Over the last decades new technologies have shed the light on an improved understanding of brain structure and function. Recently, an imaging modality referred to as diffusion Magnetic Resonance Imaging (dMRI) has emerged. By exploiting the natural motion of water molecules undergoing a thermal agitation, i.e. Brownian motion, dMRI allows to estimate the biological tissue structure. The promising potentials of dMRI relies on the fact that it is the only imaging modality that maps the architecture of the cerebral white matter in-vivo and non-invasively. In the brain, experimental evidences suggest that the tissue component mainly responsible for diffusion anisotropy in white matter is the cell membrane. Consequently, the anisotropy of molecular diffusion in the white matter can be exploited to map the structural neuronal connectivity. The study of connectivity would allow to develop a clinical comprehension of the brain function. The precise mapping of the connectivity from dMRI involves an image processing step referred to as tractography. These algorithms produce trajectories capturing coherent orientations of maximal diffusion that are likely to represent real axonal fiber bundles. Tractography will play a major role in this thesis and we will focus on two points that needs to be taken into consideration when conducting whole brain connectivity analysis and comparing controls vs patients. First, the reproducibility of the measures describing the mutual connections between a pair of brain regions across healthy subjects and scans. Second, the accuracy and reliability of these measures. A multi-center study using healthy subjects and phantoms was performed to test whether dMRI data are reproducible across different MRI scanners and subjects. Measures describing the anisotropy were investigated both using region- and tract-based approaches. The main outcomes supports the feasibility of pooling dMRI data due to a reasonably low variability of these measures. This grants us now the possibility to study pathological cases using these measures and scanners. In this framework, we conducted a whole brain connectivity analysis comparing a group of healthy subjects vs a group of patients with epilepsy. In addition, more intrinsic micro-structural features were derived to further describe the connectivity and to better understand the changes in diffusion anisotropy. We observed that global connectivity, hub architecture and regional connectivity patterns were altered in epilepsy patients. Finally, to increase the sensitivity of dMRI study for other brain disorders, we developed a global tractography algorithm that reconstructs simultaneously the fibers of the entire brain by solving an energy minimization problem. Our approach was specifically designed for connectivity analysis applications, with the following main contributions: (i) explicitly enforces anatomical priors of the tracts in the optimization, (ii) considers the effective contribution of each of them to the acquired dMRI image. This algorithm was first tested on a realistic phantom and further applied to in-vivo human brain data. The new connectivity measures were in agreement with the ground truth. These results were improved compared to the current state-of-the-art tractography methods. We believe that the findings in this thesis will be of big value for the community performing connectivity analysis on the human brain data.