Piotr Majka¹ (p.majka@nencki.edu.pl), Li Zhaoping², Marcello Rosa³; ¹Laboratory of Neuroinformatics, Nencki Institute of Experimental Biology of the Polish Academy of Sciences, ²University of Tübingen, Max Planck Institute for Biological Cybernetics, Tübingen, Germany, ³Department of Physiology and Neuroscience Program, Biomedicine Discovery Institute, Monash University, Clayton, Australia

The central-peripheral dichotomy (CPD) theory is motivated by the fact that an attentional bottleneck allows only a tiny fraction of visual inputs to reach recognition. It proposes that the peripheral visual field primarily serves to select this fraction --- by directing gaze or attention --- whereas the central field is specialized for object recognition within the selected inputs. Because of this bottleneck, the feedforward information from the primary visual cortex (V1) to downstream visual areas is limited, constraining ongoing recognition. To compensate, downstream areas send feedback to upstream regions such as V1 to query for task-relevant information. The CPD theory therefore proposes that this feedback should target mainly the central visual field. Since the ventral stream is specialized for recognition, the theory predicts that ventral stream feedback to V1 is concentrated in the central field representation. We tested this prediction by injecting cellular-resolution retrograde anatomical tracers in locations representing various visual eccentricities (2˚- 18˚) in V1 of marmoset monkeys. Given V1’s position in the visual processing hierarchy, feedforward connections to V1 originate primarily (if not exclusively) in the lateral geniculate nucleus, whereas feedback connections arise from multiple cortical areas.  As predicted, the proportion of V1-projecting connections originating from feedback sources varied systematically with eccentricity. In the ventral stream as well as V2, this proportion was approximately ten times higher for the central field representation than for the peripheral representation. Moreover, with V2 excluded, ventral stream dominated feedback to the central visual field (<5˚), whereas dorsal stream dominated feedback to peripheral V1 (up to 18˚).   These findings suggest that the brain’s architecture for vision should be understood not only in terms of distinct cortical areas, but also in terms of the qualitative differences in computation applied to the central versus peripheral visual fields in each area.

Acknowledgements: National Science Centre grant (2019/35/D/NZ4/03031) to PM; European Commission EBRAINS 2.0 (101147319 FSTP) to PM; University of Tübingen and the Max Planck Society to LZ;