# Advanced microstructure imaging at high b‐values and high resolution combining ultra‐high performance gradient diffusion imaging and model‐based deep learning demonstrated using 3D multi‐slab acquisition

## TL;DR

This paper shows how combining advanced MRI techniques with deep learning can produce high-resolution brain microstructure maps in under 15 minutes.

## Contribution

The novel combination of 3D multi-slab diffusion imaging and model-based deep learning enables high-resolution microstructure modeling in vivo.

## Key findings

- Whole-brain parametric maps were generated at 1mm isotropic resolution using a 3-shell acquisition in under 15 minutes.
- Intra-axonal diffusivities and volume fractions were consistently measured across multiple brain regions with low coefficient of variation.
- Results were validated against standard diffusion and q-trajectory encoding acquisitions.

## Abstract

To demonstrate the extended capabilities of 3D multi‐slab diffusion‐weighted acquisition (3D‐msDWI) on high‐performance gradients (HPG) to support advanced microstructure modeling for in‐vivo human studies at high resolutions.

Despite optimal SNR‐efficiency, the application of 3D‐msDWI has been limited by the long volume acquisition times (VAT) required for encoding the 3D k‐space using multi‐shot approaches. Substantial reduction of VAT is possible by employing optimized 3D k‐space under‐sampling methods. We demonstrate that with reduced VAT, 3D‐msDWI can be successfully utilized for advanced brain microstructure modeling at high resolution. HPG systems (e.g., >200 mT/m, >300 T/m/s) enable further optimization through shorter echo times at high b‐values. We evaluated the accelerated 3D‐msDWI method's ability to support diffusion studies at 1mm isotropic resolution using data collected across three shells, with b‐values extended up to 6000 s/mm2, and employing compartment models. The reconstruction employed a navigator‐based, motion‐compensated approach using a regularized, iterative model‐based algorithm.

The accelerated 3D‐msDWI framework enabled the generation of whole‐brain parametric maps of a three‐compartment model, at 1mm isotropic resolution, using a 3‐shell, 66‐direction acquisition completed in <15 min. The intra‐axonal diffusivities (in μm2/ms) and volume fractions reported from the method are as follows: 2.27 ± 0.14; 0.6 ± 0.04 in corpus‐callosum, 2.17 ± 0.09; 0.66 ± 0.03 in anterior limb of internal capsule, 2.18 ± 0.08; 0.68 ± 0.04 in posterior limb of internal capsule, 2.07 ± 0.06; 0.62 ± 0.04 in corona radiata, 2.25 ± 0.08; 0.68 ± 0.04 in cortico‐spinal tract, 2.12 ± 0.04; 0.63 ± 0.05 in superior longitudinal fasciculus, with a coefficient of variation <10% across subjects for all regions studied. The quantified values were validated using standard single‐diffusion and multi‐dimensional q‐trajectory encoding acquisitions.

The inherent optimal SNR‐efficiency of the 3D‐msDWI framework can be harnessed for whole‐brain high‐resolution advanced microstructure modeling for in‐vivo human studies, using advanced hardware and reconstruction.