Fast High-Resolution Microstructure Diffusion MRI Exploiting Data Redundancy

Project Summary/Abstract Diffusion magnetic resonance imaging (dMRI) is indispensable in everyday clinical neuroimaging due to its capacity for noninvasive whole-brain imaging. It has a significant impact on diagnosing conditions affecting millions of Americans. Furthermore, advanced dMRI holds promise for mapping microstructural tissue features, revealing hidden damage in white matter, guiding neurosurgery, and studying complex brain structures. However, lengthy acquisition times prevent advanced diffusion encodings and high-resolution imaging (≤1mm³) from reaching clinical applications. To bridge this gap, I propose a novel approach to accelerate advanced dMRI while maintaining image quality and microstructure sensitivity. This innovative method aims to achieve up to five-fold speed improvement by efficiently exploiting data redundancy. Structural MRI has only focused on spatial undersampling data acceleration techniques but dMRI datasets are intrinsically of higher dimensionality due to the multiple volumes of diffusion encodings that are acquired. My goal is to pioneer a novel imaging approach called zero-shell imaging (ZSI) that uses tissue biophysics to speed up dMRI data acquisition and improve image resolution. This technique encompasses joint undersampling in both spatial (k-space) and diffusion weighting (q-space) domains, facilitating direct reconstruction of diffusion contrasts with fewer spatial samples per diffusion encoding. Accounting for patient motion between samples and merging sequential images, I aim to optimize the allocation of scan time, to enable higher resolution and to increase the number of diffusion encodings without lengthening the overall scan time. Preliminary findings have demonstrated that ZSI can successfully undersample the diffusion acquisition space, separating diffusion weightings and directions. In Aim 1, I will optimize a method that reconstructs dMRI signal’s rotational invariants from undersampled q-space protocols. For Aim 2, I will develop a dMRI sequence that performs joint k-q-space undersampling and a motion-robust reconstruction algorithm. Both aims will include validation on healthy subjects and reproducibility assessment. In Aim 3, I will perform an evaluation of the clinical utility for tracking disease progression in multiple sclerosis (MS) patients and for detecting MS brainstem lesions. Overall, this research will develop an innovative imaging method that has the potential to transform MRI diagnosis. During the K99 phase of the award, I will benefit from the mentorship of Profs. Novikov, Fieremans, and Feng at New York University Grossman School of Medicine, by obtaining additional training in pulse sequence design and microstructure-informed image reconstruction. My proposed training plan will equip me with the necessary research and professional skills to start an independent career in the R00 phase, in which I will develop innovative solutions that harmoniously optimize image generation, data acquisition, and data modeling processes. This work is motivated by observations that the combined k-q-space samples are sparse and permit massive undersampling. In the long term, the increased speed enabled by this project may open the window to finally translate microstructure to clinic.