"Spilling-in" versus "Filling-in": disentangling bottom-up drive from predictive completion in human V1-V3

It is well established that responses in visual areas are a combination of bottom-up stimulus driven input and 'top-down' input driven by expectation. However, it has been difficult to quantify the differential strength of these components, and how they vary across the visual hierarchy. Here, we introduce a method for dissociating 'bottom-up' and 'top-down' contributions in early retinotopic areas (V1-V3). Using a log-scaled bar stimulus, we measured BOLD responses under full-field vs. central scotoma conditions. Using the full-field stimulus, we estimated pRFs in human visual cortex (V1-V3). We then used each voxel's pRF to predict the response expected from bottom-up responses alone, and compared this prediction to measured responses. Responses within the scotoma demonstrated measurable predictive filling-in that increased systematically from V1 to V3. Our results demonstrate that activity within retinotopic visual cortex cannot be explained solely by feedforward processing and provide a principled method for measuring the contributions of predictive signals in early visual cortex. Finally, we show that top-down filling-in can masquerade as cortical reorganization in which filling-in signals are accounted for through pRF displacements and size increases. Our results show that estimating top-down predictive components is critical for accurately describing cortical responses near regions of visual loss.