Magnetic Resonance Imaging: Myelin Water Imaging and Model-Based Image Reconstruction

Abstract

Magnetic resonance imaging (MRI) is a useful imaging modality governed by complicated physics. One important application of MRI is myelin water imaging (MWI), in which myelin water fraction (MWF), a measure of myelin content, is estimated. As a measure of myelin content, MWF can be used to track the onset and progression of demyelinating diseases such as multiple sclerosis. The traditional method used for MWI uses a multi-echo spin echo (MESE) scan that is prohibitively slow and has not been widely adopted clinically. One aim of this dissertation is to introduce new methods for MWI that are faster than the traditional method and that will allow for higher quality MWF maps. This dissertation introduces using the small-tip fast recovery (STFR) MRI scan, which is faster than the traditional scan, for MWI. In addition, this dissertation develops a new method for optimizing scan parameters for improved MWI by minimizing the estimation error of the estimator to be used. A set of STFR scans are optimized to be informative about MWF to enable accurate estimation of MWF. MWF is estimated using parameter estimation via regression with kernels (PERK), a recently developed learning-based method that can learn nonlinear functions with practical training time. PERK is trained with data simulated using a tissue model that incorporates parameters that are neglected in the traditional method for MWI, namely chemical exchange and off-resonance frequency differences. STFR-based in vivo MWF estimates comparable to MESE-based MWF estimates are obtained in about $1/5$ the scan time. In simulation, the normalized root mean squared error (NRMSE) of MWF estimates is reduced from 42% to 14%.

In addition to advances in MWI, another aim of this dissertation is to improve the quality of MRI images by correcting for physics-induced image imperfections. As implied by its name, MRI uses a large magnet for imaging. Spatial variations in the ideally uniform magnetic field due to magnetic susceptibility differences of different tissues can result in image artifacts, or imperfections in the reconstructed image. One such image artifact is signal loss, which occurs when the magnetic field varies too quickly across space, causing signals at different spatial locations to be out of phase with each other. Approaches exist that can correct for this effect to an extent, but not in very extreme cases. Therefore, this dissertation introduces a novel acquisition and reconstruction approach to tackle this problem. In particular, data is acquired with prephasing so the signals at different spatial locations start out of phase but then rephase as the signal is acquired. Then an image is reconstructed from the data using a novel model-based reconstruction technique that accounts for both the spatial variation of the magnetic field as well as the proposed prephasing acquisition scheme. In simulation, the NRMSE of a reconstructed image is reduced from 2.3% when reconstructed without prephasing to 0.2% when reconstructed using the proposed prephasing-based approach.