Estimating Electroencephalograph Network Parameters Using Mutual Information

AbstractStatistical parameters that measure strength, integration, and segregation of a multichannel electroencephalograph (EEG) network are evaluated using a similarity measure based on mutual information (MI) between the measured channel data. Compared with the unsigned linear correlation coefficient, MI is more robust to volume conduction and is applicable to nonlinear data. The statistical parameters estimated are node strength, average path length, and clustering coefficient. These parameters provide valuable insights into the brain network of the subject. MI is evaluated using a recently developed procedure based on the Gaussian copula. It is a computationally efficient procedure since estimation of MI is carried out analytically. This procedure is illustrated here for a 30-channel random noise and EEG network. The results are compared with those obtained using the linear correlation coefficient. The results show improvements by using MI to estimate the network properties.