Spatially dynamic nonlinear fMRI network analysis: application to schizophrenia

Motivation: Conventional fMRI functional connectivity studies have not been designed to capture nonlinear, spatially dynamic relationships which may be implicated in disorders such as schizophrenia. Goal(s): We address this gap in knowledge by estimating subject-specific, spatially dynamic nonlinear fMRI networks. Approach: We use a regression-based method along with independent component analysis to obtain explicitly nonlinear, spatially dynamic network profiles. Results: We find that explicitly nonlinear networks exhibit a higher degree of dynamicity than their linear counterparts, and that these dynamics reveal differences between individuals with schizophrenia and controls which are missed by linear functional connectivity analysis. Impact: Explicitly nonlinear fMRI networks reflect a higher degree of dynamic integration and segregation over time relative to linear networks. Moreover, the dynamic profiles of explicitly nonlinear networks reflect hidden differences between individuals with schizophrenia and controls.