The study published in the journal Neuron on August 7, 2026 describes a deep learning method that controls electrical stimulation in the visual cortex through a bidirectional cortical implant: Using two control strategies, it achieved the targeted brain activity at lower stimulation currents and elicited more consistent perceptions than conventional methods. According to SciTechDaily, the system was tested on a 27-year-old man who had lost his sight following a traumatic brain injury; a temporary implant with 96 electrodes stimulated his visual cortex at a hospital in Spain and produced flashes of light known as phosphenes. According to the Spanish newspaper elperiodic, the method is intended for people who have become blind through damage to their eyes or optic nerve and still have a functioning visual cortex.
According to the study, existing cortical visual prostheses produce crude, variable perceptual impressions and require manual calibration of each individual electrode, which is not scalable. According to the authors, this is because electrical microstimulation in the visual cortex triggers nonlinear, state-dependent responses, making it difficult to link stimulation and perception.

How the trained method controls stimulation
The method developed by the researchers was trained on neural recordings from individual trials and supports two strategies: a learned inverse network for real-time stimulation synthesis and a gradient-based optimizer for precise targeting. The deep neural network was developed jointly with researchers at the University of California in Santa Barbara and ETH Zurich; it calculates which combination of electrodes and current levels is most likely to produce the desired brain response. The artificial intelligence also takes into account the brain's state immediately before each stimulation. The implant can both stimulate neurons and record their responses. Recorded brain activity predicted the participant's reported perceptions substantially better than the stimulation parameters alone.
Limitations the researchers themselves identify
Before clinical use, the results would need to be verified with additional participants, and the relationship between brain activity and individual perception would need to be better understood. According to SciTechDaily, UCSB professor Michael Beyeler, who was involved in the study, explained that electrodes influence one another and neural responses fluctuate. What is delivered to the brain does not necessarily correspond to what the person perceives. A useful visual prosthesis would need to learn how the individual brain responds and adapt to it, rather than the other way around.
Comparable approaches since 2021: a study with one person
Comparable approaches have been around for some time: In a study published as early as December 1, 2021, researchers implanted an array with 96 electrodes in the visual cortex of a completely blind, 57-year-old person for six months. Simultaneous stimulation through multiple electrodes was associated with a statistically significant reduction in stimulation thresholds and enabled the person to identify some letters and recognize object boundaries; according to the authors, however, a clinically available cortical visual prosthesis did not yet exist at the time. The CORTIVIS study was also still recruiting as of its April 2026 verification and, according to the trial registry, was designed for five participants without blinding. According to the registry, the aim of the study, which began in 2019, is to determine whether a cortical visual prosthesis can provide people who are profoundly blind with a limited but useful sense of vision; completion of primary data collection is estimated for December 2027.
Project work and barriers to clinical use
The NeuraViPeR project had previously conducted simultaneous recording and stimulation experiments in the visual cortex of three blind people over six months. It also developed deep learning algorithms that convert camera images into stimulation patterns. In the project's own assessment, important questions remain unresolved and must be addressed before a cortical visual neuroprosthesis can be considered a viable clinical therapy. A review from March 2026 also identifies immature decoding algorithms, limited electrode performance, high device costs and legal, regulatory and ethical concerns as barriers to widespread clinical adoption.
