Cortical visual prosthesis

A cortical visual prosthesis is an implanted device that produces visual sensations in a blind person by electrically stimulating the visual cortex directly, skipping the eyes and optic nerves altogether. Each stimulated site evokes a phosphene, a small spot of light at a fixed place in the visual field, and a camera-driven processor chooses which sites to stimulate so that the pattern of phosphenes outlines what the camera sees. Because it needs neither a working retina nor an optic nerve, it is aimed at people blinded by glaucoma, optic nerve injury or loss of both eyes, for whom a Retinal implants and visual prostheses cannot work. As of October 2026 every such device remains investigational.

§How it works

The primary visual cortex holds a retinotopic map: neighbouring points in the visual field are represented by neighbouring patches of cortex, with central vision given far more tissue than the periphery. Stimulating one patch therefore produces a phosphene at a predictable location. Brindley and Lewin found in 1968 that the phosphene map they evoked "agrees roughly" with the classical maps derived from war wounds, and that phosphenes from electrodes 2.4 mm apart could be told apart.¹

Two electrode designs compete. Surface arrays sit on the cortex, as in Electrocorticography interfaces, and need relatively large currents. Penetrating arrays such as the Utah array put electrode tips inside the cortex, near the neurons. In a 1996 study at the US National Institutes of Health, microelectrodes implanted in a woman blind for 22 years from glaucoma produced phosphenes at currents as low as 1.9 microamperes, and tips 500 micrometres apart gave separate phosphenes where tips 250 micrometres apart did not.² Lower current and closer spacing are what make higher resolution conceivable.

A further complication is gaze. Phosphenes move when the eyes move, even after years of blindness, so a camera fixed to spectacles shows the brain one scene while the eye muscles imply another.³ In one Spanish trial participant who had lost both eyes, intended movements of cosmetic eyes still shifted where phosphenes appeared, and tracking those movements in real time let the participant locate targets on a screen.⁴

§Development history

1. 1929
   Phosphenes from the occipital pole
   Neurosurgeon Otfrid Foerster stimulates the exposed occipital cortex under local anaesthesia and patients report points of light.
2. 1968
   First implanted array
   Brindley and Lewin implant radio receivers wired to electrodes over the visual cortex of a 52-year-old blind patient, who sees predictable simple patterns.
3. 1996
   Intracortical microstimulation
   An NIH team implants 38 microelectrodes for four months in a blind volunteer; 34 of them evoke phosphenes.
4. 2000-2002
   Dobelle's commercial system
   William Dobelle reports a camera-to-cortex system and begins selling implants to blind patients, with surgery done in Lisbon.
5. 2018-2019
   Orion implanted
   Six blind adults receive Second Sight's 60-electrode wireless surface implant in a US feasibility study.
6. 2020
   Shapes in monkeys and humans
   A 1,024-channel prosthesis lets monkeys recognise letters, and dynamic stimulation lets blind and sighted people trace letter shapes.
7. 2022
   Wireless intracortical implant
   Illinois Tech's ICVP is implanted at Rush University Medical Center: 25 wireless stimulators carrying 400 electrodes.
8. 2024
   Blindsight designated
   Neuralink announces an FDA breakthrough device designation for its Blindsight cortical implant.

Foerster's observation that stimulating the occipital pole produced points of light, often described as stars or clouds, began the modern line of work.⁵ Brindley and Lewin's 1968 implant showed that a chronic, wireless array could deliver patterned percepts. They also documented the problems later teams inherited: some electrodes produced several spots or a cloud, and strong stimulation could leave phosphenes lingering for up to two minutes.¹

William Dobelle, working on the problem since 1968, reported in 2000 that connecting a camera and computer to the visual cortex had given a blind volunteer useful "artificial vision".⁶ His testing was done privately, outside academia, and many in the field regarded it as suspect.⁷ In 2002 Wired described patients being sent to a surgeon in Lisbon to get around the US Food and Drug Administration, and a price list of around $115,000 for the system, surgery and travel. The reporter watched one paying patient have a seizure when he was overstimulated during testing.⁷ Dobelle died in 2004,⁸ and the programme did not lead to an approved device.

The 1996 NIH study showed that penetrating microelectrodes could work at far lower currents and closer spacing than surface stimulation.² Richard Normann's group at the University of Utah pursued a penetrating implant for the same goal.⁷ In 2021 a team at Miguel Hernández University in Spain, with Normann among its authors, reported placing a 96-electrode Utah array in the visual cortex of a woman blind for 16 years from toxic optic neuropathy. Over six months, 88 of its 96 electrodes reliably evoked phosphenes, and with several electrodes at once she identified some letters and object boundaries; the authors observed no seizures, and the array was removed without complications.⁹

§Current state

The longest record in a registered trial belongs to Orion, a 60-electrode surface array driven by a camera on glasses. Six people were implanted between January 2018 and January 2019, and the study closed in March 2025. According to the developer, all six did better on spot-detection and motion-detection tests with the system on than off, fewer than 4% of electrodes lost function over the course of the study, and one seizure early in the study stopped recurring once stimulation patterns were changed. Three devices were removed after the three-year visit and one at the end of the study.¹⁰ These figures come from the company's summary of a conference presentation rather than from a peer-reviewed paper.

The Intracortical Visual Prosthesis (ICVP) uses many small wireless floating arrays of 16 electrodes each. Its first participant received 25 of them in February 2022,¹¹ and a second received 32, for 512 electrodes. The team recorded two seizures during direct cortical stimulation over three years of testing, both visible on scalp EEG.¹²

Laboratory work in animals has set the ambition for resolution. In 2020 Pieter Roelfsema's group at the Netherlands Institute for Neuroscience implanted 1,024 electrodes in areas V1 and V4 of macaque monkeys. By stimulating several electrodes at once they imposed patterns of phosphenes that the monkeys recognised immediately as simple shapes, motions or letters.¹³ These are monkey results, obtained in animals with normal vision. The same year, a Baylor-led team showed that stimulating electrodes in sequence, tracing a letter across the cortex the way a finger traces one on a palm, let sighted epilepsy patients and two blind participants recognise letter forms; static stimulation of the same electrodes merged into a few large phosphenes.¹⁴ In Australia, the Monash Vision Group's wireless Gennaris arrays were stimulated for over 2,700 hours in sheep without observed adverse effects; those published results are from animals, not people.¹⁵

§Blindsight and the claims around it

Neuralink announced on 17 September 2024 that the FDA had given its Blindsight device a breakthrough device designation.¹⁶ The company's co-founder Elon Musk wrote that day that "Provided the visual cortex is intact, it will even enable those who have been blind from birth to see for the first time", that vision would at first be "low resolution, like Atari graphics", and that it could eventually exceed natural vision and extend to infrared, ultraviolet or radar.¹⁷ The designation gives a sponsor more contact with FDA staff and prioritised review; it does not authorise a device, which must still meet the agency's usual safety and effectiveness standards.¹⁸

The animal data made public include a June 2025 conference talk in which a Neuralink engineer reported, according to Bloomberg, that an implanted monkey responded about two-thirds of the time to visual stimuli evoked by the device.¹⁹ On 30 October 2026 Musk described Blindsight as "ready from an engineering standpoint" and "Just waiting on regulatory approval", and said it would let those who "have never seen" see, this time without the earlier condition that the visual cortex be intact.²⁰ At that date none of the six studies on ClinicalTrials.gov with Neuralink as lead sponsor concerned vision, and no human Blindsight data had been published.²¹

Can a cortex that never saw learn to see?

Each of the cortical prosthesis trials described here has required prior sight: Orion asked for a documented history of useful form vision, ICVP for normal or near-normal vision through at least age ten, and the Spanish trial excludes anyone with under 12 years of good vision.²²²³²⁴ The reason is biological. In people blinded early in life, primary visual cortex becomes active during touch tasks such as Braille reading.²⁵ A study using magnetic stimulation over the occiput evoked phosphenes in every blind participant with some residual vision but in only 20% of those with none, and found the effect weakest in people without previous visual experience.²⁶ Whether an implant could give spatial vision to someone born blind has never been tested, so Musk's statement is a forecast rather than a result.

§Limitations

Resolution is the binding limit. Phosphenes do not behave like pixels: their size stops growing once current passes a modest level and grows with distance from the centre of gaze, which reflects how much cortex each part of the visual field occupies.²⁷ When several electrodes fire together, their phosphenes can fuse into a few large spots instead of the intended shape.¹⁴ Much of human V1 is folded inside the calcarine sulcus, where surface arrays cannot reach it. The 1968 and 1996 implants were each placed over one hemisphere, so their phosphenes fell in one half of the visual field,¹² and every user's phosphene map has to be measured individually.

What users report is far from ordinary sight. The first ICVP participant described his percepts as dots that tell him "there is something there", like "blips on a radar screen", with other senses needed to work out what they are.²⁸ Interviews with Orion participants found that the benefit they described could not be reduced to device performance, alongside mixed expectations and a broad range of fears about taking part.²⁹ The success claimed in these studies is measured in letter discrimination, motion detection and object localisation, not in reading or face recognition.

§Risks

Implantation carries the general risks of brain surgery shared by every implanted device in Neuroprosthetics. Stimulation adds seizures, which have occurred in Dobelle's work, in Orion and in ICVP.⁷¹⁰¹² Stimulating electrodes corrode and provoke glial scarring over time, and even highly biocompatible materials provoke some encapsulation, which limits how long an array keeps working.⁵ One Spanish case runs in the other direction: a participant blind for over three years from optic neuropathy recovered some natural perception of light and motion after stimulation began. The authors propose a plasticity mechanism and say further research must show whether it was an isolated event.³⁰

The commercial risk is already documented. Second Sight announced in 2020 that it would wind down operations, leaving more than 350 recipients of its Argus II retinal implant without support; Nader Pouratian, a neurosurgeon on the Orion trial, told IEEE Spectrum that some trial participants feared being left with unsupported devices, and one of the six had the device removed so as to have an MRI safely.³¹ The company merged with Nano Precision Medical in August 2022 to become Vivani Medical, whose lead product is a drug-delivery implant; Orion stayed in a neuromodulation division.³² The episode bears on Neurorights debates about what manufacturers owe people carrying their hardware, and in Access and inequality arguments about who carries the cost when an experimental implant outlives its maker. Parts of the blind community question the framing of blindness as a defect to repair, a view set out under Disability rights and enhancement.

§Outlook

Cortigent says its next step is a larger pivotal trial intended to seek US marketing approval, subject to talks with the FDA.¹⁰ The ICVP and Spanish studies list completion dates in 2027 and 2028.²³²⁴ Research groups are working on bidirectional arrays that record from neurons while stimulating them, so that the device can check what percept it has produced; a 2025 study in two blind volunteers found that recorded activity predicted the reported percepts.³³ That links visual prostheses to the read-out side of Brain–computer interface work and to Neural decoding.

Biological alternatives such as Optogenetics and Somatic gene therapy need surviving retinal cells and a working optic nerve, so they do not compete for the population cortical devices target. The comparison that matters is with the Cochlear implant, which became routine only after decades of refinement and shows that crude sensory input can become useful with training. Whether visual cortex can be trained the same way, and whether a company can sustain a device long enough to find out, are the open questions. Claims of vision "better than natural", or of sight for people who have never had it, belong for now under Sensory augmentation and Human enhancement speculation rather than under demonstrated results.

§See also

• Retinal implants and visual prostheses
• Neuralink
• Utah array
• Neuroprosthetics
• Brain–computer interface
• Cochlear implant
• Optogenetics
• Neurorights

§References