Abstract Property-selective Clusters Link Continuous Property Representations to Discrete Category-selective Regions in Human Higher Visual Cortex.

Abstract

Object representations in the human higher visual cortex (HVC) support complex recognition behaviors early in development, yet the principles linking continuous dimensional representations and discrete category-selective areas in the HVC remain incompletely understood. Here, we propose a "property-cluster-category" organization that bridges object conceptual dimensions and discrete category areas in the HVC. Using a large-scale naturalistic stimulus data set and voxel-wise encoding methods, we analyzed the encoding patterns of a low-dimensional abstract property space and identified distinct brain clusters with shared cortical property profiles. These clusters broadly aligned with category-selective areas, suggesting that category regions can be understood as local peaks within a continuous property topology. We further tested whether this visual organization could emerge in a visual-only Topographic Deep Artificial Neural Network trained without semantic supervision. The model recapitulated property tuning for physical and biological dimensions but showed weaker affective tuning, suggesting that affective dimensions may require embodied or nonvisual experience. Finally, model-based lesion and stimulation analyses showed that Topographic Deep Artificial Neural Network units aligned with brain clusters contributed selectively to object classification. Together, these results provide a framework for understanding how the continuous property topology and discrete category selectivity are jointly organized in the human HVC, and suggest that abstract visual properties-which shape this topological organization-can be learned in part from statistical regularities in visual input.