors, demonstrating a local increase at the lipid artifact region.
Figure 5 demonstrates human imaging at two representative slices out of 60, acquired with ×3 in-plane acceleration and ×2 slice acceleration. SAR levels were 97% with fat suppression, but only 33% without. SNR and tSNR were estimated in a resting-state fMRI scenario and demonstrated a respective increase of 30% and 12%-50% upon removal of the fat-suppression pulse.
Figure 6 summarizes the fMRI motor-task study exhibiting improved statistics using Fat Separation: the number of voxels with t-test≥2.4 (p=0.99) increased by ×1.45 in one region and ×1.9 in another.

Discussion

In this study, we demonstrated a method which allows to avoid fat suppression in EPI for 7T brain imaging. This can reduce the GRE-EPI SAR to less than a half. A robust reconstruction based on SMS-like parallel imaging was demonstrated to separate the water and lipid images. fMRI scans at 7T can benefit extensively from this SAR reduction. However, the fat separation method eventually competes for the limited resources available (multi-channel information) used to accelerate the acquisition. Further research is required to optimize the FOV shift chosen for CAIPIRINHA when combined with lipid-water separation.

Acknowledgements

No acknowledgement found.

References

[1] Schmitt, F.et al., (1998) Springer Berlin Heidelberg, [2] Uecker M, Lustig M. Proc. Intl. Soc. Mag. Reson. Med. 20 (2012), 2490. [3] https://mrirecon.github.io/bart, [4] Jona et. al, NMR BioMed 2020.