|M.Sc Student||Aharoni Tal|
|Subject||Computational platform for portable holographic|
vision restoration neural interfaces
|Department||Department of Autonomous Systems and Robotics||Supervisor||Professor Shy Shoham|
Outer-retinal degenerative diseases like Age-related Macular Degeneration and Retinitis Pigmentosa are among the most common causes of blindness. These diseases lead to the loss of photoreceptors, but relatively spare inner retinal cells including the Retinal Ganglion Cells (RGCs). Recent work has shown that combining Optogenetic sensitization to light of the remaining RGCs, with the projection of intense light patterns could offer a new path towards vision restoration by efficiently stimulating the RGCs, thus producing visual percepts in a blind patient. Here, we address several major computational challenges in the development of Holographic Optical Neural Interfaces (HONI) for vision restoration, including the generation of speckle-free holograms and the need for real-time processing in a portable, power efficient platform.
First, we introduce a new iterative phase retrieval algorithm termed ’Weighted Gerchberg-Saxton with Phase-Control’ (GSW-PC) that allows projection of amplitude and phase-controlled one-dimensional patterns and thereby solves the problem of holographic speckle, a major noise source in computer-generated holography. The new framework can also address the challenge of projecting volumetric amplitude and phase controlled patterns, by incorporating GSW-PC with the angular spectrum method. Next, we develop a simplified retinal emulation model that compensates for the loss of processing by the retinal layers prior to the RGCs. The model processes a stream of input images, and produces appropriate patterns suited for stimulation of the RGCs. Lastly, we combine GSW-PC and the retinal emulation algorithms and implement them on Nvidia’s Jetson TX1 development platform, achieving a portable, power efficient computational device for HONIs, able to produce appropriate holograms at a rate that approaches real-time.