The use of diffeomorphism is now well established in computational anatomy. In this talk, based on the joint work with Marco Lorenzi, we will present the recent theoretical and practical advances which justify the use of stationary velocity field as a parameterization of a subset of diffeomorphisms and their transport into a common reference anatomy to perform statistics. We will show that this provides a very robust computational framework which can be used to model the evolution of the atrophy in Alzheimer's disease. Results show that the mean deformation of the brain is statistically different in AD, MCI and control patients and allows to discriminate as well as state of the art methods. Moreover, the normal trajectory can be used to decompose cross-sectional intersubject deformations into a potentially accelerated aging process and an orthogonal disease specific pattern.