Constructing EEG Large-Scale Cortical Functional Network Connectivity Based on Brain Atlas by S Estimator

Large-scale functional network connectivity (FNC) reveals the neural substrate of the cognitive process on a large-scale level. Electroencephalogram (EEG) is cost effective, portable, and noninvasive and is capable of capturing brain activities at a millisecond scale. Brain atlas derived from the anatomy clearly defines reliable functional subnetworks. In this article, we proposed to construct robust EEG FNC based on brain atlas by combining EEG source imaging with multivariate synchronization analysis to mine the brain’s large-scale information exchange. We evaluated the performances of two typical methods, canonical correlation analysis (CCA) and ${S}$ estimator, in quantifying the couplings among subnetworks by both simulation and application to real EEG data set. Simulation demonstrated that, compared to CCA, ${S}$ estimator shows high robustness and adaptability to low signal-to-noise ratio (SNR) and short length data. According to the FNC of P300, we further found that the FNC network constructed by ${S}$ estimator may be more consistent with the physiological mechanisms of P300 generation, where the ${S}$ estimator-based approach emphasizes the important role of the cerebellar network that has been proved to be involved in attention-related cognition tasks. This article provides a new tool to probe information processing during the cognition process at a higher hierarchal level.

I. Introduction

The activities in human brain in the processes of cognitive behavior, even for such a simple behavior as blink or sleep, are ever changing and extremely complex. With the development of cognitive neuroscience, accumulative studies confirm that brain regions are not independent processors for specific cognitive functions [1]–[4], whereas the conjoint and the dynamic function of spatially remote brain areas work together as large-scale functional networks, such as the visual, auditory, and attention network [3], [5], also contribute to the information processing for cognitive processes. Besides the strong coupling within a large-scale functional network, recently, an increasing number of studies emphasize that the temporal dependencies among large-scale functional networks are also significant [6], [7]. By setting large-scale functional networks as nodes and the interactions among them as the edges, functional network connectivity (FNC) further quantitatively assesses the interactions (temporal dependencies) among large-scale networks, which reveals the neural substrate of brain activities in a large-scale level. For example, employing large-scale FNC, functional magnetic resonance imaging (fMRI) studies revealed the FNC changes are crucial for studying motor imagery (MI) and schizophrenia [7], [8]. To facilitate understanding, the node (large-scale functional network) in FNC is hereafter referred to as “large-scale functional subnetwork,” “functional subnetwork,” or “subnetwork.” Nevertheless, as shown above, since the high spatial resolution is required, most studies are conducted with fMRI that is expensive and definitely weak in grasping the ever-changing brain activities and probing the adaptive adjustment of the brain in the high-level cognition processes.