In this thesis, we develop some novel models and methods for the analysis of both structural and functional brain images, and their joint analysis with genetic data. In the first project, we present a suite of methods to increase the power of whole-brain genome-wide association studies. We introduce a kernel-based multilocus model to capture the interactions between single nucleotide polymorphisms (SNPs) and model their joint effect on the imaging traits. We provide a fast implementation of voxel- and cluster-wise inferences based on random field theory to take full use of the 3D spatial information in images and account for multiple comparison problems. We also propose a fast permutation procedure to increase the efficiency of standard permutation methods and provide accurate small p-value estimates based on parametric tail approximation. We explore the relationship between 448,294 SNPs and 18,043 genes in 31,662 voxels of the entire brain across 740 elderly subjects from the Alzheimer's Disease Neuroimaging Initiative (ADNI), and show boosted power of our approach by making head-to-head comparisons with previous voxel-wise genome-wide association studies. The various advantages of our methods over existing approaches indicate a great potential offered by this novel framework to detect genetic in uences on human brains. In the second project, we present a Bayesian spatial model of multiple sclerosis binary lesion maps based on a spatial generalized linear mixed model with spatially varying coefficients. Our model fully respects the binary nature of the data and the spatial structure of the lesion maps, as opposed to existing massive univariate methods, and produces regularized (smoothed) estimates of lesion incidence without an arbitrary smoothing parameter. Our model also allows for explicit modeling of the spatially varying effects of subject specific covariates such as age, gender, disease duration and disabilities scores, producing spatial maps of these effects and their significance, as well as the (scalar) effect of spatially varying covariates such as the fraction of white matter in each voxel. We apply our model to binary lesion maps derived from T2-weighted MRI images from 250 multiple sclerosis patients classified into five clinical subtypes, and determine the spatial dependence between lesion location and subject specific covariates. We also demonstrate dramatically improved predictive capabilities of our model over existing methods.
| Identifer | oai:union.ndltd.org:bl.uk/oai:ethos.bl.uk:589831 |
| Date | January 2013 |
| Creators | Ge, Tian |
| Publisher | University of Warwick |
| Source Sets | Ethos UK |
| Detected Language | English |
| Type | Electronic Thesis or Dissertation |
| Source | http://wrap.warwick.ac.uk/58416/ |
Page generated in 0.0021 seconds