Statistical estimation of visual features in retinal circuitry

Stanford University, Stanford CA

Investigators

Abstract

DESCRIPTION (provided by applicant): The overall goal of this project is to understand how retinal computations are modified in response to the structure of the visual environment. It has long been known that the sensitivity of retinal ganglion cells (RGCs) is depen- dent on recent visual input. These adaptive changes in sensitivity show remarkable flexibility; some RGCs adapt to ignore predictable components of the visual world in favor of encoding what is unexpected. This strategy is thought to maximize the amount of information transmitted to the higher brain. Recent work from our lab, how- ever, suggests that some RGCs use previous visual stimuli to predict future input. High-contrast input to these RGCs, such as a moving object, increases their sensitivity, thereby encoding the object's likely future location. It remais unknown, however, whether RGCs sensitize to properties of the visual environment beyond luminance and contrast, or, more simply, whether the retina maintains predictions about more complex features of the visual world. The first goal of this work is to understand the class of visual features to which the retina exhibits predictive sensitization. A second goal is to take advantage of the tractability of the retina to explore the neural mechanisms underlying prediction, a computation thought to be performed throughout the brain. In addition to improving our fundamental understanding of the retina, these results will have direct application. Retinal prosthetic devices hold the promise of substantially restored vision for those suffering from photoreceptor degeneration diseases, which affect more than two million Americans. As the computation of predictive sensitization is easily implemented in digital circuits, this research wil provide basic scientific insight and computational models to inform in the design of processing elements in retinal prosthetics.