Abstracts

Rationale: High frequency oscillations (HFOs, 80-500 Hz) provides crucial information in the delineation of the epileptogenic zone. Due to the short duration and low amplitude, visual inspection of HFOs in continuous intracranial EEG (iEEG) recordings is cumbersome. We applied a detection approach based on time-frequency analysis and clustering, that automatically identifies HFOs in iEEG, to facilitate localization of seizure onset zone (SOZ). Methods: Two pediatric patients with medically intractable focal epilepsy underwent intracranial electrode implantation for surgical evaluation. Fifty-six minutes of iEEG data was collected from 105 electrodes with 2 kHz sampling rate. Three HFO-distinguishing features were extracted from the time-frequency plane, including high-band to low-band power ratio, entropy, and frequency corresponding to maximum peak to notch ratio. These features were used for Gaussian Mixture Model clustering. Results: The detector isolated HFOs from other arbitrary waveforms. In patient 1, one cluster composed of faster oscillations (above 200 Hz) was originated from the first two contacts of superior and inferior central operculum, which precisely indicated the SOZ identified by the neurologists. In patient 2, two HFO clusters were identified with their energy below 200 Hz. One cluster in the right lateral parietal-occipital was consistent with SOZ. Conclusions: In these pediatric patients the seizure onset areas corresponded to the HFO generating regions supporting previous findings. Both patients were seizure free following the resection surgery. Our observations indicate an automatic detector with unsupervised clustering can be applied efficiently in quantitative HFO investigation, and can be used to assist in determining the seizure focus in pediatric epilepsy cases. Funding: NA