Acquisition and reconstruction methods for hybrid 2D/3D diffusion MRI

Diffusion magnetic resonance imaging (MRI) has been increasingly used in neuroscience studies, particularly for mapping white matter tracts in the brain. Typically, diffusion MRI is acquired using a two-dimensional (2D) single-shot echo planar imaging sequence, which allows rapid acquisition and reduced sensitivity to subject motion. However, conventional 2D diffusion MRI faces many limitations, such as long TR (repetition time) which lead to low SNR (signal-to-noise ratio) efficiency, and long scan times with advanced diffusion protocols which usually require a large number of diffusion directions and/or b values. These limitations become more acute at high spatial resolution. Three-dimensional (3D) methods have also been developed for diffusion MRI, but they are not widely used for acquiring data in vivo due to the challenges in subject motion. Hybrid 2D/3D methods, including 3D multi-slab acquisition and simultaneous multi-slice acquisition, have been recently proposed to address the limitations of conventional 2D and 3D methods. However, 3D multi-slab acquisition faces the problem of slab boundary artefacts, which could decrease the image quality and propagate into diffusion quantifications. Simultaneous multi-slice acquisition would suffer from significant noise amplification when in-plane under-sampling is also applied. The work in this thesis seeks to develop acquisition and reconstruction methods to improve hybrid 2D/3D diffusion MRI. A new method is proposed to correct slab boundary artefacts in 3D multi-slab imaging, which jointly estimates the slab profile and underlying image using a nonlinear reconstruction. Correction results demonstrate superior performance compared with previously proposed methods. A k-q acquisition and reconstruction approach is developed to accelerate diffusion MRI based on Gaussian process methods. Here, we target improvements to simultaneous multi-slice imaging, demonstrating high acceleration factors in combination with in-plane under-sampling. Combinations of hybrid 2D/3D acquisitions with the ultra-high field of 7T are also investigated, which could enable high resolution diffusion MRI without substantially compromising SNR. The methods developed in this work are expected to improve the data quality and scan efficiency of diffusion MRI