Computational constraints underlying theemergence of functional domains in thetopological map of Macaque V4

Abstract

V4, an intermediate visual area in the ventral visual stream of primates, is known to contain neurons tuned to color, complex local patterns, shape, and texture. Neurons with similar visual attribute preferences are closely positioned on the cortical surface, forming a topological map. Recent studies based on multielectrode arrays and calcium imaging revealed the macaque V4 has neuronal columns tuned to specific natural image features, and these columns are clustered into various functional domains. There are domains tuned to attributes generally associated with object surface properties such as texture or color, as well as domains associated with the shape and form of object boundaries reminiscent of the blobs and inter-blobs in the primary visual cortex. Here, we explored the computational constraints underlying the development of the V4 topological map. We found that the map learned based on self-organizing principles constrained by neuronal columns tuning and retinotopy position can account for many characteristics of the observed V4 map, including the interwoven organization of texture and shape processing clusters. These anatomical clustering, with the implied local recurrent connectivity, might facilitate a modular parallel processing of surfaces and boundaries of objects along the ventral visual system.