Abstract
Sensory input and internal models combine to generate perception of the world. In vision, internal models can influence processing of feedforward sensory input in the primary visual cortex (V1) through cortical feedback. Whether such cortical feedback is retinotopically specific is still a matter of debate. Here we simultaneously recorded BOLD signal and eye-movements to study the spatial precision of cortical feedback in V1 during saccades. Subjects were shown images of natural scenes and instructed to execute a saccade across visual hemi-fields. During the saccade, the scene stimuli remained the same, changed, or disappeared. Retinotopic localizers were used to identify the processing region in V1 following the saccade. We trained support vector machines (SVM) on one-second time-windows at the post-saccade processing region to assess the extent of feedback related to the pre-saccadic scene. Integration of the relocated feedback and the post-saccadic feedforward signals was expected to affect SVM performance. After eye-movement, we observed lower SVM accuracy to scenes that changed across saccades in comparison to scenes that remained the same. These results suggest an interference of the feedback for the expected post-saccadic content with the processing of the newly presented scene. The decrease in SVM accuracy co-occurred with a univariate increase in BOLD activity at the post-saccadic region, indicative of a predictive coding error signal (Alink et al., 2010; Kok et al., 2012). Classification analysis did not reveal feedback to new retinotopic regions when the scene disappeared with the saccade. We suggest that a post-saccadic reference frame is necessary to support the remapped feedback in V1 across saccades. Our results demonstrate that with each saccade cortical feedback projects to the new relevant retinotopic regions to integrate the expected content with new sensory information. An interaction of predictive coding, saccadic remapping, and visual attention is likely to account for feedback relocation.
Meeting abstract presented at VSS 2015