Linear models replicate the energy landscape and dynamics of resting-state brain activity
The brain is constantly and spontaneously active even in the absence of sensory stimulation or motor tasks. The spatiotemporal dynamics of this resting-state brain activity are commonly assumed to be non-stationary and non-linear. It has been proposed that dynamic features (e.g., brain states) extracted from measured resting-state brain activity can better explain subject-specific phenotypes (e.g., fluid intelligence score) than static features. However, several recent studies reported that some analysis methods used to extract dynamic features from resting-brain activity as measured by functional magnetic resonance imaging (fMRI) do not truly reflect non-stationarity and/or non-linear statistical properties of the data. In this study, we examined energy landscape analysis (ELA) to assess whether dynamic features extracted by ELA accurately represent the non-stationary and/or non-linear aspects of resting-brain fMRI activity. To specify the statistical properties being extracted, we applied ELA to both real fMRI data and surrogate data generated by linear and stationary models. We found that ELA results obtained with both the real and simulated data were almost indistinguishable, suggesting that the features extracted by ELA can be explained by covariance and autocorrelation structures. These results corroborate a recent proposal that resting-state fMRI activity can be adequately described by linear models.