Abstracts

Rationale: Responsive neurostimulation for epilepsy relies on early seizure detection. In the search for potential electrographic precursors of ictal events, we investigated the change in interelectrode phase coherence before and during ictal and interictal discharges. Methods: We analyzed awake and sleep intracranial recordings of patients with intractable epilepsy undergoing presurgical evaluation. From each recording, we created epileptiform (containing ictal and interictal discharges) and non-epileptiform (control) segments. Then we computed the Hilbert transform of the bandpass-filtered EEG signal at gamma (40-79 Hz), ripple (80-120 Hz) and conventional (1-25 Hz) frequency bands for each channel. Next, we computed the probability of pairwise change in phase coherence between each electrode and its neighbors. Next, we mapped the spatial distribution of these phase transitions in the electrode space relative to the brain. Finally, we computed the temporal distribution of local phase transitions relative to the time of ictal onset and computed the sensitivity of the detection at patient level. Results: There were 32 epileptiform and non-epileptiform segments from 8 patients available for analysis (9 awake seizure, 4 sleep seizure, 9 awake, 10 sleep segments). Our main findings were: 1) phases of gamma, ripple and theta oscillations change before epileptiform discharges at certain electrodes relative to the phases of the same oscillations at other electrodes (mean sensitivity=95%, mean specificity=69%); 2) although gamma, ripple and theta phase transitions are separable temporally and spatially, these transitions remain consistent within their respective frequency bands; 3) gamma phase transitions predict the impending epileptiform discharges more reliably than theta or ripple oscillations (p<0.05); 4) gamma and ripple phase transitions precede epileptiform discharges by 500-1500 ms (median 651 ms). Conclusions: We interpret these results in the context of macroscopic network dynamics where the interelectrode phase coherence is associated with functional connectivity. In this model, a subtle focal phase decoherence indicates a reconfiguration of the functional network. Our results demonstrate that default phase coherence is maintained over prolonged periods of time during resting state; when this state changes due to local phase perturbation (i.e., decoherence), it results in a rapid transition into an expanding hypersynchronous oscillation, which then terminates in a series of ictal discharges. Our results also show that these phase decoherences occur sufficiently early before the ictal onset such that they could serve as markers for triggering closed-loop neurostimulation for terminating seizures.