Abstract
Recent research on the ventral temporal cortex (VTC) has found clusters with differential preference to certain mid-level visual features such as rectilinear and curvature, which overlapped with the classical scene-preferring regions and face-preferring regions, suggesting a feature-based account for the seeming object domain-organization. This notion calls for a more comprehensive understanding of the visual feature topography in the VTC, considering the inter-correlation nature of different features, and its relation with the domain organization. We mapped out the sensitivity to 20 visual features (various visual shapes, color hues, and Fourier power features) across all VTC voxels using a parametric modulation paradigm. The fMRI responses of 95 object photographs were collected in 29 individuals. Photograph computational vision models were used to obtain the weighting of the 20 visual features. There were 3 main findings: 1) Association between visual features and domain-preference VTC clusters: The full parametric modulation model (all variables entered simultaneously) showed multiple significant visual features clusters overlapped with the three domain-preferring regions (PPA for large objects, latFG for animals, LOTC for tools) (see Figure 1). 2) Association between visual features’ VTC distribution pattern (beta values across the 3 domain-clusters) and the natural image statistics in the 3 object domains (weights in large image sets): Mixed linear model F=13.49, p < 0.001. 3) Visual feature effects independent of domains: Part of visual featural effects remained largely stable when the categorical structure was regressed out, and when presented in isolation as simple visual features. in experiment 2, the effects of the shape features (right angle, curvature) in the domain-preferring regions still remained whereas the effect of color hues disappeared. These findings depicted a comprehensive VTC visual feature topography map, which can be explained by their domain-associated natural image statistics, but with specific shape feature effects being domain-general.