Neuroanatomy-Based Strategies for the Statistical Analysis of Brain Imaging and Tractography Data

The analysis of brain images is one of the most common methods used in clinical neurosciences to investigate the structure, the function and the biological substrate associated to different brain pathologies. Most brain images consist of arrays of voxels that together provide a three-dimensional representation of the space occupied by a brain. Depending on the imaging modality, voxels are assigned with quantitative measurements that correspond to some biophysical phenomena detected at the different spatial locations specified by the voxels. The ability of the different statistical methods and image modalities to detect and characterise specific brain pathologies seems to be affected by the underlying brain anatomy. At the same time, image modalities like diffusion MRI tractography provide rich anatomical information that can be used to inform the statistical analysis of the images. In this thesis, I used in-vivo data and computational simulations to investigate the effect that the brain anatomy and the underlying microstructure has on the statistical properties of different diffusion MRI metrics. Subsequently, I developed a new framework for the representation and analysis of neuroimaging data that I named tract-based hypervoxels based on the anatomical and connectivity information provided by diffusion MRI tractography. To demonstrate the use of the new framework, I implemented the tract-based hypervoxel version of classical cluster-level inference analysis. I compared the new hypervoxel-based method with the voxel-based counterpart to detect group differences in the images from three different clinical studies ( Motor Neurone Disease, Autism and Huntington Disease) using two imaging modalities (diffusion MRI and PET). The results show the benefits of using the new hypervoxel framework in the analysis of neuroimaging data by increasing the sensitivity of the analysis and the anatomical specificity of the results.