To demonstrate high-resolution, high SNR diffusion imaging using 3D multi-slab EPI at 7T, with improved slab-boundary correction and auto-calibration acquisition.Optimal SNR efficiency for spin-echo diffusion imaging is associated with very short repetition times (TR=1-2s), which even the state-of-the-art simultaneous multi-slice methods struggle to achieve. By comparison, 3D multi-slab imaging is highly compatible with this regime^1,2, and can also achieve thin slices. These properties make it an attractive method for high-resolution diffusion imaging. Nevertheless, 3D multi-slab imaging does have unique challenges, several of which are the focus of recent developments by our group and others^3,4. Signal variations at slab boundaries can lead to incorrect quantification of diffusion parameters, for which we developed a non-linear correction method⁵. Further, at this meeting we present a multi-slab adaptation of FLEET⁶, which reduces sensitivity of auto-calibration scans (ACS) to subject’s motion and respiration. Using these advances, we now aim to explore the potential for 3D multi-slab imaging to achieve high-resolution, high-SNR diffusion imaging. In particular, we demonstrate the benefits of 7T for 3D multi-slab imaging, to our knowledge the first time these highly compatible technologies have been combined.Three healthy subjects were scanned on a Siemens 7T scanner. Scan parameters: FOV 210×210×117 mm³, matrix size 204×204, 14 slabs, 10 slices/slab, slice thickness 1.03 mm, 20% slab overlap, in-plane GRAPPA 3, partial Fourier 6/8, echo spacing 0.82ms, BW 1442 Hz/pixel, TE/TR=71/2700ms, b=0, 1500 s/mm², 64 diffusion directions and 8 b=0 (phase encoding: 6 A-P, 2 P-A). One set of Slice-FLEET ACS data was acquired. Briefly, ACS scans are acquired with a 2D GRE-EPI sequence, with slice excitations matched to the reconstructed slices of the final diffusion imaging volumes. ACS data for each slice is segmented to match the in-plane acceleration factor (here, three segments) and all segments are acquired for one slice before moving on to the next, minimizing motion sensitivity. In NPEN, the slab boundary artefacts correction is formulated as a nonlinear inversion problem, and slab aliasing and saturation artefacts are corrected by iterative reconstruction. For initial estimation of slab profile (required by NPEN), one set of b0 data was oversampled by a factor of 2 along kz. The total scan time is around 35mins. Image distortions and eddy-current effects were removed using topup^7,8 and eddy, respectively. Diffusion tensor was calculated using FSL’s FDT toolbox. BedpostX was used for multi-fibre estimation within imaging voxels⁹. Figure 1 shows isotropic-resolution b=0 and DWI images acquired at 7T, demonstrating high SNR under a challenging protocol with both high spatial resolution (~1mm isotropic) and b-value (1500 s/mm²) in 27s scan time. However, there are clearly some residual artefacts at the slab boundary (green circle in the b=0 image), which the same NPEN reconstruction was able to robustly correct at 3T. The location of these artefacts suggests that they relate to severe B0/B1 inhomogeneity (which may interfere with the NPEN assumption that slab boundaries are periodic along z direction). It should be also noted that the low signal intensity at temporal lobes (yellow circles) is due to B1 inhomogeneity, which could be improved with parallel transmit technology. Despite these artefacts, we are able to reconstruct high quality visualizations of the underlying fibre architecture from this data. Figure 2 demonstrates fibre reconstructions in the pons, where the corticospinal tracts (running superior-inferior) interdigitate with the decussation of the pons (crossing right-left). On the top, a single-fibre reconstruction is dominated by the corticospinal tract (blue), indicating that the spatial resolution is insufficient to resolve this interdigitation. However, the multi-fibre reconstructions on the bottom are able to robustly estimate both populations within a voxel, demonstrating high contrast and SNR in the data. Figure 3 demonstrates high-resolution depiction of cortical anisotropy, which is highly consistent across three subjects. Figure 4 shows directionally-coded colour map of a sagittal slice, in which thin tracts can be consistently captured from the high-resolution data (red arrow) from all three subjects.In this work, we combined two powerful techniques to improve diffusion MRI: 3D multi-slab acquisition (optimal SNR efficiency) and 7T (high SNR). With several technical developments, including Slice-FLEET to improve robustness of GRAPPA and NPEN to correct slab boundary artefacts, we were able to acquire ~1mm isotropic resolution diffusion data with high SNR and high b-value. Analysis of the high resolution data demonstrates its ability to resolve crossing fibers and capture thin fibre tracts.