Biomechanical brain models offer a physics-based means of simulating tissue deformation and stress under physiological and pathological conditions. However, most existing models still rely on coarse, region-wise material properties derived from invasive ex-vivo experiments. Such simplifications neglect the spatial heterogeneity of brain tissue and may limit the fidelity of individualized simulations. In this work, our objective is to advance subject-specific brain biomechanics by incorporating personalized, spatially varying material parameters derived from in-vivo imaging. We develop a three-dimensional finite element model of the human brain and compare two material parameterizations implemented on the same computational mesh. In the first, homogeneous constitutive properties are assigned to nine anatomical regions using published ex-vivo mechanical data. In the second case, the properties of the material vary at the voxel level and are derived from a correlation between the fractional anisotropy (FA) based on diffusion tensor imaging and the regional stiffness of the tissue. A linear relationship between FA and shear modulus is established from region-averaged data and subsequently applied voxel-wise to obtain a smoothly varying stiffness distribution that reflects local microstructural variability. Both parameterizations are subjected to an identical atrophy-driven loading scenario. On a global scale, the models predict similar overall loss of brain volume, suggesting that regional parameterizations may be sufficient to capture global deformation trends. However, on the local scale, clear differences emerge. Using voxel-wise parameters leads to altered displacement and strain patterns and predicts distinct regional responses, particularly in periventricular areas, indicating a pronounced sensitivity of local mechanics to spatial variations in stiffness. This study shows that incorporating diffusion-derived microstructural information into biomechanical brain models can substantially influence local predictions while preserving global behavior. The proposed framework provides a non-invasive method toward more personalized brain mechanics and represents a step toward patient-specific digital twins in neuromedicine.