One of the main challenges in identifying people at risk of dementia is their clinical heterogeneity. One hypothesis is that the clinical symptoms may be the result of different biological processes. We applied a data-drive clustering approach on structural and perfusion brain-MR imaging on a cohort of 141 MCI subjects in order to elucidate homogeneous structural and perfusion profiles and we observed the correspondent clinical features. Unsupervised clustering identified 6 different clusters on both ASL and gray matter volume data. Perfusion and atrophy showed to be variable in the different clusters and showed dissimilar patterns at subcortical and cortical levels.