Two separate studies published today in Nature indicate that, in the future, brain-to-computer interfaces (BCI) could help restore communication for people who can’t speak due to severe paralysis. In both studies, researchers used brain implants that could pick up brain signals, which were then translated into sentences on a screen using algorithms. While this isn’t a new concept, the exciting thing is that both research teams were able to do this much faster and more accurately than existing technologies.
In the study from Stanford, researchers implanted electrodes into the brain of a patient with amyotrophic lateral sclerosis (ALS) in two areas associated with speech. The BCI was designed to detect brain activity when the patient was trying to speak. Those signals were then fed into an algorithm that associated certain brain activity patterns with phonemes — the sounds that make up speech. To train the algorithm, the researchers had the patient attempt to vocalize or silently mouth sample sentences across 25 sessions lasting roughly four hours each.
In the UC San Francisco and UC Berkeley study, researchers surgically placed a paper-thin sheet containing 253 electrodes onto the brain of a person with severe paralysis from a brainstem stroke. Like the Stanford study, researchers had the patient train the algorithm by attempting to speak so it could recognize which brain signals were associated with different phonemes. Those signals were then translated into facial expressions and modulated speech on a digital avatar.
While the studies used slightly different approaches, the results were similar in terms of accuracy and speed. The Stanford study had an error rate of 9.1 percent when limited to a 50-word vocabulary and 23.8 percent when expanded to a 125,000-word vocabulary. After about four months, the Stanford algorithm could convert brain signals to words at about 68 words per minute. The UC San Francisco and Berkeley algorithm was able to decode at a median rate of 78wpm. It had an error rate of 8.2 percent for a 119-word vocabulary and roughly a 25 percent error rate for a 1,024-word vocabulary.
Although a 23 to 25 percent error rate isn’t good enough for everyday use, it’s a significant improvement over existing tech. In a press briefing, Edward Chang, chair of neurological surgery at UCSF and co-author of the UCSF study, noted that the effective rate of communication for existing technology is “laborious” at five to 15wpm when compared to the 150 to 250wpm for natural speech.
“Sixty to 70 wpm is a real milestone for our field in general because it’s coming from two different centers and two different approaches,” Chang said at the briefing.
That said, these studies are more proof of concept than a technology that’s ready for prime time. One potential issue is that these treatments require long sessions to train the algorithm. However, researchers from both teams told press at a briefing that they were hopeful that algorithm training would be less intensive in the future.
“These are very early studies and we don’t have a big database of data from other people. As we do more of these recordings and get more data, we should be able to transfer what the algorithms learn from other people to new people,” says Frank Willett, a research scientist at the Howard Hughes Medical Institute and co-author of the Stanford study. Willett did note that wasn’t guaranteed, however, and more research was needed.
Another issue is that the tech has to be easy enough for people to use at home, without requiring caregivers to go through complicated training. Brain implants are also invasive, and in these particular studies, the BCI had to be connected via wires to a device on the outside of the skull that was then attached to a computer. There are also concerns about electrode degradation and the fact that these may not be permanent solutions. To get to consumer use, the tech will have to be rigorously vetted, which can be a lengthy and expensive process.
Furthermore, the studies were conducted on patients who still had some lingering ability to move. Some neurological conditions, like late-stage ALS, can result in what’s called “locked-in syndrome.” In this state, a person still has the ability to think, see, and hear but can only communicate by blinking their eyes or other small movements. People with locked-in syndrome are most in need of this type of tech, but more research is needed to see whether this method would be effective.
“We’ve crossed a threshold of performance that we’re both really excited about because it crosses the threshold of usability,” says Chang, noting that the potential benefit of this tech is tremendous if it can be safely and widely implemented. “We are thinking about that quite seriously and what the next steps are.”