+ than 15 yrs ago, a man who was 1-ly 20 yrs old had a massive stroke when a major artery supplying his brain stem burst. the incident left him unable to control his limbs or any muscles rel8d to speech. witha device that relied on his head motions to control a keyboard, he ‘d produce bout 5 words a minute, one toon at a time. the typical rate when some1 is speaking fluidly can be up to 200 words a minute.
now he tis 1st person ever to produce whole words via a computer intermediate that decodes his brain’s messages. a processor connected to an array of electrodes implanted onnis brain receives the messages and transl8s them into words displayed na' screen. as researchers reprted on jul 14 inna new england journal of med, the man, who is now onnis l8 30s, used this brain-computer interface, or bci, to produce whole words outside of his brain for the 1st time since his stroke. in fact, witha suite of at least 50 words, he ‘d even transmit up to 1,000 complete sentences.
earlier generations of neuroprostheses ‘ve relied on communications from the brain to the limb or hand muscles to activate letters na' keyboard. messages are relayed round the unresponsive muscles to a processor that transl8s them into single-letter keystrokes. as with using head movements, word production is slo and tedious and often produces just a few words a minute. now researchers ‘ve decoded the origin of brain signals controlling speech and created the new neuroprosthesis that facilitates the production of whole words, yielding a faster word-per-minute rate.
“this is a big step, one big step among many that we’ll be able to take,” says vikash gilja, an associate professor of electrical and computer engineering atta university of california, san diego, who was not involved inna study. a key advance, he says, tis “incredible proof of concept” that some1 who s'been unable to speak for + than a decade and a ½ can still generate speech signals to use with these interfaces.
“this was not like an overnite kind of thing, where we just plugged it in,” says the study’s senior author edward chang, chair of neurosurgery atta university of california, san francisco. he and his colleagues 1st spent many yrs sorting out how the brain controls speech-rel8d muscles, pinpointing the messages and movements associated with each vowel and consonant inna english α-bet. when they launched the bci restoration of arm and voice (bravo) study to test the 128-electrode brain implant they had developed, the 1st pticipant was the man who had suffered a stroke at age 20, who goes by the pseudonym “bravo-1.”
he worked through 50 sessions offa ½ hr or so each during 81 weeks of the study. inna sessions, researchers ‘d present a target word or sentence na' screen. when bravo-1 engaged his brain to send the rel8d speech signals, the processor picked them up through the implanted electrodes and transmitted their message to a computer. the computer side of the device decoded bravo-1’s messages correctly 74.5 % of the time (logging + than 90 % accuracy occasionally)—n'it produced a median rate of bout 15.2 words per minute.
that is, course, nowhere near the fluidity offa fast-talking teenager. achieving better performance and accurate message decoding will require combining the high accuracy of devices that signal through the upper limb and this “crit demonstration that the speech signals are present and t'they can be leveraged,” gilja says.
chang says that for his group, the nxt steps are to see “if this is better, worse or the same in + pplz” while the researchers also use a larger vocabulary to train the machine that decodes the brain’s output. the vocabulary has already expanded beyond the 50 words reprted in this study, he says, and “it’s exciting to see things grow in that kind of way.”
when asked bout how bravo-1 responded to the success he’s experienced sfar, chang says, “i think he was really thrilled and excited, and this is really just the beginning.”
original content at: rss.sciam.com…
authors: emily willingham