Abstract
Disentangling the neural computations performed through feedforward and feedback information flow in visual processing is challenging. Backward masking is an efficient experimental approach for this purpose, as it effectively interferes with reentrant feedback processing but leaves feedforward processing relatively intact. We used backward masking to dissect the spatiotemporal flow of visual information in the human brain. We briefly presented natural objects which were followed by a dynamic visual mask to participants. The mask could appear shortly after the object (16.7ms), resulting in low visibility, or with a substantial delay (600ms), resulting in high visibility of the object. We performed multivariate analysis on EEG (n=32) and fMRI (n=27) data to characterize temporal and spatial object representations in these two visibility conditions. While visual representations changed rapidly in time for both conditions, object information could be better decoded at later time points in the high than in the low visibility condition. To assess the temporal stability of the representation, we used time-generalization analysis. We observed less generalization and stronger pattern dynamics across time in the low compared to the high visibility condition. This suggests that backward masking fundamentally alters the visual recurrent processing in time. Furthermore, we combined fMRI and EEG responses using representational fusion to assess information flow in space and time simultaneously. While backward masking did not alter neural dynamics in the early visual cortex, we observed a decreased correspondence between EEG responses and the lateral occipital complex in later processing stages (i.e., from around 400ms on). This suggests a difference in recurrent processing at object-selective stages of the visual hierarchy due to efficient backward masking. Overall, our results characterize how feedforward and feedback information flow mediate visual object processing.