RT Journal Article
SR Electronic
T1 Mice use robust and common strategies to discriminate natural scenes
JF bioRxiv
FD Cold Spring Harbor Laboratory
SP 156653
DO 10.1101/156653
A1 Yiyi Yu
A1 Riichiro Hira
A1 Jeffrey N. Stirman
A1 Waylin Yu
A1 Ikuko T. Smith
A1 Spencer L. Smith
YR 2017
UL http://biorxiv.org/content/early/2017/06/27/156653.abstract
AB Visual perception is essential for animal survival in natural environments. Mice use vision to navigate and avoid predators. However, the spatial resolution of mouse vision is poor compared to primates and carnivores, and mice lack a fovea. Thus, it is unclear how well mice can discriminate ethologically relevant scenes. Here, we examined natural scene discrimination in mice using an automated touch-screen system, which measured the discriminability for sets of target and distractor scenes. We found that a conventional metric for estimating image resemblance, structural similarity (SSIM), reasonably predicted the discrimination performance. Mouse-to-mouse consistency was high, and the performance of each mouse was better predicted by the population mean than SSIM. This high intermouse agreement indicates that mice use common and robust strategies to discriminate natural scenes. We tested several alternative image metrics to find an alternative to SSIM for predicting mouse discrimination performance. We found that a primary visual cortex (V1) model-inspired approach predicted mouse performance with fidelity comparable to the inter-mouse agreement. The model was based on convolving the images with Gabor filters, and its performance was orientation-specific. The orientation-specificity was stimulus-dependent. These results indicate that, compared to artificial parameters like SSIM, neurophysiological-based models of V1 processing can better predict visual discrimination behavior of naturalistic stimuli.