Project Summary: Neurodegenerative disorders are a significant public health and economic problem and are
the leading cause of disability worldwide. Understanding the specific degenerative processes that are actively
progressing over the course of the illness is crucial for developing targeted drugs therapies and deciding
treatment options. Additionally, understanding the structural connectivity changes to tease apart the specific
circuitry affected is crucial in developing circuit specific non-invasive brain stimulation therapies. Diffusion-
based MRI assays can provide microstructural measures that are highly sensitive to (i) the neurodegenerative
processes and (ii) connectivity changes. Advanced modeling approaches can be utilized to further enhance the
specificity of the microstructural measures to the underlying neurodegenerative processes. However, their
utility is often limited to pure white matter regions. At the typical spatial resolution of diffusion MRI (~2mm
isotropic voxel size), significant partial volume effects exist in most brain voxels (e.g., voxels with multiple
tissue types, heterogenous fibers with different properties). In whole brain studies, this compromises the
specificity of the disease processes identified by the advanced modeling approaches; it also contributes to
inaccurate connectivity mapping. Additionally, the diffusion parameter encoding space is currently limited to
one or two shells of low b-values (b<2000s/mm2), which limits the unique determination of several relevant
microstructural parameters. The main objective of the proposal is the development, validation and clinical
translation of a diffusion MRI assay that enable efficient encoding of diffusion parameter space at sub-
millimeter voxel resolution for joint microstructure and connectivity mapping in the whole brain. Our overall
hypothesis is that the proposed framework can significantly improve the validity of microstructural modeling in
most brain voxels. The proposed development will make use of SNR-efficient 3D multi-slab acquisitions.
Coupled with time-efficient sparse k-q sampling, the encoding will span over multiple b-shells. To allow the
unique determination of several relevant microstructural parameters, multicompartmental T2 information will be
utilized. The proposed developments will be enabled by two advanced reconstruction methods: structured low-
rank matrix completion, a novel integrative framework for MRI reconstructions that enables several capabilities
including multi-echo imaging and self-calibrating reconstruction; and model-based deep learning, a novel deep
architecture to solve MR reconstruction algorithms using neural networks in a systematic fashion. These
methods overcome several inefficiencies associated with extending the 3D multi-slab acquisition for multi-
dimensional imaging in the k-q-TE space. To ensure scientific rigor, we will comprehensively validate our
technology on dedicated diffusion phantoms along with healthy volunteers using different quantification
metrics. We also validate the capability of the dMRI assay using a multi-modal MRI study in a cross-sectional
study on a cohort of Huntington's disease.