In this thesis algorithms are proposed for quantification of pathology in Diffusion Weighted Magnetic Resonance Imaging (DW-MRI) data. Functional evidence for brain diseases can be explained by specific structural loss in the white matter of the brain. That is, certain biomarkers may exist where the disease inhibits improper functioning. Axonal and myelin sheath damage hamper neural connectivity. This can be assessed in vivo by measuring a change in the diffusion of water molecules. DW-MRI may deliver such biomarkers by capturing subtle changes in the diffusion process in an early disease stage. Diffusion tensor derived scalar measures such as the Mean Diffusivity and Fractional Anisotropy (FA) quantify this process. When comparing such measures on group level, patients may be found to significantly differ from healthy controls. Conventional analysis treats measurements in each voxel independently. Due to high inter-brain connectivity, we hypothesize that multiple brain regions are involved in complex brain diseases such as schizophrenia. We introduce a new machine learning framework for performing such analysis in comparative studies on volumetric data. By ‘shaving' the mapping computed by discriminant analysis, a characteristic set of regions is automatically extracted. We opt to ‘prune' the dataset by iteratively discarding misclassified subjects from the cohort, such that the mapping is based merely on representative prototypes. Then a comparison is made between a linear and non-linear kernel discriminant analysis, to identify the dimensionality of the problem. Methods are proposed for modeling the 30% of white matter that cannot be adequately described by a single tensor. These regions are currently disregarded in comparitive studies. More complex diffusion models that are introduced need to be adequately evaluated. We propose a new method for creating experimental phantom data of fiber crossings, by mixing the DWI-signals from high FA-regions. These models demand HARDI (High Angular Resolution Diffusion Imaging) data acquired with higher SNR, diffusion weighting and angular resolution. In comparitive studies, scanning time may be insufficient for meeting the requirements. We propose a method to create a dual tensor atlas from multiple coregistered non-HARDI datasets. The random fluctuations in the pose of subjects in the scanner as well as anatomical heterogeneity contribute to an increased angular resolution. Finally, we build an optimization framework for estimating both diffusion shape and orientation in fiber crossings. This work sets fundamental limits for comparative studies to correctly analyze crossing white matter structures. Firstly, it assesses the precision and accuracy with which parameters may be estimated. Secondly, the optimal acquisition parameters are selected in order to do so. This model allows for estimating consistent FA-profiles along crossing tracts. Fiber tracts provide a specific frame of reference for computing statistics. We perform 3D tract-based comparison of tensor-derived indices between groups. Inter-subject correspondence is achieved by non-rigid registration based on a joint clustering and matching. The clustering delivers atlas points that serve as a frame of reference for performing the analysis. Spatial consistency is taken to reflect a significant difference between groups. The potential of our methods is demonstrated in two comparative studies: on Childhood Cancer survivors and Amyotrophic Lateral Sclerosis respectively. High throughput analysis of our data is realized by adopting a grid computing approach. The grid provides fast and easy access to shared resources. We present our progress over the past four years in the development and porting the DW-MRI analysis pipeline to grids. By doing do, our algorithms and results can also be accessed by fellow researchers.