To develop and evaluate a method for reducing motion sensitivity of auto-calibration data for parallel imaging in 3D MRI, with application to high-resolution diffusion imaging at ultra-high field.3D multi-slab acquisition can achieve high isotropic-resolution DWI with optimal SNR efficiency^1,2. Using single-shot EPI for each k_z partition, 3D multi-slab DWI can be as fast as conventional 2D multi-slice DWI. In this case, in-plane parallel undersampling is necessary to reduce TE and T₂^* blurring, especially at ultra-high-field. GRAPPA auto-calibration data for EPI are typically acquired in a segmented manner for a matched echo spacing with the imaging scans, and it has been shown that GRAPPA reconstruction in 2D EPI is significantly improved by the recently proposed FLEET method³: all segments of a given slice are acquired in immediate succession to minimise motion and respiration induced phase corruptions between segments. It has also been shown that FLEET-like ACS acquisition improves reconstruction in single-slab 3D EPI⁴. We therefore expect that FLEET-ACS will also benefit GRAPPA reconstruction for 3D multi-slab EPI. In this work, we implemented several possible versions of FLEET for 3D multi-slab EPI DWI, to determine how to best acquire FLEET-ACS data. In vivo results show that “slice-by-slice FLEET” provides the best reconstruction compared with other methods, enabling high-SNR DWI at 1mm isotropic resolution with a high b value of 2000s/mm².

Three possible FLEET-ACS schemes for use with 3D multi-slab imaging are depicted in Fig.1, along with the expected benefits of different approaches. These can broadly be characterised as: FLEET-like (exciting slabs aligned to the centre of the subsequent multi-slab imaging data with fully k_z phase-encodes), slab-FLEET (FLEET-like without k_z phase-encodes) and slice-FLEET (using 2D multi-slice acquisitions corresponding to all slices' locations in the imaging volume). Long acquisition times required by the FLEET-like method might undermine the goal of reduced motion sensitivity, and this option was not considered further.

We implemented and compared four methods for ACS acquisition: spin-echo single-shot (SSH) EPI, conventional spin-echo multi-shot (MSH) EPI, Slab-FLEET and Slice-FLEET. SSH and conventional MSH based methods excite the same slab as in imaging scan and one set of 2D ACS data was acquired for each slab. Slab-FLEET: for each image slab, the FLEET-ACS scan excites a slab aligned to the centre of the imaging data, but using a thinner-slab excitation to reduce signal loss due to intra-voxel de-phasing. Slice-FLEET: for each image slab, the FLEET-ACS scan acquires a set of ACS data for each slice. With thin-slice excitation, the intra-voxel de-phasing could be minimised and local coil sensitivity can be better captured.

Three healthy subjects were scanned on a Siemens 7T scanner with 32ch RX Nova coil. Scan parameters: FOV 210×210×117mm³, matrix size 204×204,14 slabs,10 slices/slab, slice thickness 1.03mm, 20% slab overlap, in-plane GRAPPA 3, partial Fourier 6/8, echo-spacing 0.82ms, BW 1442 Hz/pixel, TE/TR=71/2600ms, b=0,1000,2000 s/mm². For SSH, conventional MSH and Slab-FLEET, GRAPPA kernels calibrated from ACS data were applied to all k_z partitions. For Slice-FLEET, the GRAPPA kernel for each slice was calculated separately and a Fourier transform along k_z was applied before GRAPPA reconstruction. We acquired Slab-FLEET ACS with different slab thicknesses (100%,50%,20%,10%), and Slice-FLEET ACS with 1× and 2× imaging slice thicknesses (relative to the final reconstructed 3D volume).Fig.2 compares GRAPPA reconstructions using different ACS-acquisition methods. SSH reconstructions show residual aliasing and strong noise amplification because the distortions are not matched. While the distortions are matched in the conventional MSH method, the phase error between in-plane segments can be large, leading to aliased reconstructions. Slab-FLEET improves on this, although the choice of slab-thickness is a problem: if it is thin (10%-20%) the performance at the boundary slices degrades, as the kernel does not capture the local coil sensitivities well enough. If a thick slab is used (50%-100%), stronger noise amplification can be seen, particularly in the DWI images (red ellipses). Slice-Fleet works well in all scenarios, with a slightly better performance for the 2x slice thickness version (boundary slice of the b=0 image). Fig.3 shows a DWI image at the slab centre (b=2000 s/mm²), also demonstrating the SNR improvement in Slice-FLEET over Slab-FLEET.The SNR efficiency of 3D multi-slab DWI makes this method attractive, but data quality is easily compromised by poor-quality ACS data. We have adapted the FLEET-ACS approach for accelerated 3D multi-slab reconstruction, and demonstrate less residual artefacts and increased SNR compared to the typical ACS schemes. We find that slice-by-slice FLEET-ACS provides the highest SNR and lowest residual artefacts, as was here illustrated in application to high-resolution high-b-value 3D multi-slab diffusion imaging at 7T.