Abstracts

Rationale: The presence of high frequency oscillations (HFOs) in intracranial recordings at seizure onset and interictally is well-described (Jirsch, Worrell, Gotman), and has been shown to have a localizing value and perhaps a prognostic value for the surgery. In previous work, HFOs were found by painstaking visual analysis of filtered and magnified EEG records. We present a simple method of automated analysis of HFOs using continuous wavelet transform of multichannel EEG data. Wavelet analysis is well-suited to neurophysiological signals in general and to EEG in particular, due to nonstationarity and wideband nature of these complex signals. We propose that wavelet-derived HFOs are useful for seizure onset localization, and may be superior to the human eye. Methods: 29 patients with intractable epilepsy of frontal semiology and no radiological abnormalities underwent intracranial EEG monitoring. Multiple seizures were recorded for each subject. Seizure localization was performed by visual expert reading of the ictal EEG: 15/29 were deemed localized, and 14/29 nonlocalized. Continuous wavelet transform was computed retrospectively on multichannel referentially recorded EEG epochs of 5 seconds centered at the expert-defined electrographic seizure onset of localized subjects, or at the electrographic or clinical onset, whichever came first, of nonlocalized patients. Wavelet coefficients for 0.1-250 Hz were plotted and quantified for each channel. Interictal 5-second epochs were also analyzed. All findings were confirmed in bipolar montage recording. Results: HFOs at 125-250 Hz were present in all channels both interictally and periictally, however there was a significant difference of amplitudes, by an order of magnitude, for the HFOs in expert-defined “hot” channels at seizure onset as compared to the “cold” channels (Fig.1). This correlation of high-amplitude HFO activity to ictus in channels of visually detected onset was consistent for the 15 "localized" subjects. For activity under 125 Hz we have found a weaker correlation, and activity under 60 Hz appeared to have no correlation with the epileptogenic zone. For several nonlocalized patients, we have found strong evidence that HFOs were present focally at the clinical seizure onset, and this was missed by visual analysis (Fig.2). Conclusions: Intracranially recorded HFOs, as predicted from visual analysis, are indeed correlated with the recording electrode’s proximity to the epileptogenic focus as defined visually by an expert. However, multichannel analysis using wavelet transform suggests that the correlation is not as robust as expected: HFOs are present in all channels interictally and periictally, and are likely to represent normal neuronal activity or are an epiphenomenon of volume conduction. Only comparatively high-amplitude HFOs are correlated with seizure onset, and are useful for intracranial seizure localization. A sudden focal change in HFO amplitude appears to be a reliable marker of seizure onset, and may be used to define an epileptogenic focus where the golden standard of visual analysis had failed.