Brain State-Space Model Parameters Estimation During Non-Invasive Stimulation

Abstract

Brain dynamic modeling is essential in understanding neural mechanisms and also developing neurotechnologies such as closed-loop brain stimulation systems that are used in a broad range of neurological disorders. In this paper, intending to model brain’s dynamic in the presence of non-invasive stimulation, we present a dynamic model of electroencephalography (EEG) activity under transcranial magnetic stimulation (TMS). In this regard, using collected data from the conducted TMS/EEG experiment and performing special preprocessing steps on that we build a multi-input multi-output (MIMO) linear state-space model (LSSM) for the temporal dynamics of EEG signal. To further investigate LSSM's performance, we also use a multilayer perceptron (MLP) model structure and evaluate its prediction ability. Results illustrate that despite the low signal-to-noise ratio in EEG signal, especially during the stimulation, LSSM as a general linear model performs well in predicting EEG dynamics. Also, choosing an MLP nonlinear structure with more complexity does not improve the prediction performance. The present study can be considered as a preliminary approach in modeling neural signals during stimulation. Expanding the proposed method in modeling other features of neural signals during the stimulation procedures can be a promising step toward designing closed-loop stimulation systems.