Loading ...

An advanced brain-computer interface (BCI) that can translate brain signals into speech with up to 97% accuracy, has restored the speech of a man with amyotrophic lateral sclerosis (ALS). 

"Previous speech BCI systems had frequent word errors. This made it difficult for the user to be understood consistently and was a barrier to communication. Our paper demonstrates the most accurate speech neuroprosthesis ever reported," study investigator David Brandman, MD, PhD, co–principal investigator and co-director of the University of California (UC) Davis Neuroprosthetics Lab, said in a news release. 

The system has allowed a man with severely impaired speech due to amyotrophic lateral sclerosis (ALS) communicate with friends, family and caregivers.

The technology has been "life-changing" for the patient. "He is able to speak to his 4-year-old daughter, something he couldn't do before due to his condition," Sergey Stavisky, PhD, co–principal investigator and co-director of the UC Davis Neuroprosthetics Lab, noted in interview with Medscape Medical News. 

The investigators describe the research in a paper published online on August 14 in The New England Journal of Medicine.

New Milestone 

BCI technology has allowed people with paralysis to communicate by transforming brain activity associated with attempted speech into text on a computer screen. However, communication with BCI systems has been restricted by lengthy training periods and limited accuracy, making it hard for the user to be understood consistently.

The new system rapidly reached a level of performance not reported before.

The 45-year-old man with ALS with tetraparesis and severe dysarthria underwent surgical implantation of four microelectrode arrays into his left ventral precentral gyrus 5 years after the onset of the illness. These arrays record neural activity from 256 cortical electrodes. 

The BCI interprets brain signals when the user tries to speak and turns them into text. The decoded words are displayed on a screen and read aloud by the computer in a voice that sounds like the patient. The voice was created using software trained with audio samples of his pre-ALS voice. 

The patient successfully used the system in both prompted and spontaneous conversations. In both cases, speech decoding happened in real time. 

On the first day of use (25 days after surgery), the neuroprosthesis achieved 99.6% accuracy with a 50-word vocabulary. 

On the second day, after 1.4 additional hours of system training, the neuroprosthesis achieved 90.2% accuracy using a 125,000-word vocabulary. 

With further training, the system maintained 97.5% accuracy over a period of 8.4 months. For comparison, with previous speech BCI systems, "accuracy of decoded speech was around 75%," Stavisky told Medscape Medical News.

The patient has used the system to communicate in self-paced conversations at a rate of roughly 32 words per minute for more than 248 cumulative hours. 

This level of accuracy represents "a new performance milestone for speech neuroprostheses," Edward Chang, MD, wrote in an accompanying editorial.

The performance achieved "approximates the accuracy of modern speech-recognition systems, which are trained to transcribe audio into text sentences. The decoder could learn rare words and could be trained rapidly and recalibrated online, which has not been shown previously," noted Chang, with the Department of Neurological Surgery, University of California, San Francisco.

"Perhaps most important," said Chang, the researchers provide evidence of "real-world usage, in which the speech neuroprosthesis was used in interactive conversations."

The UC Davis Health team cautions that the results represent the experience of one patient who had retained some residual speech function. They plan to expand the trial to include more patients and further refine the technology. 

In addition, the long-term durability of the system, particularly as ALS progresses, remains uncertain.

Long-Term Data 

In a separate report published online on August 14 in The New England Journal of Medicine, researchers describe longevity (over 7 years) of independent, at-home use of a BCI system implanted in a woman with ALS in 2016.

As previously reported by Medscape Medical News, the BCI system used in this patient consists of four subdural electrode strips implanted over the sensorimotor and dorsolateral prefrontal cortex, a transmitter placed subcutaneously beneath the left clavicle, a receiving antenna placed on the chest over the device, a receiver, and a tablet computer.

The 7-year results "indicate that subdural, electrocorticographic, electrodes can provide a high-quality signal for many years, offering stable and accurate BCI performance, suitable for unsupervised at-home use by people with severe motor impairment," Nick Ramsey, PhD, and Mariska Vansteensel, PhD, with the Brain Center, University Medical Center Utrecht, the Netherlands, stated in a joint email to Medscape Medical News. 

For several years, the woman used the BCI system and relied on it for communication. In later years, however, she used it less often; this coincided with a decline in "click-command" accuracy, which was attributed to ALS progression. The woman stopped using the BCI device when control of the BCI became unreliable, Vansteensel and colleagues noted in their paper. 

"The manufacturer (Medtronic) has retired the device, but it served our study very well," Ramsey and Vansteensel said. 

Rapid Progress; What's Next? 

In his editorial, Chang noted that, over the past decade, the concept of a speech neuroprosthesis has gone from "science fiction to reality."

In his view, these two reports provide "compelling new evidence of rapid progress in clinically viable, practical applications of brain-computer interfaces to restore communication to persons living with paralysis."

"These and other studies have made it clear that persons with paralysis can meaningfully benefit from communication brain-computer interfaces, and it is clear that the development of newer devices and neural interfaces is critically needed," Chang wrote.

The ideal interface, Chang said, will provide "high-performance decoding and reliability over the long term (at least a decade), which will require that it be fully implantable (wireless), with a large bandwidth to capture information-bearing cortical signals that are critical to decoding."

"In addition, the device should be easily implanted (and explanted) without injury to the brain. Many academic and commercial efforts are under way to address this challenge. The convergence of neuroscience, artificial intelligence, and engineering of new neural interfaces continues to bring us closer to realizing the goal of restoring naturalistic communication function for persons with paralysis," Chang concluded. 

This research had no commercial funding. Disclosures for the research teams are available in the original articles.