Behavioral and neural correlates of impaired scene perception following saccadic eye movements

The visual input projected to the retina shifts drastically across saccadic eye movements. Although we are not aware of it, perception of simple visual stimuli presented around the time of a saccade is impaired (Burr et al., 1994). Meanwhile, the extent to which the post-saccadic impairment influences high-level visual scene perception remains unclear. We conducted behavioral and fMRI experiments examining processing for scene images containing different spatial frequency content presented at different delays following a saccade. First, subjects performed a 6-way scene categorization task (beach, mountain, etc.) on images presented 5, 16, 50, 158, or 500 ms after saccade completion. We found lower scene categorization accuracy at 5 ms and 16 ms post-saccadic delays compared to longer delays, for both low- and high-spatial frequency filtered images, suggesting broadly impaired scene perception lasting less than 50 ms after a saccade offset. To further investigate what visual information is impaired, and in the absence of an explicit categorization task, we conducted an fMRI experiment where subjects performed a 1-back task on scene images while making saccades. Short and long post-saccadic delay trials were sorted post-hoc using eye-tracking data. Using RSA-based decoding analysis, we assessed scene category information (urban vs nature) in scene-selective brain areas, and low-level visual information (high vs low spatial frequency) in the early visual cortex. We found decreased scene category information in the posterior parahippocampal place area on short versus long post-saccadic delay trials, consistent with the behavioral impairment. Interestingly, lower-level visual information of a scene image was less impaired; spatial frequency information in the early visual cortex was not significantly different between short and long post-saccadic delay trials. Taken together, the current study presents novel evidence for impaired processing of complex scenes following saccades that may be driven by selectively interrupted neural representations of high-level scene content.