Detecting changes in dynamical structures in synchronous neural oscillations using probabilistic inference.

ABSTRACT

Recent neuroscience studies have suggested that cognitive functions and learning capacity are reflected in the time-evolving dynamics of brain networks. However, an efficient method to detect changes in dynamical brain structures using neural data has yet to be established. To address this issue, we developed a new model-based approach to detect change points in dynamical network structures by combining the model-based network estimation with a phase-coupled oscillator model and sequential Bayesian inference. By giving the model parameter as the prior distribution, applying Bayesian inference allows the extent of temporal changes in dynamic brain networks to be quantified by comparing the prior distribution with the posterior distribution using information theoretical criteria. For this, we used the Kullback-Leibler divergence as an index of such changes. To validate our method, we applied it to numerical data and electroencephalography data. As a result, we confirmed that the Kullback-Leibler divergence only increased when changes in dynamical network structures occurred. Our proposed method successfully estimated both directed network couplings and change points of dynamical structures in the numerical and electroencephalography data. These results suggest that our proposed method can reveal the neural basis of dynamic brain networks.