Date of Award
Doctor of Philosophy
Electrical and Computer Engr
Eddie L. Jacobs
Wilburn E. Reddick
Diffusion Tensor Imaging (DTI) is a powerful medical imaging technique that provides a unique method to investigate the structure and connectivity of neural pathways. DTI is a special magnetic resonance imaging (MRI) modality that combines the principles of magnetic resonance with molecular diffusion to trace the motion of water molecules. In the central nervous system, where nerve fibers are packed in highly-directional bundles, these molecules diffuse along the orientation of the fibers. Hence, characterizing the motion of water with DTI delivers a non-invasive in vivo technique to capture the connectivity of nerves themselves. Despite its promises and successful clinical applications for nearly thirty years, problems with validation and interpretation of measurements still persist. Most validation studies attempt to generate ground-truth data from animal models, phantoms, and computer models. This dissertation proposes a novel validation system, FiberBlender, capable of reproducing three-dimensional fiber structures and simulating the diffusion of water molecules to generate ground-truth synthetic DTI data. In particular FiberBlender contributes to: (i) creating more biologically accurate representations of fiber bundles with the inclusion of myelin and glial cells, (ii) examining the effect of demyelination and gliosis on DTI measurements, (iii) optimizing acquisition sequences, and (iv) evaluating the performance of multi-tensor models for the study of crossing fibers. FiberBlender strays away from the “one size fits all” approach taken by previous studies and uses computer algorithms in conjunction with some limited manual operations to produce brain-like geometries that take into account the random spatial location of axons and correct distributions of axon diameters, myelin to axon radius, and myelin to glia ratio. In this way no two models are the same and the system is capable of generating structures that can potentially represent any region of the brain and encompass the heterogeneity between human subjects. This feature is essential for optimization as the performance of DTI acquisition sequences may vary among subjects and the type of scanner used. In addition to better accuracy, the system offers a high degree of flexibility as the geometry can be modified to simulate events that cause drastic changes to the fiber structure. Specially, this dissertation looks at demyelination (an extensive loss of myelin volume), gliosis (a proliferation of glial cells), and axon compaction (a condensation of axons due to a loss of total brain volume) to determine their effects on the observed DTI signal. Simulation results confirm that axon compaction and partial remyelination have similar characteristics. Results also show that some standard clinically used acquisition sequences are incapable of capturing the effects of demyelination, gliosis and compaction when performing longitudinal studies. A novel sequence optimization technique based on Shannon entropy and mutual information is proposed to better capture demyelination. Optimized sequences are tested on a number of non-identical models to confirm their validity and can be used to improve the quality of DTI diagnostics. Finally this work looks at crossing fibers for the validation of multi-tensor models in their ability to characterize crossing diffusion profiles. The performance of multi-tensor models from CHARMED, Q-ball and spherical deconvolution that are widely used in both research and clinical settings are evaluated against ground-truth data generated with FiberBlender. The study is performed on a number of different crossing geometries and preliminary results show that the CHARMED model is the most comprehensive approach.
Dissertation or thesis originally submitted to the local University of Memphis Electronic Theses & dissertation (ETD) Repository.
Salan, Teddy, "FiberBlender: A Realistic Computer Model of Nerve Bundles for Simulating and Validating the Acquisition of Diffusion Tensor Imaging" (2017). Electronic Theses and Dissertations. 1784.