Abstract
Abstract: :
Purpose: To test an algorithm for generation of visual pattern–specific codes as stimulation signals for epiretinal multi–neuron stimulation electrodes. To simulate the visual system as visual decoder (VD) of epiretinal multiple cell stimulation signals as visual percepts albeit with reduced resolution. Methods: A: Part of the central visual system was simulated as visual decoder (VD). A 34 x 34 array formed the VD input with an even distribution of selective inputs 'expecting' ganglion cell codes exclusively from either P–On–, P–Off–, or M cells. B: The receptive field (RF) properties of these cell types were simulated by a corresponding retina encoder (RE) with an even distribution of a total of 34 x 34 P–On–, P–Off–, and M cell filters to encode visual input patterns with 64 x 64 pixels. These three filter types generated distinct temporal output signal codes in response to visual stimulation of specific RF–related pixel combinations. C: VD was designed to decode the 34 x 34 temporal codes into a visual percept to mimick a corresponding visual input pattern with 64 x 64 pixels (Eckmiller, Neumann, Baruth: ARVO,2004). D: The decoding properties of VD were studied for simultaneous stimulation of three or more neighboring ganglion cells using specific temporal codes with + / – 12 discrete amplitudes and 4 time values. Results: (1) Whereas single ganglion cell stimulation was recently postulated to require the removal of the high code redundancy (Eckmiller, Neumann, Baruth: ARVO,2004), multiple cell stimulation was successfully simulated here by using only one of the three distinct RF–center– or of the three total RF (center + periphery) codes . (2) Due to the VD processing rule that for a given VD input channel only legitimate codes (e.g. P–On center code only for P–On inputs) were considered, the P–On, P–Off, and M channels in common contact with a given electrode could be stimulated separately. (3) The visual percept simulation as VD output yielded a reduced spatial resolution, which was limited by the distance between neighboring RF–centers of the RE input array. (4) Visual percept quality could be improved by sequential stimulation with RF–center– vs. total RF codes and by selective switching between the codes of the three cell types. Conclusions: A. The interpretation of the central visual system as visual decoder (VD) rather than pattern recognizer offers novel explanations for the principal ability of epiretinally stimulated blind subjects to perceive visual objects rather than noise. B. Dialog–based tuning of temporal stimulation codes is important for retina implants especially in cases of electrode size–caused multiple ganglion cell stimulation.
Keywords: electrophysiology: non–clinical • shape, form, contour, object perception • computational modeling