Heterogeneous Orientation Tuning Across Sub-Regions of Receptive Fields of V1 Neurons in Mice

Abstract

A key feature of the responses of V1 neurons is their selectivity for stimulus orientation. Recent work using deep neural predictive models found that the most exciting images (MEIs) for neurons in mouse V1 can exhibit complex spatial structures deviating from Gabor-like filters. This finding suggests that orientation selectivity may not be a stable property across the receptive field. Here, we characterized the spatial profile of orientation tuning and found that it varied up to 90 degrees across sub-regions of the receptive field. Moreover, the more the MEI deviated from Gabor-like stimuli, the more their orientation tuning diverged from a single-peaked tuning function when larger stimuli covering the receptive fields were used. Complex spatial feature selectivity, deviating from Gabor-like receptive fields, seems to be emerge earlier in the visual pathway of mice compared to primates, which could provide a faster, albeit less general, method to extract task-relevant causal variables.

Keywords: orientation, mouse primary visual cortex, deep learning