Diffusion Weighted Imaging has massive potential to noninvasively detect explicit neurobiological properties, beyond what is possible with the resolution of conventional neuroimaging. However, there is very little known about what neurobiological properties these metrics, especially those derived from newer analytical approaches like NODDI, correspond to. While these diffusion metrics do not promise any inherent cell type specificity, different brain cells and even cell states have varying morphologies, which could influence the diffusion signal in many ways. This relationship is currently not well-characterized. Understanding the possible cytoarchitectural signatures of these measures would enable them to estimate different cell counts, potentially resulting in a very powerful clinical diagnostic tool. In this paper, using advanced diffusion imaging and a large cytoarchitectural atlas in the mouse brain, we demonstrate that different regions have unique relationships between cell counts and diffusion metrics. We then take advantage of this exclusivity, and introduce a framework to create region-specific models, which can be used to predict densities of different cell populations- including neurons and glia.