Age-related macular degeneration is a leading cause of incurable blindness, destroying the central vision, which is essential for reading, face recognition and making eye contact. A new retinal implant — the PRIMA system — provides central prosthetic vision sufficient for reading, but faces remain out of reach: patients describe them as blurred or “cloud white.” While pixel size can be decreased to improve resolution, this project asks a different question: given the implant's limitations in reproducing the retinal code, what is the smartest possible way to prepare an image before sending it to the device?

Working at the intersection of neural engineering and computational neuroscience, we combine a precise model of how the implant stimulates the retina with a model of how the brain processes images to recognize faces and emotions, then use them together to find the image processing that will produce the most faithful face perception, not the most faithful picture on the retina. Early computer simulations show that this approach roughly triples face-recognition performance compared to standard processing, with no changes to the implant itself. The three aims of this proposal: build the full system, test it in normally sighted volunteers viewing simulated prosthetic vision, and finally deploy it in implanted patients. Because the improvement is purely in software, any benefit can reach existing patients immediately. The same principle, encoding information optimally for a low-bandwidth brain interface, should apply broadly to other sensory prosthetics.