In the race for Brain implants, it is not only speed that counts, but also precision, especially for decoding speech! Several devices recently presented manage to restore the possibility of communication to people who are paralyzed, suffering from amyotrophic lateral sclerosis (Charcot's disease) or following a cerebrovascular accident (CVA). Arriving for example at decode 70 words per minute, about half the typical speaking rate (around 150 words per minute).
But these systems still had a relatively high error rate, with about 25 % errors, that is, one word in four was wrong. Other systems achieved better accuracy (94 % success per character for this one), but with a very low rate of just seven words per minute.
A new study, led by researchers from the University of California, Davis (USA), has managed to combine precision and fluidity, paving the way for more efficient devices. Their results were published on August 14, 2024 in the journal New England Journal of Medicine.
An implant that “reads” what we want to say
This new approach involved "reading" a large region of the cerebral cortex associated with the movements needed for speech (among other things), called the precentral gyrus. The importance of this region had previously been identified with functional magnetic resonance imaging, by visualizing brain activity when the patient tried to speak.
The implant has four chips, each containing 64 electrodes (256 in total), inserted one millimeter deep into the cortex. These chips are linked to connectors, fixed to the skull, into which HDMI cables can be plugged to transmit the decoded information to a computer.
“ What we detect is their intention to move the muscles to speak, explains in a communicates Sergey Stavisky, author of the study. We record activity of the part of the brain that tries to send these orders to the muscles. And we translate these brain activity patterns into phonemes, such as syllables, and then In the words they want to say."
Accuracy close to 100 %
The implant was tested in a 45-year-old man who had suffered from Charcot's disease for five years.
Read alsoCharcot's disease: discovery of a new treatment avenue
Due to the disease, he is partially paralyzed, requiring assistance to move and eat. He was still able to speak, but for his speech to be understandable, he had to speak very slowly (about seven words per minute). With the implant, the system decoded 30 words per minute, with an accuracy of 97.5 % after 15 training sessions with the device (about 16 hours in total).
To put this failure rate of only 2.5 % into context, current software used by smartphones to recognize speech is wrong about 5 % of the time. According to his relatives, the voice used by the system to read his words and sentences was close to the one the patient had before he had symptoms.
“ This technology is transformative because it gives hope to people who want to talk but can't.", enthuses neurosurgeon David Brandman. A feeling of optimism shared by patient Casey Harrell: No p"Being able to communicate is so frustrating and demoralizing. It's like being trapped. Technology like this will help people reintegrate into their lives and society."
English 
French