A bayesian model for Diffusion Tensorial Magnetic Resonance Images denosing and reconstruction is considered. This is based on a Tikhonov like-functional for Total Generalized Variation and Rician likelihood which is described in a variational framework. A primal-dual algorithm is implemented and accurate numerical solutions of the associated saddle-point formulation are computed. An automatic parameter selection rule is proposed to facilitate practical clinical usage and diagnostic of neurodegenerative disorders.
展开▼
