This dissertation describes three primary contributions to the field of medical imaging: (1) a mathematical model ("Blockhead") of the macroscopic and microscopic structure of the white matter of the human brain; (2) a technique for computing the parameters of such a model from MRI scans of a given subject; and (3) an application of existing statistical tools to compare instances of any different tissue models according to their accuracy and parsimony with respect to MR images of the subject. The Blockhead model has both discrete and continuous parameters. As such the fitting method demonstrates a novel synthesis of techniques from combinatorial and numerical optimization: namely, the inclusion of short gradient-descent steps into the neighborhood of a local-search solver. The thesis of this work is that this multi-scale tissue model admits of an instance that, for certain inputs with simple geometry, has better fit to the input (in the sense of accuracy and parsimony) than current voxel-oriented techniques. Furthermore the fitting method is capable, in some circumstances, of computing this model instance. The dissertation also details the shortcomings of this model and proposes future refinements to better represent realistic tissue geometry.
Miles, Jadrian,
"A Multi-Scale Model of Brain White-Matter Structure and Its Fitting Method for Diffusion MRI"
(2015).
Computer Science Theses and Dissertations.
Brown Digital Repository. Brown University Library.
https://doi.org/10.7301/Z0K64GG0
