Impact of ultra-high gradients and signal modeling on diffusion MRI fiber tractography

Diffusion MRI and derived diffusion metrics have been widely applied in research and in clinic for over four decades. While diffusion tensor imaging (DTI) has provided robust diagnosis-aided metrics (e.g. in stroke), there are also more sophisticated approaches. Spherical deconvolution methods, for instance, can estimate the orientational information more accurately. These higher-order models can better fit the diffusion MRI signals obtained from high b-value data sets. The fiber orientation distribution estimated by the constrained spherical deconvolution and damped Richardson-Lucy methods could resolve crossing fibers and provide reliable information for the fiber orientations in the process of streamline fiber tracking. However, there are also inaccuracies in the estimations because of hardware limitations like the gradient nonlinearities and modeling limitations such as identifying appropriate response functions. In this thesis, several important factors that could affect the accuracy of the analysis and modeling of diffusion MRI data have been investigated. In addition, a general framework for estimations from multi-shell multi-tissue datasets has been proposed. Spherical deconvolution with the computation of fiber orientation distributions has been the primary approach used in the studies. All of the analyses are aimed to provide improved fiber tractography results from diffusion MRI data. Finally, we have also investigated the termination distributions in streamline fiber tracking. The influences of the seeding and prolongation parameter settings in the fiber tracking process have been quantitatively studied to provide references in developing and selecting tractography methods.