The Construction of Large-Scale Cortical Networks for P300 from Scalp EEG
Fali Li,Chanlin Yi,Yuanling Jiang,Yuanyuan Liao,Yajing Si,Dezhong Yao,Yangsong Zhang,Peng Xu
DOI: https://doi.org/10.1109/access.2018.2879487
IF: 3.9
2018-01-01
IEEE Access
Abstract:Multiple studies, which studied cortical generators and cerebral networks of P300 based on electroencephalogram (EEG) and functional magnetic resonance imaging, have investigated the possible mechanism underlying the inter-subject variability of P300 component. However, few studies have investigated the large-scale networks of P300 as well as the potential relationships between large-scale networks and P300, particularly in EEG signal. When using group independent component analysis (gICA) to investigate the large-scale EEG networks, some resting-state networks, such as the default mode network, have been observed, while the related task-state networks (TSNs) of P300 are still unknown. In this paper, we estimated the time courses of cortical voxels following EEG source localization and then adopted gICA at the group level to identify the related P300-TSNs. Thereafter, we constructed and further investigated the potential relationships between the properties of functional network connectivity (FNC) and P300 amplitude. Our findings revealed six best-fit common functional networks that participated in the generation of P300, namely, the primary visual network, visual network, central executive network, left parietal network, attention network, and sensorimotor network. Specifically, the FNC properties significantly related to the P300 amplitude, as well as some FNC connections. In addition, when using FNC properties to classify the high- and low-amplitude groups, a relatively high-classification accuracy of 87.26% for linear discriminant analysis and of 86.31% for support vector machine were achieved. These findings confirmed the validation of gICA in mining large-scale networks in cortical EEG and may help us better understand the inter-subject P300 variability.