Abstracts

Rationale: Identification of consistent distinguishing features between preictal and interictal periods in the EEG is an essential step towards performing seizure prediction. Recent studies suggest that high frequencies (80-500Hz) in intracranial EEGs play a role in epileptogenesis and could potentially underlie seizure precursors. In this study, a continuous wavelet transform approach is proposed to derive discriminating features between the preictal and the interictal state in intracranial EEGs from patients with mesial temporal lobe epilepsy. Methods: EEG signals (filtered at 500Hz and sampled at 2000Hz) from 6 randomly selected patients with mesial temporal lobe epilepsy, recorded from bilateral contacts in the hippocampus, amygdala and parahippocampus, were processed. 10 interictal segments of 1 hour and up to 6 preictal segments of 22 minutes were analyzed using a continuous wavelet transform in 4 frequency bands (50 to 450Hz). Wavelet entropy and energy were computed in the time-frequency domain for each segment. We defined a phase space of entropy and energy and hypothesized the existence of a reference state in this space defined in the 90s immediate preictal period. We characterized this reference state by a geometric region in this space. The parameters of this region were learned from training preictal datasets. We then compared the dynamics of the interictal and preictal distributions by computing their distance to the reference region and the inclusion rate, defined by the proportion of points in the phase space inside the reference region during a segment of 22 minutes. Results: When computing the average distance to a reference state defined simply by the mean point of the immediate preictal distributions, 4 patients showed a consistent discrimination between preictal and interictal segments on at least one EEG channel and one frequency band (figure 1). Distances computed from interictal segments were consistently higher than those from preictal segments. Inclusion rates were higher in preictal periods on channels where discrimination was observed with average preictal/interictal rates of 73.5/18.6 for patient 1, 85.6/57.2 for patient 2, 68.2/1.6 for patient 3 and 93.4/20.4 for patient 4. We thus confirmed the existence of a reference state characterizing the immediate preictal period. Random samples of 90s taken from interictal and preictal segments produced distributions of mean points substantially different (at 5% confidence level) from actual immediate preictal values. Conclusions: There exists a reference state behaving as an attractor in the immediate preictal period from which most seizures emanate. We proposed a discrimination technique between interictal and preictal periods based on the distance to this reference state. First results suggest that preictal periods are closer to the reference state than interictal periods in the entropy-energy phase space of a specific frequency band, opening the door to discriminating preictal from interictal periods. Supported by CIHR MOP-10189, RSC-NSERC CHRPJ 323490-06