Linear models replicate the energy landscape and dynamics of resting-state brain activity

The spatiotemporal dynamics of resting-state brain activity can be characterized by switching between multiple brain states, and numerous techniques have been developed to extract such dynamic features from resting-state functional magnetic resonance imaging (fMRI) data. However, many of these techniques are based on momentary temporal correlation and co-activation patterns and merely reflect stationary, linear features of the data, suggesting that the non-stationary dynamic features extracted by these techniques may be misinterpreted. To examine whether such misinterpretations occur when using techniques that are not based on momentary temporal correlation or co-activation patterns, we addressed Energy Landscape Analysis (ELA), a statistical physics-inspired method that was designed to extract multiple brain states and dynamics of resting-state fMRI data. We found that the ELA-derived features were almost identical for real data and surrogate data suggesting that the features derived from ELA were accounted for by stationary and linear properties of the real data rather than non-stationary or non-linear properties. To confirm that surrogate data were distinct from the real data, we replicated a previous finding that some topological properties of resting-state fMRI data differed between the real and surrogate data. Overall, the present finding that the energy landscape and activity dynamics of resting-state fMRI data were well captured by stationary and linear surrogate data supports the notion that linear models sufficiently describe the dynamics of resting-state brain activity.