Authors :
Presenting Author: Su Liu, PhD – University of Miami
Naiara Garcia Losarcos, MD – University of Miami
Andres Kanner, MD – University of Miami
Rationale:
High-frequency oscillations (HFOs; >80 Hz) have emerged as promising biomarkers of the seizure onset zone (SOZ) in intracranial EEG (iEEG). However, translating this biomarker to non-invasive scalp EEG remains challenging due to its low signal-to-noise ratio and susceptibility to artifacts, particularly in adults. To address this, we adapted and applied an unsupervised HFO detection algorithm to scalp EEG data, aiming to evaluate whether it could automatically identify clinically relevant scalp HFOs above 60 Hz in adults with epilepsy.
Methods:
We applied the algorithm to three adult cohorts: (1) A publicly available dataset with simultaneous scalp EEG and iEEG, including known SOZ/resection zones (n=9); (2) A pilot clinical cohort undergoing simultaneous iEEG and scalp EEG for sleep monitoring (n=4); and (3) A separate group with pre-implant scalp EEG recordings prior to iEEG monitoring (n=5). The scalp EEG signals were filtered in the high-frequency range ( > 60 Hz). Feature extraction was performed using the short-time Fourier transform (STFT), Channels with temporally and spectrally distinct t-f components in the HFO range were identified. An unsupervised clustering approach distinguished putative neuronal HFOs from muscle and movement artifacts without requiring manual channel selection.
Results:
In both the public and clinical datasets, the algorithm successfully detected scalp HFOs that spatially aligned with clinically determined SOZ. In the first two cohorts with simultaneous intracranial recording, scalp HFOs also temporally corresponded with iEEG-detected HFOs within the SOZ. In seven cases, the algorithm generated separated HFO clusters, effectively distinguished SOZ-related HFOs from other physiological high-frequency activities and non-neuronal artifacts based on spectral and time-frequency features, despite variable noise levels.
Conclusions:
Our results demonstrate that scalp HFOs can be automatically detected in adult epilepsy patients using an adapted unsupervised algorithm. These HFOs show spatial concordance with epileptogenic zones and hold promise as non-invasive biomarkers to support seizure focus localization. This work supports the feasibility of using automated HFO detection in scalp EEG for potential future applications in epilepsy surgical planning and EEG-guided BCI systems.
Funding: None