AbstractDiffusion tensor imaging (DTI) is an advanced technique of magnetic resonance imaging (MRI) which is able to measure the diffusion of the water inside the brain tissues. Developing statistical methods for accurate grouping and modeling data objects with complex nature, such as the space of diffusion tensors, is needed to improve disease diagnosis and surgical planning. In this thesis, new statistical methodologies for DTI of human brain are developed.
The corpus callosum (CC) is a great fiber bundle in the white matter of the brain. Accurate segmentation of the CC is an important aspect of clinical medicine and is used in the diagnosis of various brain disorders. An accurate automated method for two and three-dimensional segmentation of the CC using DTI is developed. Hartigan’s K-means, an accurate K-means algorithm, is generalized for use with f-mean metrics (e.g. Cholesky, root Euclidean and log Euclidean). Then the generalized algorithm is used to provide a segmentation of the CC. The segmentation results using different metrics are evaluated to determine which metrics lead to the most accurate segmentations.
The von Mises-Fisher distribution (vmf) is a probability distribution for modeling directional data on the unit hypersphere. Multiple sclerosis (MS) is a neuroinflammatory disease that affects the brain and spinal cord and it is considered
the most common neurological disease that cause disabilities in young adults. We modeled the diffusion directions of the CC as a mixture of vmf distributions for MS and healthy subjects. Higher diffusion concentration around the mean directions and smaller sum of angles between the mean directions are observed on the normal-appearing CC of the MS as compared to the healthy subjects. An individual-based curvature threshold for stopping criteria of fiber tracking in the CC is proposed.
Fiber tracking is an important tool for visualizing white matter pathways and detecting brain abnormalities. Deterministic fiber tracking is highly affected by the noise in diffusion weighted imaging (DWI) and abnormalities in the brain which cause errors in the propagation of the track. The proposed stopping criteria, which is based on individual’s curvature thresholds, improves fiber tracking by terminating the fiber to prevent the deviation out of the original pathway. Quantitative measures (fractional anisotropy (FA), mean diffusivity (MD) and length of fibers) are compared in the healthy and MS subjects using three fiber tracking techniques (FACT, Bayesian and Wild Bootstrap). Significant differences in length of fibers in the healthy and the MS subjects are obtained using the three tracking methods.
Overall, the main contributions of the thesis is the development of new statistical methods for accurate clustering and modeling of data using DTI taking into account the complex nature of the data.
|Date of Award||2018|
|Supervisor||Andrew Fish (Supervisor), Roma Chakrabarti (Supervisor) & Diwei Zhou (Supervisor)|
Computational statistics for human brain diffusion tensor image analysis
Elshheikh, S. (Author). 2018
Student thesis: Doctoral Thesis