Algorithmic framework for enhanced artificial sight

Stanford researchers at the Chichilnisky lab have patented an artificial retina framework for dynamic electrical stimulation to improve the performance of electronic visual implants. The prototype surpasses existing approaches by using a closed-loop device with so-called "greedy dictionary-based encoding", for precise electrical stimulation of retinal ganglion cells (RGCs) in a way that is algorithmically designed specifically to maximize visual performance. This critical step addresses the vision restoration challenge of accurately targeting multiple RGC types to overcome vision loss.

Researchers developed a simple visual perception model and created calibrated RGC activity dictionaries through electrical stimulation combined with recording. Using these personalized dictionaries to process incoming visual signals, they achieved near-perfect efficiency through a greedy algorithm that optimized electrical stimulation patterns dynamically over time to achieve the most acute visual perception. The algorithmic framework produces less heat, and decreases lag time of visual communication.

This approach could dramatically improve the performance of implants for sight restoration, and has broader applications for high fidelity neural implants in ophthalmology, neurology, and audiology.

Visual Encoding Steps in a Healthy Retina and with a Retinal Prosthesis
(Image courtesy the Chichilnisky Lab)

Stage of Development - Prototype:
Proof of concept prototype was tested via a primate retina and dense electrode array. Work is ongoing to improve image quality, refine algorithms and dictionaries of multi-electrode stimulation patterns, and delve further into the retinal signals of the brain.