Research On Brain White Matter Axonal Fiber Shapes Via MRI Data

Being a center of neuroal system, brain has the most complicated structure andfunction. The axonal fibers in the white matters are in charge of bio-signal delivery andexchange among neurons located in the gray matters, and thus are critical elements ofbrain functional network. Also, in the process of brain development and along the courseof evolution, the white matter fibers are positively correlated to the formation of cerebralcortex folding patterns. These two white-matter-centered topics are extensively studiedand discussed in this dissertation by mainly using multi-modal MRI data, including T1-weighted MRI, dMRI and fMRI. The related results are summarized as follows:To study the coorelation of cerebral cortex folding patterns with axonal fibershape patterns, we firstly developed two parameterization methods:Cerebral cortex folding patterns are parameterized. A parameterizationmethod is proposed to describe cerebral cortex folding patterns. This methodis developed by fitting a simple polynomial function to the vertices within aregional surface patches under a local coordinate system. The advantage ofusing local coordinate system is that it can transform a variety of regionalsurface patches of differentfoldingpatternsintoa uniformcoordinate system,so that they and their parameterized representation can be compared, even ifthe surface are reconstructed from database with very different imagingprotocols. Polynomial is adopted parameterize surface patches so that theresulted representation is compact, so that it will be effective to conductanalysis, such as clustering parameters of patches obtained from database oflarge size. Via a model-driven method, surface patches are classified intoeight primary folding patterns.White matter axonal fiber shapes are parameterized. A parameterizationmethod is proposed to describe axonal fiber shape and connectional patterns.By taking each single fiber tract reconstructed from dMRI as an individualidentity, four categories of descriptive features are extracted from it after itis normalized. This method can be applied indepently on dMRI databases ofdifferent imaging parameters. The parameterized representation can be usedto conduct analysis across databases such as shape clustering. Four primaryaxonal fiber shapes are identified via clustering method. Base on the previous two methods, we jointly analyze the co-localization andcorrelation between cerebral cortex folding patterns and white matteraxonal fiber shapes. Based on the previous two parameterization methods andother related methods, joint analysis is performed on three primate species(macaque, chimpanzee and human) and it is found that axonal fiber shapepatterns and cerebral cortex folding patterns are positively correlated with eachother. It is found that orientaions of axonal fiber terminations closely follow thegyrus orientation on the circumferential directions. Abundant U-shape axonalfibers can be found beneath all major sulci and they connect neighboring gyralcortial regions. Also, cortical regions connected by U-shape fibers are morelikely to be have more complicated folding patterns. Generally, those results, thepositive correlation between cortial folding patterns and axonal fiber shapes, canbe preferably reproduced on the three primate species, but different correlationstrength, however, can be found across species as well. The difference in acertain extent may help to explain the role axonal fiber plays along the line ofevolution.To study the correlation between white matter axonal fibers and brainfunction. The close relationship between axonal fibers derived brain structureand brain function has been widely reported in the literatures. Based on thiswidely accepted concept, a novel method is proposed to use axonal fibers shapepatternsto predictbrainfunctionalregion locations. The frameworkisconductedunder a machine learning scheme. After identifying the corresponding brainfunctional regions across training subjects, fiber shape models are learnt fromthose emanating from those regions. Those models are then applied on testingsubjects, so that the corresponding functional regions can be predicted. Theframework does not only show further supports to the close relationship betweenbrain structure and function, but also show promise in a variety of applicationscenrios. That is, only brain structure data of small cost is needed to generatepreferably accurate brain functional regions, which can only be obtained viaconventional method from functional data, the cost of which is dramaticallyhigher than sturture data.The framework is extended to incorporate cerebral cortical folding informationinto the functional regionlocation prediction. Informationfromaxonalfibersandcortical cortex, complementing each other, is fused via a novel framework to further improve the brain functional region location prediction accuracy.