Accelerated Acquisition of Quantitative MRI Using Parametric Redundancy

摘要

Magnetic Resonance Imaging (MRI) is a non-invasive technique that can be utilized to obtain Quantitative T1 images of the brain. Unfortunately, the acquisition of Quantitative MRI (Q-MRI) is an extremely slow process, and this has prevented applications of Q-MRI in many clinical situations where low scan times are critical. Current approaches to speed up the acqui- sition of Q-MRI are inadequate and result in images that have artifacts. This thesis develops and validates novel compressed sensing algorithms that exploit information in spatial and para- metric dimensions for improved MRI T1 mapping.;In the first project, two different sampling patterns are used to acquire MRI data at multiple Flip Angles (FAs). The two sampling patterns are the traditional stack-of-stars (SOS) and the recently introduced vastly undersampled isotropic projection (VIPR). It is shown that when used in parametric dimension regularized compressed sensing algorithms, at high accelerations, VIPR outperforms SOS when estimating T1 maps.;The second project considers the Inversion Recovery acquisition, in which data is acquired at multiple inversion time points. Here, a novel parametric dimension compressed sensing regularizer, Total Generalized Variation (TGV) is developed, and for this data, it is shown that using TGV as a regularizer yields better T1 estimates than the current approach, which uses no regularization. TGV also outperforms the classical Total Variation (TV) regularizer by eliminating its well known staircase artifacts. In addition to continuous signals, it is also shown that TGV works well on discontinuous signals, since it does not oversmooth the discontinuities.;In the third project, we develop and validate a novel technique to estimate the T1 values of all voxels in the brain simultaneously using a Total Variation (TV) regularizer, and show that this outperforms the current approach of estimating T1 values of all voxels independently. Three different approaches to perform TV based T1 estimation are proposed, and simulations are performed to determine the best one.