Accurate EEG-based linear state-space model of cortical dynamics during a passive motor task with exogenous input

OBJECTIVE: Revealing the dynamics of the brain in the presence of exogenous inputs is crucial for mechanistic understanding, interventions and regulations. We study state-space mathematical modeling of brain's cortical dynamics using electroencephalography (EEG) data in an experiment with continuous exogenous input. APPROACH: We use subspace identification techniques to model the brain dynamics in a passive motor task experiment. The EEG data were recorded while participants were seated with their wrist strapped to a haptic manipulator. The manipulator imposed a continuous multisine angular perturbation to the wrist as the exogenous input to the brain. MAIN RESULT: Using two subspace identification algorithms, we show that linear time-invariant state-space models accurately represent the measurements, in a latent space, assuming that the EEG data are the models' output. The model is verified and validated using data from seven participants. Moreover, we construct linear maps to relate the latent space dynamics to the neural source space. We verify that the findings of our model align with those identified in previous studies. SIGNIFICANCE: Accurate linear model in the presence of exogenous input provides a pathway for designing parsimonious and simple controllers. In addition, it enables development of online monitoring, connectivity analysis and control relevant to neurorehabilitation, motor learning, and brain-computer interfacing.