Brain Dynamics During the Resting State

Functional MRI in the absence of tasks reveals dynamic large-scale configurations. Yet important gaps in our understanding remain: (1) What type of dynamics are observed during brain states? (2) What type of dynamics are observed between brain states? We address these questions by developing a multi-level Switching Linear Dynamical System (SLDS) model, which jointly estimates state-specific equations of motion and probabilistic state transitions. Within states, the inferred dynamics were stable attractors: model estimated fixed points closely matched observed fMRI activity. A central finding was that dynamic states did not map one-to-one onto canonical static large-scale networks; instead, the relationship was many-to-many, with each state engaging multiple networks and each network participating in several states. For example, a default-mode related state also engaged attention/control networks. Between states, transitions were structured (i.e., non-random) and heterogeneous. While some state transitions were relatively abrupt in terms of fMRI activity change (e.g., between a default-related network and an attention/control-related network), others were much smoother. Indeed, vector-field analyses in latent space quantified both toggle-like (abrupt) switches and smoother changes, revealing distinct transition pathways. To link systems-level organization to brain regions, we introduced region-level dynamics importance and state transition importance measures. Subcortical regions, particularly basal ganglia structures, dominated the dynamics of the highest-occupancy state, suggesting corticostriatal loops may scaffold a baseline regime from which excursions into other states arise. Importantly, regions most influential for withinstate dynamics sometimes differed from those driving state transitions, demonstrating that fMRI signal magnitude alone does not uniquely uncover how regions contribute to system-level properties. Our SLDS framework provides a principled bridge between discrete-state and continuous-trajectory perspectives of resting state dynamics, clarifying what the brain does while in a state and how it moves between states during rest.