Optimal Filtering in the Salamander Retina

Abstract

The dark-adapted visual system can count photons with a reliability limited by thermal noise in the rod photoreceptors — the processing circuitry between the rod cells and the brain is essentially noiseless and in fact may be close to optimal Here we design an optimal signal processor which estimates the time-varying light intensity at the retina based on the rod signals. We show that the first stage of optimal signal processing involves passing the rod cell output through a linear filter with characteristics determined entirely by the rod signal and noise spectra. This filter is very general; in fact it is the first stage in any visual signal processing task at low photon flux. We identify the output of this first-stage filter with the intracellular voltage response of the bipolar cell, the first anatomical stage in retinal signal processing. From recent data on tiger salamander photoreceptors we extract the relevant spectra and make parameter-free, quantitative predictions of the bipolar cell response to a dim, diffuse flash. Agreement with experiment is nearly perfect. As far as we know this is the first instance of a predictive theory of neural dynamics.