Identification of epileptic high frequency oscillations in the time domain by using MEG beamformer-based virtual sensors

Abstract

Objective: High frequency oscillations (HFOs, >80. Hz) are biomarkers for epileptogenic cortex in invasive and non-invasive electroencephalography (EEG). Identification of HFOs in magnetoencephalography (MEG) is hindered by noise. Computing spatial filters using beamforming to reconstruct time series for selected brain regions, so-called virtual sensors (VS), can increase the signal-to-noise ratio. We identified HFOs in MEG in time domain using VS. Methods: Fifteen minutes of MEG data were selected from 12 patients. VS were placed around the epileptic spikes (affected region) and in the contralateral hemisphere. VS and physical sensors were reviewed for HFOs and spikes. HFO locations were compared to spikes and other clinical parameters. Results: Eight patients showed 78 time points with 575 HFOs in VS, 513 were in the affected region. HFOs could not be identified in physical sensors for 61 of the 78 VS time points. HFOs overlapped with presumed epileptogenic areas and were also visible in unfiltered VS signals. Conclusion: Beamformer-based VS analysis can help to identify epileptic HFOs that are not discernable in physical MEG sensors. Significance: This approach can be extended to enable localization of non-invasively recorded HFOs. This would help surgical planning and reduce the need for invasive diagnostics.