Abstract
This work proposes $\mu$GUIDE: a general Bayesian framework to estimateposterior distributions of tissue microstructure parameters from any givenbiophysical model or MRI signal representation, with exemplar demonstration indiffusion-weighted MRI. Harnessing a new deep learning architecture forautomatic signal feature selection combined with simulation-based inference andefficient sampling of the posterior distributions, $\mu$GUIDE bypasses the highcomputational and time cost of conventional Bayesian approaches and does notrely on acquisition constraints to define model-specific summary statistics.The obtained posterior distributions allow to highlight degeneracies present inthe model definition and quantify the uncertainty and ambiguity of theestimated parameters.