Figure 1.

Clustering functional connections from dynamic epileptic networks. A, We identify ictal and interictal epochs from ECoG signals collected from patients with drug-resistant neocortical epilepsy implanted with intracranial electrodes. An ictal epoch is the period between seizure-onset, as characterized by the EEC (Litt et al., 2001), and seizure termination. An interictal epoch is defined to be a continuous, 5-min period at least 2 h preceding or following seizure onset. To measure time-varying functional networks, we divide each epoch into 1-s time windows and estimate functional connectivity in each time window. In our model, each electrode sensor is a network node, and the weighted functional connectivity between sensors, interpreted as degree of synchrony, is represented as a network connection. B, For each epoch, we estimated functional connectivity by applying a magnitude normalized cross-correlation between each pair of sensor time series in each time window. C, For time-varying functional connectivity, we extract all pairwise connections between nodes and concatenate them over time windows to generate a time-varying network configuration matrix. D, We apply NMF to the time-varying configuration matrix from each epoch, resulting in subgraphs that capture frequently repeating patterns of functional connections, and their expression over time.