Decoding peripheral saccade targets from foveal retinotopic cortex

Human vision is characterized by frequent eye movements ¹ . This causes continuous shifts in visual input, yet visual perception appears highly stable ^2–4 . A potential mechanism behind this stability is foveal prediction, involving feedback from higher cortical areas during saccade preparation. ^5,6 However, it remains unknown (1) whether information is fed back to early visual areas, (2) whether feedback is specific to stimulus features, and (3) which brain regions mediate this effect. To dissociate neural processes associated with stimulus presentation from those related to foveal feedback, we designed a gaze-contingent fMRI paradigm, where saccade targets are removed before they can be foveated. To determine the content of the neural representation, we used natural images as saccade targets and independently manipulated object shape and category. Multivariate analyzes revealed reliable decoding of stimuli from foveal retinotopic areas as early as V1, even though the stimulus never appeared in the fovea. Decoding was sensitive to shape but not semantic category, indicating that only low-to-mid-level information is fed back. Cross-decoding to a control condition with foveal stimulus presentation yielded reliable decoding, indicating a similar neural representation between foveal feedback and direct stimulation. Eccentricity-dependent analyzes showed a u-shaped decoding curve, confirming that these results are not explained by spillover of peripheral activity or large receptive fields. Moreover, fluctuations in foveal decodability correlated with activity in the intraparietal sulcus, a candidate region for driving this foveal feedback. These findings go beyond trans-saccadic remapping ^7,8 by suggesting that peripheral saccade targets are encoded in the foveal cortex in a feature-specific representation ⁵ .