Source-informed segmentation: Towards capturing the dynamics of brain functional networks through EEG

In this paper, we propose a data-driven framework to objectively evaluate the dynamics of brain functional connectivity, during task execution, from electroencephalography (EEG) recordings. The proposed framework consists of two main steps: first, EEG recordings, in the sensor space, are segmented into intervals during which the spatial distribution of functionally connected networks, in the source space, stays quasi-stationary. This “source-informed segmentation” is achieved through statistically assessing the temporal variation in the dominant left singular subspace of the EEG recordings matrix. Second, recordings within each identified segment are transformed into the source space, where networks of functionally connected cortical points are localized based on their correlation to a common dominant right singular vector of the source data matrix. The proposed framework is capable of identifying multiple networks of functionally connected cortical points, per segment, with each network corresponding to an orthogonal time-course of estimated neuronal activity. Theoretical derivations are presented. The proposed framework is also used to examine the dynamics of networks of functionally connected cortical points during the execution of a visual oddball task, and results are discussed.