WHEN Delvin Kehoe looked at himself in the mirror earlier this year, he was appalled by what he saw. Instead of the smiling face of a man in his 30s, he was greeted instead by the grimace of a 59-year-old. But Kehoe was not suddenly feeling his age, nor even reviving from a long coma. He was simply seeing his reflection for the first time in 20 years.
His is one of several miracle tales sweeping the ophthalmological community. Years after losing their sight to a degenerative disease – in Kehoe’s case, retinitis pigmentosa – patients have been given an eye implant that restores their vision. These “visual prosthetics” work by artificially stimulating the retina with electrical signals that are fed down the optic nerve into the visual cortex. Of the 20 people who have received such implants worldwide, some claim that they deliver such clear and detailed images that they can not only see their reflections, but can make out objects as tiny as ants on a pavement. One man even felt confident enough to drive his car, albeit on private property.
But if the implants really work that well, why aren’t millions of blind people lining up for them? The truth is that, despite the compelling stories, many researchers are starting to question how well the current technology really works. For every Delvin Kehoe, there are several others whose visual prosthetics deliver little more than dots and fuzzy patches. For these people, the environment around them resembles at best a game of space invaders, full of grainy, dotted images that bear little resemblance to reality.
Advertisement
Now one researcher is proposing a fundamental rethink of the design principles used in existing retinal implants. Kwabena Boahen, a bioengineer at the University of Pennsylvania, says that instead of merely stimulating the retina we should be simulating it. He’s created an artificial retina that copies the function of a real one down to the firing of individual neurons. Although he has yet to try his implant in a person, he’s confident that his design can one day deliver full natural vision to blind people.
There are many reasons for doubting that current technology can produce anything like natural vision. All the implants undergoing trials at the moment are based on the 1950s discovery that stimulating the retina with electrical pulses produces phosphenes, or vision sensations, even in blind people. In most cases, people become blind because their retinas no longer sense light, but the rest of their visual processing system – neurons in the retina, optic nerve and visual cortex in the brain – works fine. If an implanted chip can stimulate enough meaningful phosphenes in the brain, the theory goes, blind people will be able to see again.
But how do you design the implant so people get useful phosphenes? There is no clear consensus. Perhaps the most obvious technique, called the epiretinal approach, is to cover the surface of the retina with an array of electrodes, each of which can stimulate the ganglion nerve cells underneath it to fire. The electrodes receive their signals from a transmitter that monitors visual information through a camera on the patient’s glasses.
But critics of this approach do not like the implicit assumption that ganglion cells are simple on-off switches. In a healthy retina, these cells do not receive signals directly from photoreceptors. The photoreceptors pass their information to layers of other cell types – bipolar, amacrine and horizontal cells – which then process it and relay it to the ganglions. In the epiretinal chips, by contrast, there’s no pre-processing. “It’s truly artificial,” says Mark Humayun, whose team at the University of Southern California in Los Angeles gave one patient an epiretinal implant earlier this year.
And this pre-processing step is a crucial element of vision. The ganglion cells use this network of cell layers to compare differences between the signals from nearby photoreceptors and only relay data on striking changes, such as a border to an area of fixed intensity or its motion across the field of vision. Less interesting information is ignored. So the signals that ganglion cells relay to the optic nerve are already partially processed. In a prosthetic that doesn’t carry out this processing, the brain loses a source of useful information. This can lead to contradictory signals, if, for example, the chip directly stimulates a ganglion that normally only fires when it detects a border but also stimulates neighbouring cells in a way that indicates a region of constant brightness.
It is a wonder, then, that Humayun’s patient can see 16 dots of light in a four-by-four array corresponding to the locations of the electrodes on his retina. The resolution does improve slightly if the patient moves their head to scan the visual field but “high resolution and near-natural vision is a long way off”, says Humayun.
Another technique, the subretinal approach, looks more promising. Here, the implant sits at the back of the retina and stimulates the bipolar cells. Because these are upstream of the ganglion cells, subretinal implants have the potential to tap into some of the processing power of those parts of the retina that are still working.
Subretinal implants seem particularly promising because they would let the rest of the eye function as normal, allowing the lens to focus images directly onto photodiodes on the chip, which convert the light into electrical energy. But with this comes another difficulty.
The electrodes must be separated by a large enough distance so that they don’t interfere with each other’s signal. As a result it is still not possible to address individual bipolar cells, and therefore large numbers of cells are stimulated simultaneously and resolution is lost, says Eberhart Zrenner, a neuro-ophthalmologist at the University of Tübingen in Germany who is developing subretinal chips. He also says that subretinal chips do not incorporate a full understanding of how the bipolar, amacrine and horizontal cells process visual data. He does not expect the approach to do much more than provide blind people with some basic awareness of their surroundings. Their visual acuity will always fall a long way short of natural vision, he says, and will always be slightly distorted (see Graphic).
However, Vincent and Alan Chow, whose company Optobionics in Wheaton, Illinois, implanted Kehoe’s chip, claim to have taken subretinals further. Optobionics’ chips contain 5000 photodiodes, so should in theory give at least three times the resolution of Zrenner’s, probably enough to see your reflection in a mirror.
But it is unclear to what extent Optobionics’ successes are due to the implants. In May, the company reported that its chips have had an unusual and unexpected side effect. Rather than just stimulating groups of cells near each of the electrodes, their patients were experiencing improved vision in areas of the retina well beyond the physical reach of the chips. Previously damaged parts of their retinas were suddenly working again, even in places not covered by the chip.
So what’s going on? The Chows argue that the current generated by the chip is reactivating photoreceptors across large areas of the retina beyond the reach of the chip. While arguably this would not qualify the implant as a prosthetic, because the patients aren’t directly using it to see, it would be a truly remarkable effect.
But others working in the field prefer a different explanation that gives even less credit to the Optobionics chips. In many retinal diseases, including retinitis pigmentosa, it is not unusual for faulty photoreceptors to start working again, says Zrenner. This “rescue effect” can be caused by the trauma of surgery itself. Tissue experts agree. “The eye is very sensitive,” says Peng Tee Khaw, an expert in tissue repair at Moorfields Eye Hospital in London. “The moment you fiddle around with it you get a release of all these growth factors. Even if you inject water into the eye you get rescue.”
If the critics are right, Kehoe and fellow patients at Optobionics could have the same visual experiences if surgeons carried out dummy operations on their eyes where they didn’t bother to implant the chips at all. Alan Chow disagrees. He says a healing effect triggered by the trauma of surgery couldn’t last as long and be as robust as the results he has seen in his patients.
Perhaps the most promising work in visual prosthetics, in terms of what the patient actually sees, has been from cortical stimulation, or stimulating the brain directly. William Dobelle, of his own Dobelle Institute in Commack, New York, believes that visual prosthetics should not just be developed for people with retinal diseases but for those who have suffered trauma to the eye or optic nerve. If a device is to cater for a wide range of causes of blindness, it will have to bypass the eye.
Dobelle’s technique involves implanting electrodes into the visual cortex. The electrodes then receive a video-like feed down a wire from a camera mounted on the patient’s glasses. “Initially the phosphenes appear almost randomly,” says Dobelle. But he improves the images by adjusting signals delivered to the electrodes while the patient describes what they’re seeing. He also uses algorithms to find the edges of objects in the camera image and ensure that light-dark boundaries appear in the phosphenes at the right point in the visual field.
But in the 30 years Dobelle has been experimenting with human vision, the visual acuity of his patients has not improved beyond about 100 pixels. In comparison, normal visual acuity is at least 10,000 times that. For all the sophistication of his software it is questionable whether he could ever match the efficiency of the retina in producing natural vision. His electrodes stimulate hundreds if not thousands of neurons simultaneously, but to get natural vision he’ll need to increase the number of electrodes dramatically and find a way to prevent their signals from interfering with each other.
Enter Boahen. He says the problem with existing prosthetics is that they are not based on any real understanding of how a healthy retina functions. Rather than merely relaying signals to the brain, the retina performs a huge amount of pre-processing. Instead of treating the retina purely as a light-detecting surface, Boahen says you have to think of it as part of the brain. Until we understand how the retina “thinks” we’ll never build good prosthetics.
So Boahen has spent the past few years trying to fathom the colossal computations that go on in a healthy retina. To get a sense of how much processing the retina does, consider that there are nearly 130 million photoreceptors in each human retina, but only about a million neural pathways in the optic nerve. The eye couldn’t possibly cram all the information it receives down such a narrow pipe. Instead, the retina filters and interprets the information and sends only the interesting stuff. “Most of the time [ganglion] cells are not firing at all,” explains Boahen. “They only respond to changes.”
Boahen is now developing a silicon device that tries to copy the function of all of these layers of cells. It’s a tall order and many of those who are already implanting less complex chips can’t shake the feeling Boahen is trying to run before he can walk. “We don’t want to over-engineer it,” says Humayun. Boahen’s “retinomorphic chip” has several layers of different, interconnected “cells” etched in silicon. Like their biological counterparts – the bipolar, amacrine, horizontal and ganglion cells – these silicon cells perform different and very specific signal processing functions.
Boahen’s chip already detects light and performs the same kind of edge and motion detection functions as a real retina, ultimately resulting in an electrical signal rather like that in a healthy optic nerve. “We more or less have a complete retina model,” says Boahen. “We have been able to develop outputs that match the outputs of the optic nerve.”
Processing a great many signals in a device simultaneously is no mean feat. Silicon chips use roughly 100,000 times more components and connections to run a given set of computations than the human nervous system. That’s because they run computational tasks one after the other, instead of running several in parallel like the brain can do.
Wiring the retinomorphic chip to run enough computations without generating too much heat is a remarkable piece of engineering. Chips generate a lot of heat and biological tissue doesn’t take too kindly to this. But Boahen has found a way to process thousands of signals in parallel by getting single transistors to behave like individual neuronal synapses. In chips, transistors normally behave as simple switches, digital on-offs that form simple binary logic circuits. Boahen’s transistors produce analogue signals that are not simply on or off, but can be stronger or weaker as well. Over time, the transistors are relaying a waveform-like signal that is a function of the waveform inputs they get from the silicon cells.
This solves the heat problem because, with the richer analogue signals, about 1000 times fewer transistors are needed to convey the same information. But Boahen could not implant the chip right now because in its current form the chip still does not match the resolution of the retina. With roughly 6000 photoreceptors feeding into about 1500 ganglion cells, its components need to shrink by a factor of about 10,000 to match the capabilities of a real retina. Although Moore’s Law is on Boahen’s side – every 18 months the number of transistors chip makers can cram onto a chip doubles – one problem the chip industry won’t solve is how to engineer the face of the chip so that it interfaces to living tissue at high resolutions.
Solving these problems will take several years, which is no great comfort to people who are already coping with blindness, but Boahen believes his sophisticated chips will be worth the wait. He hopes that pioneers like Zrenner, Dobelle, Humayun and Chow will be using his technology when his first volunteers open their eyes to truly natural vision. This is not to criticise the effort that’s gone into current technology – discovering what kind of stimulation gets different nerves to fire is important. The difference now is that Boahen’s chip is talking the brain’s language.