Modelling the spatio-temporal dynamics of functional connectivity in resting-state fMRI. – DynaSTI

Functional Magnetic Resonance Imaging (fMRI) is the dedicated modality for studying functional brain connectivity (FC). The analysis of fMRI data (3D + time) allows the identification of brain regions whose temporal activity, also called timecourse, is strongly correlated: these regions form what are called functional networks. In this project, we propose to develop a model of spatio-temporal dynamics of FC for the longitudinal detection of neurodegenerative or neuropsychiatric pathologies in resting-state fMRI data at the single subject level. The objective is to define a new framework from the fMRI data of a subject to the modelling of the spatial and temporal dynamics. This will make possible to highlight the presence (or absence) of spatio-temporal patterns in functional connectivity at rest and to compare these patterns at different stages of the pathology in order to identify markers of the pathology's development. Our preliminary work [Bhanot2019, Bhanot2021] has allowed us to verify that it is possible to work with fMRI data at the scale of a very fine anatomical segmentation: for example, for the mouse brain we have an atlas of the brain divided into approximately 600 anatomical regions. We therefore propose to work with connectivity matrices calculated at a much finer spatial scale than functional networks in order to define what we will call functional connectivity units (FCUs). The FCUs will allow the introduction of spatial dynamics by connecting and disconnecting during the whole fMRI scan. The objective is to model the list of connectivity matrices by a sparse decomposition on the dictionary of FCUs.