Information flow in EEG source networks in epileptic children with focal seizure activity

Scalp electroencephalographic (EEG) signals are influenced by several factors, including volume conduction and low spatial resolution, which can jeopardize the validity of brain connectivity analysis performed on the raw recordings. One possible solution is to identify, starting from scalp EEG signals, the underlying cortical source activations, and to apply connectivity metrics on the reconstructed source time series. In this work, the dynamics of information flow between cortical EEG signals obtained after source reconstruction were assessed in children suffering from focal epilepsy. In a group of 10 children with focal seizures, 5-second windows of the 19-channel EEG were obtained in the baseline, pre-ictal, and post-ictal phases. After filtering and artifact removal, 19 baseline, 19 pre-ictal, and 12 post-ictal stationary trials were selected for the analysis. Source reconstruction was performed combining a common spatial pattern algorithm with linear modeling and Indiapendent component analysis. Finally, linear measures of functional connectivity (information storage, total and conditional information transfer) were obtained from vector autoregressive models of the source signals. While the average information stored in the nodes of the source EEG network did not change significantly across conditions, the total information transferred to each node increased significantly just before the seizure onset (p=0.001) and remained high after the seizure (p=0.009). The number of directed links in the network (statistically significant values of the conditional information transfer) also increased comparing the pre-ictal and post-ictal phases with the baseline period (p=0.134, p=0.109). These results indicate that a reorganization of the source EEG network, characterized by dense topology and increased information transfer, occurs before the onset of focal seizures, which is promising for seizure prediction algorithms