Multi-session extension of the joint-detection framework in fMRI

Modern cognitive experiments in functional Magnetic Resonance Imaging (fMRI) involve the repetition of the same paradigm over several short sessions (or runs) since long fMRI acquisitions usually place the subject in an uncomfortable situation and generate motion artifacts. Also, shorter sessions enable to better control the subject's cognitive state and guarantee his attention during task. The Joint Detection-Estimation (JDE) framework which aims at detecting evoked activity and estimating hemodynamic responses jointly, has been developed so far to treat each session independently and then build average contrasts of interest as already done in other packages (SPM, FSL). Here, we extend JDE to the multi-session context by proposing a new hierarchical Bayesian modeling including an additional layer to describe the link between session-specific and mean evoked activity. In contrast, the HRF shape to be estimated in each region is assumed constant across sessions. Our results on simulated and real multi-session datasets show that the proposed extension outperforms its ancestor both in terms of activated areas and HRF recovery.