Epilepsy is fundamentally a disease that affects neuronal networks. Focal seizures, in particular, may arise from the abnormal dynamics of networks within the epileptogenic zone, as well as abnormal network dynamics outside this zone that facilitate seizure propagation to other brain regions. Structural connections are assumed to play a permissive role in both normal and abnormal dynamics, while physiological interactions among neurons and their networks play an active role in shaping network dynamics. These physiological interactions can be conceptually divided into functional connectivity and effective connectivity, both of which are potentially important to understanding the normal and abnormal dynamics of human cortical networks.
Functional connectivity is usually measured with band-limited phase-locking values or coherence. Effective connectivity has been measured in model-free systems (e.g. EEG/MEG, ECoG) with a variety of methods, including transfer entropy, permutation conditional mutual information, and those based on causality in the Granger sense, where one can consider an observed time series x(t) to have a causal effect on another time series y(t), if knowledge of x(t)'s past significantly improves prediction of y(t) (Granger, 1969). Using the concept of Granger causality, we have created an event-related causality measure (ERC) for our ECoG studies of human cortical networks and their dynamics during cognitive tasks and during seizures. In our studies of the network dynamics of spoken word production, for example, we have reasoned that neural activity at any given processing stage is more likely critical to task performance if it has a causal impact on activity at downstream processing stages, ultimately culminating in response (e.g. articulatory) processing. In our studies of epileptic networks, on the other hand, we have reasoned that the dynamics by which ictal activity propagates will reveal the effective connectivity of epileptic brain networks and identify nodes in these networks that are relevant for surgical planning. In both cases, we have focused on high frequency activity because of its established correlation with population firing rates, as well as the preponderance of high frequency oscillations (HFOs) in the ictal onset zone. In reviewing the results of these studies, we explore the possibilities of discriminating epileptic networks from the networks that support normal cognition.