Eigenvector centrality reveals the time course of task-specific electrode connectivity in human ECoG

Abstract

Connectivity measures provide a quantification of information flow across electrodes in human subject electrocorticography (ECoG). They do not, however, lend themselves to direct interpretation due to the combinatorial size increase of the feature space. We utilize time-varying dynamic Bayesian networks (TV-DBN) as a model of the individual ECoG electrode activity based on the activation of the electrode array. Using the high gamma power TV-DBN connectivity matrices, we determine if eigenvector centrality can objectively highlight the important interactions between electrodes. The statistically thresholded centrality measure reveals task-related differences in the significant electrode subsets during distinct task phases (p<;0.05; 13 significant electrodes overall: 2 exclusive to the cue processing phase, 3 exclusive to the motor output phase). These results suggest that TV-DBN and centrality analysis can be used in an online brain-mapping system to show regions of the brain relevant to real-time task performance.