An offline evaluation of the autoregressive spectrum for electrocorticography.

Abstract

Electrical signals acquired from the cortical surface, or electrocorticography (ECoG), exhibit high spatial and temporal resolution and are valuable for mapping brain activity, detecting irregularities, and controlling a brain-computer interface. As with scalp-recorded EEG, much of the identified information content in ECoG is manifested as amplitude modulations of specific frequency bands. Autoregressive (AR) spectral estimation has proven successful for modeling the well-defined and comparatively limited EEG spectrum. However, because the ECoG spectrum is significantly more extensive with yet undefined dynamics, it cannot be assumed that the ECoG spectrum can be accurately estimated using the same AR model parameters that are valid for analogous EEG studies. This study provides an offline evaluation of AR modeling of ECoG signals for detecting tongue movements. The resulting model parameters can serve as a reference for related AR spectral analysis of ECoG signals.