Subject- and behavior-specific signatures extracted from fMRI data using whole-brain effective connectivity

ABSTRACT The idea of fMRI-based personalized medicine is emerging to characterize brain disorders at the patient level and to develop tailored therapeutic protocols. A main limitation in this direction, as well as with fundamental studies of brain function, is the variability observed in fMRI data. In practice, such measurements exhibit a mixture of session-to-session variability with subject- and condition-related information. Disentangling these contributions is crucial for real-life applications where only a few sessions per subject are available. The present study aims to define a new reliable standard for the extraction of signatures from fMRI data. These signatures correspond to subnetworks of directed interactions between brain regions (typically 100 covering the whole brain) supporting the subject and condition identification for single fMRI sessions. The key is to step from functional connectivity to effective connectivity, which is estimated using a whole-brain dynamic model. Our method shows that as few as 4 sessions per subject are sufficient to perfectly identify more than 40 other sessions of 6 subjects. We also demonstrate the good generalization capability for 30 subjects. Using another dataset with resting state and movie viewing, we show that the two extracted signatures correspond to distinct subnetworks, suggesting some sort of orthogonality. Our results set solid foundations to follow longitudinally a subject’s condition from fMRI data.