MRI scans are primarily performed and evaluated in a qualitative way using contrast-weighted images (e.g., with T1, T2 or proton-density weighting). This image weighting is a nonlinear function of one or more of these intrinsic MR tissue parameters as modulated by external scanner settings and imperfections. In quantitative mapping of MR tissue parameters, we attempt to unravel this complex combination to provide a direct characterization of the tissue parameter in absolute units. This has potential to improve direct comparisons of scans across different institutions and/or scanners, and also facilitates the understanding of disease progression and treatment for a single patient across time. Although the potential of quantitative MRI has long been recognized, its use has been limited by lengthy acquisition times. Magnetic resonance fingerprinting (MRF) is a recent breakthrough in quantitative MRI that enables simultaneous measurements of multiple tissue parameters in a single experiment, dramatically shortening acquisition time to ~15 sec per imaging slice and providing intrinsically registered maps. However, this can still result in unacceptably lengthy acquisitions for high-resolution, volumetric quantitative imaging. For example, MRF can take up to 20 min for a volumetric whole-brain acquisition with a spatial resolution of 1.21.25 mm3, a resolution which, itself, falls short of that needed for structural neuroimaging analysis. The major deficiency is due to the sub-optimal data acquisition and image reconstruction schemes currently employed. In this application, we will optimize the data acquisition and image reconstruction for MRF by a rigorous statistical signal processing framework, with an overall goal of improving the accuracy and speed of for volumetric neuroimaging. In particular, we will exploit the tremendous flexibility/freedom inherent to volumetric acquisition and image reconstruction to improve accuracy and efficiency. Specifically, we will address the image reconstruction problem with a principled statistical reconstruction approach that incorporates (1) a data model for multi-channel acquisitions, (2) a low-rank tensor image model for volumetric time-series images, and (3) a statistical noise model. We will characterize the reconstruction performance (e.g., error bars) by calculating the constrained Cramer-Rao bounds (CRB) under low-rank tensor models. We address the data acquisition problem, by utilizing the constrained CRB as metrics to optimize MRF data acquisition parameters (e.g., flip angle and repletion time schedule) and k-space trajectories (e.g., stack-of-spiral trajectories) for improved SNR efficiency. Together, we expect that the proposed technique produces 2x more accurate MR tissue parameter maps, enabling a desirable resolution (e.g., isotropic 0.8 mm3) and a whole-brain coverage in a short acquisition time (e.g., 3 minutes). Finally, we will systematically validate the performance of the proposed technique and its utility for ageing studies, for which quantitative imaging biomarkers enabled by rapid, whole-brain MRI are playing an increasingly important role.
We plan to improve the capability of calibrating MRI images so that they are directly comparable across hospitals/institutions and also between time-points when following disease progression. We also aim to speed up the imaging process. Together we hope these two advances will improve our ability to study the changes in the human brain as it ages, and also when burdened with Alzheimer?s disease.
