Reverse Radial Bias: Temporal Orientation Bias Compensation in Early Visual Areas Revealed by MEG

Abstract

Human visual system is organized sophisticatedly but it is also known that it has some response biases (e.g. color sensitivity, orientation selectivity) depending on locations of the visual field. Nevertheless, we can perceive an isotropic visual world without notifying any selectivity biases of the underlying neural responses. Therefore, a long-standing question is how our visual system compensates the visual feature selectivity anisotropies. Here, using human MEG measurements (Elekta Neuromag 360ch, 1000 Hz, N=12) and a machine learning classification (a linear SVM) technique, we explored a temporal compensation mechanism of orientation selectivity bias (aka, Radial Bias) in early visual areas. Specifically, we measured cortical responses for -45 and 45 deg gratings and visualized the dynamic changes of the orientation selectivity in each of retinotopic cortical positions during and after the stimulus on/offsets. The two orientations could be discriminated (90% accuracy) at 100 ms after the stimulus onset. Furthermore, the orientation selectivity in V1-V3, computed from SVM classifier weights, followed the presumed radial bias preference in each of retinotopic locations. However, interestingly, we found the reversal of the radial orientation bias 120 ms after the stimulus offsets, which may suggest a temporal neural compensation mechanism for the non-preferred tangential orientations. Follow-up analysis showed that the reversal was more likely to come from the temporal alteration of neural selectivity rather than the spatial response pattern shift. To explore a functional meaning of the reverse radial bias, we run additional psychophysics to measure orientation sensitivity thresholds in the reverse period and found that the orientation detection inefficiency corresponding to the radial bias was recovered exactly in the reversal period. Taken together, those findings indicate that our visual system compensates the neural selectivity bias temporally, as well as spatially, to capture transient changes of objects seen in the dynamic visual world. Meeting abstract presented at VSS 2018