Brain tissue viscoelasticity measured with magnetic resonance elastography¹ (MRE) has continually shown promise in assessing neurodegenerative conditions^2,3 and intracranial tumors^4,5. To develop the ability of MRE methods to capture local property measures, many recent methodological advancements have focused on the pursuit of high-resolution viscoelastic maps through improved imaging techniques^6-8. The challenge in acquiring high-resolution MRE data rests in balancing total scan time, signal-to-noise ratio (SNR), and distortions from field inhomogeneity. Our previous MRE efforts have introduced multishot spiral readouts⁶ to reduce distortion and 3D encoding to maximize SNR efficiency⁷. In this work we incorporate a multiband excitation to enable parallel imaging acceleration in the slice direction using a 32-channel head coil⁹ and further reduce scan time. We also introduce a nonlinear motion-induced phase error correction scheme¹⁰ for phase errors not handled by linear correction, such as from cardiac pulsation. The result is a flexible sequence for fast, high-resolution, whole-brain MRE that produces 2x2x2 mm³ images in 3 minutes.

Figure 1 presents the proposed sequence, which includes standard MRE features such as flow-compensated motion encoding gradients. Details of novel imaging features are:

(a) Multiband excitation: The multiband pulses excite and refocus multiple slices at once, each separated in z and with different phase to reduce peak RF power¹¹. The resulting volumes are excited in an interleaved fashion to cover the entire brain. This provides access to optimal SNR efficiency due to its use of short TR⁷.

(b) 3D distributed slab encoding: The resulting volume is encoded with a 3D stack-of-spirals that includes multishot constant density spirals¹² in-plane with k_z-blips. The total number of designed shots per volume is equal to the number of in-plane interleaved spirals times the number of excited slices. Parallel imaging acceleration with SENSE¹³ is accomplished by undersampling both in-plane (k_xy) and thru-plane (k_z).

(c) Nonlinear phase error correction: The low-resolution 3D navigator image of the distributed volume is acquired using a k_z-blipped spiral-in trajectory¹⁴. The use of a navigator before readout reduces acquisition time and RF energy deposition by removing a second refocusing pulse. Phase error maps for each shot are recovered by comparing with the mean of all navigator images for a given volume⁷. Nonlinear correction is performed by applying the negative of these phase errors during iterative image reconstruction¹⁰.

We acquired MRE data with 2x2x2 mm³ isotropic resolution using the proposed sequence with a Siemens 3T Trio and32-channel head coil. The specific sequence parameters included: 4 band excitation; 60 total slices (15 volumes); 4 in-plane k-space readouts; FOV = 240 mm; matrix = 120x120; TR/TE = 1800/75 ms. Vibrations were generated at 50 Hz using the Resoundant pneumatic actuator system and 4 phase offsets were acquired. Iterative image reconstruction included correction for field inhomogeneity distortions¹⁵. We reconstructed datasets with various undersampling patterns and with/without motion correction, and compared resulting OSS-SNR¹⁶ and shear stiffness maps from nonlinear inversion¹⁷.

Figure 2 presents reconstructed images from a single volume with different sampling patterns. From magnitude difference maps, areas of large errors due to residual aliasing and g-factor penalties are lowest in the two-direction undersampled case (R_xy/R_z = 2/2) as compared to either in-plane or thru-plane undersampling only. This is further evidenced by the 2/2 case having the smallest NRMSE relative to fully-sampled. The ability to undersample in both directions allows for higher acceleration factors without the associated artifacts that compromise MRE results.

Figure 3 presents the MRE results from the undersampled dataset compared with the fully-sampled case. This dataset still maintains a high OSS-SNR¹⁶ for inversion stability (7.6) due to the SNR efficiency of the sequence. The resulting shear stiffness map is very similar to the fully-sampled case, with an NRMSE of only 5.0%. The majority of error is concentrated at ventricles, which are not valid regions for MRE analysis given their model-data mismatch.

Figure 4 demonstrates the performance of the nonlinear motion-induced phase error correction. Compared with magnitude images, complex displacement images, and shear stiffness maps from the uncorrected dataset, the corrected results exhibit clear improvement. This is especially evident through an SNR improvement of 18% (OSS-SNR: 7.6 vs. 6.4) and corrected phase inconsistencies in the slice direction.

We have demonstrated the performance of the proposed multiband sequence in acquiring high-resolution, high-SNR brain MRE displacement data quickly. By enabling parallel imaging acceleration both in-plane and thru-plane, we created a flexible sequence that can be used to improve the clinical adoption of high-resolution MRE methods or further push the bounds of achievable resolution.