National audienceAnalysing functional Magnetic Resonance Imaging (fMRI) data is mainly done using the general linear model (GLM) in which the activation of a brain area is supposed to depend on all delivered stimuli (e.g. motor, visual, etc.) although activation is likely to be induced by only some of them in specific brain areas. Inclusion of irrelevant events may degrade the results, particularly when the Hemodynamic Response Function (HRF) is jointly estimated. In addition, a prior selection of relevant condition for each brain region is not always possible (e.g. pathology). To face this issue, we propose an efficient variational procedure that automatically selects the conditions according to the brain activity they elicit. It follows a...