Retina Encoder Inversion for Retina Implant Simulation.pdf

Retina Encoder Inversion for Retina Implant Simulation M Becker
M Braun
R Eckmiller Informatik VI Neuroinformatik
 Universitat Bonn D Bonn F R Germany email mbnero unibonn de URL http www nero unibonn de Abstract To test tuning methods for the Retina Encoder RE
of a Retina Implant RI
as a visual prosthesis for blind subjects with retinal degenerations a suitable simulation of the patients evaluative response to RE state al terations must be provided RE simulates real time retinal information processing and consists of several hundreds of spatiotemporal receptive eld RF
lters to generate electrical signals for ganglion cell GC
stim ulation We propose a neural network to reconstruct the RE input from a number of consecutive RE output frames The network can be inter preted as a simulation of a part of the central visual system with GC signals as input and perception visualization as output We present rst results using Evolution Strategies for neural network weight optimization Problem Formulation Retina Implants for blind subjects with retinal degenerations are currently under development  In the future the blind subject wears a Retina Encoder RE
in a frame of glasses to observe the visual environment and to process image signals in real time according to a suitable model of retinal informa tion processing Intact retinal ganglion cells GC
are stimulated electrically with the simulation output signals to elicit visual perceptions to regain a mod est amount of vision RE consists of receptive eld lters RF
 one for each GC to be stimulated Each RF has several spatial and temporal parameters and can be tuned continuously within a wide range of GCtypes observed in retina physiology The set of RF parameters forms the RE state space During stimulator implantation the actually contacted GCtypes remain unknown and must be determined in a dialog with the implant carrying subject Learning algorithms have been