Abstract
Integrating image features across space is fundamental to visual function. Previous fMRI studies investigating retinotopy and related spatial properties have assumed implicitly or explicitly that responses to spatial contrast patterns sum linearly over the visual field. To test this assumption, we measured BOLD activity in human visual cortex while subjects viewed brief presentations of contrast patterns seen through horizontal and vertical apertures. A variety of different apertures were presented in random order in an event-related design. Throughout visual cortex the response to a large aperture was less than the sum of the responses to two parts of the aperture shown in separate trials. This sub-additive effect was larger in extrastriate areas than in V1. We modeled the sub-additivity by incorporating a compressive nonlinearity into a basic linear model of population receptive fields (Dumoulin and Wandell, 2008; Kay et al., 2008). Using cross-validation to obtain unbiased measures of model accuracy, we found that the nonlinear model systematically outperforms the linear model, explaining up to 98% of the variance in the amplitudes of the responses to the various apertures. Moreover, the nonlinearity of the model was more pronounced in extrastriate areas, consistent with the larger degree of sub-additivity in these areas. The nonlinear effect we have described can be viewed as spatial saturation in the sense that stimulating only a portion of a population receptive field is sufficient to saturate the response. Spatial saturation reduces sensitivity to changes in the position of a viewed object, and may be a fundamental computation that underlies the emergence of position tolerance in extrastriate areas.
NIH grant EY019244, NEI grant RO1-EY03164.