Computational constraints underlying the functional domains in Macaque V4

V4, an intermediate area within the primate ventral visual pathway, comprises neurons specialized for diverse attributes, including color, shape, texture, and faces. Recent wide field imaging studies have revealed that macaque V4 neurons are spatially organized into cortical columns, each selectively tuned to distinct natural image features. These columns form functionally specialized domains arranged topologically across the cortical surface, spanning a spectrum from boundary-defined shapes (low dispersity) to broadly distributed surface features like texture and color (high dispersity). In this study, we applied the self organizing map algorithm to systematically explore the computational constraints underlying this complex functional architecture. Our results demonstrate that a balance between two primary constraints—feature tuning similarity and retinotopy—is both necessary and sufficient to reproduce the observed spatial arrangement of these domains. Specifically, we found that retinotopic constraints significantly fragment large clusters defined solely by feature similarity, resulting in an interleaved hypercolumn-like organization reminiscent of V1 blobs and interblobs. Additionally, we compared the biological V4 map with topological maps generated by topographic deep artificial neural networks (TDANNs). Although TDANNs replicated key aspects of cortical topography, they exhibited a pronounced bias towards texture features, differing notably from the balanced representation of shape and texture observed empirically in V4. Our findings underscore the critical role of combined feature continuity and retinotopic constraints in shaping cortical map topology, providing insights into the biological mechanisms underlying visual feature integration and offering directions for refining computational models to better align with cortical organization.