Purchase this article with an account.
J. L. Wyatt, Jr., A. Eisenman, S. Valavanis, S. Fried, S. Stasheff, J. F. Rizzo; Estimating Visual Motion From Neural Output of Rabbit Retina. Invest. Ophthalmol. Vis. Sci. 2007;48(13):2564.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
This in-vitro study is related to the efforts of the Boston Retinal Implant Project to develop a sub-retinal prosthesis to restore vision to the blind. It quantitatively explores the hypothesis that retinal output from directionally selective (DS) ganglion cells permits more accurate determination of local motion parameters than output from other ganglion cell types, e.g., on-transient cells.
We recorded from the flat-mount New Zealand white rabbit retina using a 60 channel multi-electrode array (MEA) with 200 µm spacing between electrodes. Moving bar stimuli were used to confirm cells as directionally selective (DS) and to determine preferred directions, and a 3- second separation was maintained between leading and trailing edges. After spike sorting, cell locations were determined by comparing the known location of a moving optical edge to the mean time of the resulting burst of spikes, with edge motion in opposite directions and comparison to recording electrode position used to check for consistency. The unknowns to be estimated for a second (unknown) moving optical edge, the speed and orientation, consisted of two components. The differences in positions and response times for each pair of cells gave a single constraint the two components must satisfy, and imprecision in the measurements caused imprecision in each individual constraint. A novel algorithm greatly reduced the imprecision by combining all these approximate constraints using a matrix pseudo-inverse to find a least-squares solution for the speed and direction of a second moving optical edge.
Reasonably accurate indications of speed and angle were obtained using larger numbers of widely separated on-transient ganglion cells, but the accuracy fell dramatically when smaller numbers of closely spaced cells were used. More specifically, data from 9 pairs of cells separated by about 1500 um yielded speed errors of 2% - 3.5% for an edge moving at 850 um/sec and angle errors of 0.5% - 2.5%. In contrast, data from 5 pairs of cells separated by about 300 um yielded speed errors of 26% - 323% and angle errors of 4.7% - 22%. Other preliminary results suggest much greater accuracy for data taken from 5 pairs of closely spaced DS cells.
These initial measurements suggest that under our testing conditions on-transient cells are unable to provide accurate data for motion in local regions, while DS cells provide greater accuracy in local motion estimation.
This PDF is available to Subscribers Only