Purchase this article with an account.
D. Bradley, M. Lusignan, P. Troyk, S. Cogan, D. McCreery, E. Schmidt, M. Bak, R. Erickson, D. Curry, L. Towle; Simulated Object Recognition with a Cortical Visual Prosthesis. Invest. Ophthalmol. Vis. Sci. 2007;48(13):2346.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Cortical visual prostheses hold the promise of restoring basic vision to patients blinded by retinal malfunction or optic nerve damage. Recent animal work has shown that stimulating implants with 250 or more electrodes are possible. To assess the potential benefit of an array this size, we simulated an object recognition task where sensory input was derived solely from microstimulation currents.
The hypothetical experiment involved a subject facing a screen and a camera that transformed visual signals into a distribution of stimulation currents. These were delivered through an electrode array implanted in striate cortex (V1). Each electrode was assumed to activate a cluster of neurons with overlapping receptive fields. These fields were modeled as Gaussian functions with SD equal to published field widths. Each Gaussian was used as a kernel to integrate luminance on the screen; in turn, current was sent through the corresponding electrode with magnitude proportional to the integrated luminance. The subject was assumed to make 5 saccades such that the distribution of receptive fields (Gaussians) relative to the object being displayed was variable. This effectively increased the size of the array. In the simulation, the distribution of input "luminances" (microstimulation strengths) was compared to a learned distribution, and a Bayesian rule was used to determine which of 8 objects in memory best matched the current input.
Model performance was 49%, substantially and significantly greater than the 13% expected for guessing (p<.001).
Based on currently feasible technology and reasonable assumptions, it appears that a cortical visual prosthesis could provide at least rudimentary visual capability.
This PDF is available to Subscribers Only