A DEFORMABLE ATTRACTOR MANIFOLD ORGANIZES HUMAN RESTING-STATE BRAIN DYNAMICS

This study aims to uncover the dynamical organization of the resting-state brain by characterizing how spontaneous fMRI activity self-organizes on a low-dimensional manifold. Using phase-based instantaneous functional connectivity from high-temporal-resolution resting-state fMRI data (HCP and ADNI), we identify two antagonistic dynamical regimes: a low-coherence state (LCS) marked by stable, incoherent activity, and a transient high-coherence state (HCS) characterized by globally coordinated dynamics. A two-state hidden Markov model segregated these regimes, revealing that HCS occupies a significantly lower-dimensional space yet accounts for greater cognitive fluidity and a richer repertoire of co-activation patterns. The manifold structure of each state and resting-state network was quantified via stationary probability densities and interpreted through the stationary solution of the Fokker–Planck equation, yielding effective potential functions. Remarkably, a quartic normal-form potential with only two parameters—criticality and bias— captured network-specific dynamics. Results show that LCS is strongly monostable across networks, whereas HCS spans bistable to near-critical regimes along the functional gradient, with core networks—especially the default mode network—exhibiting the slowest timescales and realizing a hierarchical self-organization of HCS. Importantly, only HCS-specific criticality parameters differentiated controls from individuals with mild cognitive impairment, highlighting the potential of this biomarker. State transitions occurred mainly through the deformation of the underlying attractor manifold. Rare direct inter-modal HCS jumps also occurred; notably, half of them involved previously reported high-amplitude cofluctuations. Overall, these findings identify an organizing geometric and dynamical principle of resting activity, linking large-scale cortical coordination, cognitive variability, and vulnerability to pathology.