Visual scene segmentation in the mouse

Abstract

Mechanistic studies of the rodent visual system have exploded in the past few years owing to the wealth of molecular, genetic, and chemical tools available for neural circuit perturbations. Although mice have lower spatial acuity than primates, responses of neurons in mouse visual cortex show a marked resemblance in their tuning to elementary image features (e.g. spatiotemporal frequency, direction and orientation) when differences in acuity are taken into account. An outstanding question in the field concerns how these elementary visual feature representations are further transformed by the rodent visual system. For example, does the mouse visual system explicitly extract object boundaries from visual scenes to facilitate object and scene perception? In an attempt to answer this question, we trained mice on a figure-ground segmentation task where figures were defined by either iso oriented gratings in the foreground/background, cross oriented gratings, or naturalistic textures, moving in counterphase to the background. Mice were readily able to report the side of the stimulus containing a texture-defined figure, doing so most readily and with highest accuracy for figures defined by cross oriented gratings. These animals, were able to generalize reports of a visual figure when presented with novel orientations, suggesting that they were able to learn abstract rules about the visual stimulus. Similar, albeit reduced performance could be measured using iso oriented gratings and performance was further reduced for naturalistic textures. We also recorded visual responses when presenting the same iso/cross oriented gratings and naturalistic textures in V1 and extrastriate visual areas RL and LM using both 2-photon calcium imaging and electrophysiology revealing, diverse contextual surround effects. Coherent spatial preference across all texture conditions presented, indicative of a texture-invariant figure response, was never observed. Taken together, these findings serve to inform the limits of the rodent as a model system in the study of scene segmentation.