Fig. 1

Identification of the most relevant seizure onset windows (SOW) and seizure onset zone (SOZ) detection is illustrated with one exemplary seizure. Signals are band-pass filtered in pre-defined bands of interest spanning the whole spectrum. For each recording site, a mean activation index (MA) is obtained over different time-frequency windows of interest using the Hilbert transform method and averaging the instantaneous power across time. For each window, the MA profile is characterized by two summary measures: the global activation (GA) and the activation entropy (AE), that quantify the magnitude of spectral changes with respect to the pre-ictal epoch and the spread of these spectral activations, respectively. SOW detection is then achieved by finding time-frequency windows that maximize GA under the constraint of low AE. For each SOW, most active regions are considered to be part of the SOZ