Abstracts

Rationale: : Improvements of real-time seizure detectors and of algorithms of focus localization involve two very different strategies. However the possibility of a real-time detector of epileptic seizures during the presurgical evaluation of patients in monitoring unit would benefit from increased efficiency in both these paradigms. We investigate here statistics which can optimize either or both goals. Methods: A selection of methods assessing the complexity, entropy and spectral features of a signal were selected to investigate their respective efficiency to quantify the onset of partial seizures. These measures include Shannon entropy (ShEn), spectral entropy (SpEn), sample entropy (SampEn), permutation entropy (PermH), Hjorth complexity, Lamberti complexity (CJS), LZC algorithmic complexity (Lempel-Ziv), signal structure complexity (GAD), Higuchi fractal dimension (HFD), total signal power (P), spectral edge frequency (Fe) and the power in spectral bands. Data include 331 partial onset seizures recorded from 45 consecutive patients with either mesial temporal (MT), neocortical temporal (NeoT) or extra-temporal (NeoXT) onset seizures. Normalized changes relative to an interictal baseline level were calculated and preictal-ictal contrasts were used to quantify the amplitude of the change occurring in the early phase of the seizure. Relative efficiency was estimated for MT and NeoXT groups. Detection power was assessed by the average increase at onset of all partials seizures. Discriminative power was assessed by the contrasts between the increases averaged over each region.Results: Spectral power bands present a continuum of relative efficiency from low detection power/low discriminative power for ? to high detection power/low discriminative power for high-frequency bands (up to 300Hz). Intermediate bands such as ?, ? and ? present the greater detection power of the frequency bands with moderate ability to discriminate mesial temporal region of onset. For the entropies, ShEn, SpEn and SampEn are unable to differentiate MT and NeoXT onset seizures. PermH however present ictal decreases specific to MT onset seizures. For the complexities, GAD, LZC and CJS exhibit moderate detection power but no discriminative power. HFD has moderate detection power and moderate discriminative power. Hjorth shows low detection power and low discriminative power.Conclusions: Measures with responses specific to seizures patterns originating from certain regions of onset (eg. HFD) and measures which provide unequivocal indication of seizure onset regardless of the region (eg. power in ? frequency range) would be ideal candidates to combine for a real-time detector usable in the context of presurgical evaluation for the detection and localization of partial seizures.