Decoding Rich Spatial Information with High Temporal Resolution

Abstract

New research suggests that magnetoencephalography (MEG) contains rich spatial information for decoding neural states. Even small differences in the angle of neighbouring dipoles generate subtle, but statistically separable field patterns. This implies MEG (and electroencephalography: EEG) is ideal for decoding neural states with high-temporal resolution in the human brain.

Keywords: Neural decoding; electroencephalography; magnetoencephalography; multivariate pattern analysis; orientation tuning; spatiotemporal information.

Figure 1

The Proposed Basis for Decoding Stimulus Orientations in EEG and MEG is Analogous to fMRI. (A) Orientation-selective neurons are clustered to form orientation-specific cortical columns in the visual system (orientation-preference coloured coded in the middle panel). Although fMRI cannot resolve activity from specific columns, each voxel samples an uneven distribution of columns resulting in weak but reliable voxel-wise preferences. Stimulus orientation can be decoded from the ensemble pattern of subtle preferences across a number of voxels (9 voxels schematised here; figure adapted from [10]). (B) A similar principle could explain orientation decoding with MEG/EEG. Variation in the angle of neighbouring dipoles could generate separable signals at the scalp surface. i. For example, three dipoles approximately 2 mm apart but with very different angles result in easily distinguishable MEG (upper row) and EEG (lower row) topographies. ii. Such differences will tend to average out with increasing numbers of dipoles (e.g., thirty dipoles, each tuned to one of three stimulus orientations, distributed along visual cortex generate very similar field topographies). However, just like decoding with fMRI, multivariate pattern analysis can differentiate stimulus-orientation by pooling orientation-specific information contained in the subtle biases for each sensor.