Abstract
Introduction. Recently, intracranial EEG recordings in human and microelectrode recordings from animals have shown that sensory stimuli evoke a broadband elevation in the spectral power of field potentials. This broadband signal is of interest as it correlates with multiunit activity and the fMRI signal. Here we asked whether stimulus-related broadband responses can be reliably measured using MEG. Because extracranial measurements like MEG have multiple global noise sources and relatively low signal-to-noise ratio, we developed a denoising technique that helps reveal the broadband signal of interest. Methods. Subjects viewed 12-Hz contrast-reversing patterns presented either in the left, right, or full visual field, up to 11º eccentricity. Responses in MEG sensors were summarized as an evoked response (12-Hz amplitude) and broadband responses (mean of the log power between 60–150 Hz, excluding stimulus harmonics). A denoising algorithm developed for fMRI (‘GLMdenoise’; Kay et al. 2013) was adapted for MEG. The algorithm separated MEG sensors into those that were visually responsive and those were not (the ‘noise pool’) based on the evoked responses. Using PCA, the noise pool time-series were then summarized as several global noise regressors, which were projected out from the time series of all sensors. Finally, the broadband responses were re-computed from the denoised data. Results. In all subjects, broadband responses were reliably measured in sensors over occipital cortex. The spatial pattern of activation depended on the stimulus, with lateralized stimuli producing bigger responses in contralateral sensors. The signal-to-noise of the broadband responses in visual channels more than doubled as a result of denoising. Conclusions. Spatially localized broadband responses can be measured with MEG from human visual cortex. A new algorithm for denoising MEG data facilitates detection of this signal. Future applications of the denoising algorithm include other measures such as narrow-band oscillations and extension to EEG.
Meeting abstract presented at VSS 2015