UC San Diego

Owing to its exquisitely sensitive contrast mechanism, diffusion magnetic resonance imaging is a powerful non- invasive approach for studying the microstructural properties of the human brain in vivo. Magnetic resonance images are made sensitive to the microscopic displacements of water molecules that take place in brain tissue as part of the natural, physical diffusion process. Tissue water is used as an intrinsic probe, revealing important clues into the subtle architectural features of normal and pathologic brain tissue. Typical inferences include the intravoxel orientation distribution of neuronal fibers and changes in diffusion resulting from cell swelling in acute stroke. However, despite the many important advances made in the field of diffusion magnetic resonance imaging over the past decade, quantitative inference in the human brain remains somewhat limited due to the lack of direct quantitative validation against realistic biological architectures and practical limitations in data collection due to sub-optimal design parameters and artifacts caused by patient motion during scanning. In addition, current methods to resolve neuronal fiber orientations are unable to disambiguate fiber structures at different microscopic length (size) scales. In this dissertation I present a series of studies addressing each of these important limitations, starting with a general real-time image-based technique for motion correction in magnetic resonance images and ending with a series of studies on inferring complex fiber orientations from diffusion data, addressing issues such as quantitative histological validation, optimal acquisition, and improved multi-scale analysis