AI-Powered Deep Learning Improves Precision of Bionic Eye Technology

Aug 7, 2026

Researchers from UC Santa Barbara, ETH Zurich, and Miguel Hernández University have demonstrated that a deep-learning AI model can improve the precision of bionic eye technology. In a 2024 clinical test in Spain, researchers used the AI model to design electrical stimulation patterns for a 96-channel electrode array temporarily implanted in the visual cortex of a blind 27-year-old man. The AI-designed patterns accurately reproduced targeted brain activity with less electrical current, adapting to the brain's fluctuating resting states to improve phosphene perception.

AI-guided visual cortex stimulation

▪The AI-designed electrical stimulation patterns reproduced targeted brain activity patterns more accurately and required less electrical current than alternative stimulation approaches.
▪Researchers from UC Santa Barbara, ETH Zurich, and Miguel Hernández University used a deep-learning model to design electrical stimulation patterns for electrodes implanted in a blind participant's visual cortex.

Deep learning neural prediction

▪Researchers trained a deep neural network to predict the patterns of brain activity produced by different combinations of electrical stimulation settings in the visual cortex.
▪The deep-learning model incorporated measurements of the participant's resting brain activity immediately before each test to predict neural responses to electrical stimulation.

Phosphene perception patterns

▪Recorded patterns of neural activity in the study participant's brain were more informative about his perceptual experiences of phosphenes than the electrical stimulation parameters themselves.
▪Electrical stimulation of the study participant's visual cortex electrodes caused him to perceive phosphenes, which are spots or shapes of light resembling flashes, stars, or fireworks.

Visual cortical prosthesis development

▪The study participant received a temporary brain implant composed of a 96-channel array of electrodes in his visual cortex, which was removed after six months.
▪The proof-of-concept study, published in the journal Neuron in August 2026, represents a step toward developing visual cortical prostheses that bypass damaged eyes and optic nerves.
▪The clinical testing was conducted in 2024 at Hospital IMED Elche in Spain on a 27-year-old male participant who lost his vision following a traumatic brain injury.

Adaptive brain-responsive stimulation

▪The deep-learning model adjusted stimulation patterns based on the study participant's resting brain activity, aiming to keep visual prostheses reliable as neural responses fluctuate daily.
▪UC Santa Barbara associate professor Michael Beyeler stated that a useful visual prosthesis must learn how an individual brain responds and adapt the stimulation accordingly.