Abstract
Orientation is one of the most basic visual features encoded in the brain, and many neuroimaging studies have used decoding techniques to reveal how orientation information is represented in the brain. Previous MEG studies have successfully decoded orientation from event-related activity induced by oriented gratings (eg, Cichy et al., 2015; Pantazis et al., 2018). However, the gratings used in these studies are typically very large, extending from fovea to periphery in both visual fields. Further, subjects are generally only shown gratings with a small, fixed number of orientations. It remains unclear if the approach for decoding these large, fixed-orientation gratings can work for small peripheral Gabor patch targets, which are more realistic and psychophysically useful. In this study, we showed 21 subjects small, randomly-oriented Gabor patches at 7 degrees eccentricity in the right visual field, with MEG signals recorded concurrently. We also collected structural MRI data for each subject to augment MEG data with source localization analysis. We used a sliding logistic regression model to decode Gabor orientations at various timings relative to the stimulus onset. Our model achieved an average decoding accuracy that is modestly above chance, but revealed a peaking structure in decoding accuracy over time that is significantly different from chance, which was confirmed with a permutation test. From 0-125ms, decoding accuracy sat at the chance level. From 125-225ms, decoding accuracy consistently increased, then decreased back to chance accuracy by 350ms. This structure mirrors the structure of activations in the contralateral visual cortex after stimulus onset, which peaked around 150-225ms, on average. Our results suggest that increases in decoding accuracy correspond to increased activity in the visual cortex, and that MEG is viable for decoding small peripheral targets. This may improve our future ability to decode more complex peripheral visual stimuli from MEG data.