A brain implant gives speech to a paralyzed man

summer”, a paralyzed patient silently responds to the question “what is your favorite time of year” displayed on the screen. The display of these few apparently simple words is the result of a technical feat published in the journal NatureCommunications by researchers at the University of California in San Francisco (USA). Because the man cannot speak, it is thanks to a cerebral implant paired with a computer that he succeeds in expressing himself in writing.

Electrodes that capture mental spelling in “alpha bravo”

Deprived of any movement in the limbs and the vocal apparatus by a cerebrovascular accident (CVA), the voluntary patient had a chip implanted with 128 electrodes. Placed against the area of ​​the cerebral cortex involved in the articulation of language, the chip transmitted the brain activity detected to the computer to convert it into sentences on a screen. The principle is simple. Since in our brain, imagining doing something activates the same areas as actually doing it, the subject simply had to speak in his head. Not quite “speak” in truth, but spell each word letter by letter, using the NATO code says “alpha bravo”. When the patient thought “alpha” or “echo” for example, the computer wrote “A” or “E” respectively. To signal that he has finished speaking, the patient must simply imagine moving his hand, a signal easily distinguished from the others that marks the end point.

A success rate of 94% per character on more than 1,000 words

With spectacular results, since the letters displayed on the screen were 94% accurate, among a vocabulary of 1,154 words – enough for everyday conversations – where the previous version of these experiments barely exceeded 75% among 50 words. One of the keys to this impressive leap forward lies in computer models processing brain signals, explains neuroscientist David Moses, the first author of this work. “The natural language model tells us which sequences of words and letters are more likely than others, using statistical information from the English language“Without this model, the error rate could reach 35%, far from the 6% of the final result.”This means that we could correctly decode about two out of three letters from the brain signals”, explains the researcher. “This is a very sensitive and significant improvement.“This natural language model was thus able to predict which letters were most likely in second, third, and up to the nth position.”It is very difficult to create perfect decoders with brain signals, given the complexity of these signals“, explains David Moses, also the use of these models makes it possible to “considerably improve” the result. And transform the result of the neural decoding “Thankytu” into understandable “Thank you”.

7 words per minute

While spelling the words may seem tedious, it should be remembered that many paralyzed patients, “including our participant himself”, points out David Moses, are generally experienced users of communication aids currently on the market, precisely based on spelling. Based on eye movement or other residual motor capacity, none of these commonly used interfaces currently involve brain implants. Thanks to this new technology, the man managed to express himself at a rate of about 7 words per minute (about 30 characters).

Three times slower than another recently tested interface, this time based on detection of brain signals from hand movements. By imagining writing, the patient sent signals interpreted by the computer as handwritten sentences, at the rate of 90 characters per minute – comparable to the 115 words per minute produced by able-bodied people on a smartphone. But this technique required an electrode implanted deep in the brain, a much more cumbersome operation than in this new work. “In general, invasive brain-computer interfaces – which require an implant in the skull – perform better than non-invasive ones, as they can directly record brain activity with better signal quality.“, explains David Moses.

A system capable of operating on more than 9,000 words

Another advantage of his device is that it is based directly on attempts to speak, an approach which may prove to be more intuitive and natural than others, based on writing for example, for many patients. In a model carried out subsequent to their work, the researchers have already shown that their system could decode sentences from an extended vocabulary of 9,170 words with only 8% error! For comparison, 9,000 words are enough to read English fluently, and in French we use an average of 5,000 words to make ourselves understood fluently. “We even think that future versions of our system could use letter-by-letter spelling and even allow certain very frequent words to be spoken directly.”, anticipates David Moses. Ultimately, it will be up to each patient and their healthcare team to weigh the advantages and disadvantages of each approach to decide what suits them best, concludes the scientist.

A brain implant gives speech to a paralyzed man – Sciences et Avenir