Large-scale calcium imaging reveals a systematic V4 map for encoding natural scenes

Abstract

Biological visual systems have evolved to process natural scenes. A comprehensive understanding of visual cortical functions requires characterizing how natural stimuli are encoded by the massive population of neurons in each visual area. Here, we utilized widefield calcium imaging to sample primate V4 cortical response to tens of thousands of natural images. With this large dataset, we developed a deep-learning model of a substantial region of V4 that allowed us to map the natural image preference of neural populations at 100-µm scale. This map contains a variety of domains encoding distinct shape and surface features of natural scenes, including both previously reported attributes and previously missed ones like facial features or textures. We validated these model predictions with additional widefield calcium imaging and single-cell resolution two-photon imaging. Our study illuminates the V4 representation for natural scenes and provides an unbiased approach for characterizing the functional organization of primate visual cortex.