Neuroprosthesis may restore ability to communicate in paralyzed persons with anarthria

Using deep-learning and natural-language models, words and sentences were directly decoded from cortical activity during attempted speech in an individual with anarthria (ie, inability to articulate speech) and spastic quadriparesis.

Anarthria may spring from conditions such as stroke and amyotrophic lateral sclerosis (ALS). [Augment Altern Commun 2007;23:230-242] It impedes communication, leading to impaired quality of life. [Amyotroph Lateral Scler Frontotemporal Degener 2016;17:179-183]

Although cognition may remain intact in individuals with anarthria, some may have limited oral movement and indistinguishable vocalizations when trying to speak. For anarthric individuals with paralysis, the accompanying paralysis may render them unable to operate assistive devices, the researchers noted.

From letters to words

A paralyzed individual implanted with an intracortical device and audiovisual interface was able to generate phonemes and vowel sounds, but not full words. [PLoS One 2009;4:e8218; Front Neurosci 2011;5:65] Other advances have been made to allow speech-impaired individuals to spell out messages. [Sci Transl Med 2014;6:257re7; N Engl J Med 2016;375:2060-2066] However, relying on just sounds or letters is slow and takes effort. “A more efficient and natural approach may be to directly decode whole words from brain areas that control speech,” said the researchers.

To establish whether speech can be decoded directly to produce language from the neural activity of a 36-year-old male with limb paralysis and anarthria caused by a brainstem stroke, 22 hours of cortical activity were recorded over 48 sessions while the participant attempted to articulate individual words from a 50-word English vocabulary set*. [N Engl J Med 2021;385:217-227]

“We implanted a subdural, high-density, multi-electrode array over the area of the sensorimotor cortex that controls speech,” said the researchers. “We decoded sentences from the participant’s cortical activity in real time at a median rate of 15.2 words per minute, with a median word error rate of 25.6 percent.”

On post hoc analyses, 98 percent of the attempts by the participant to produce individual words were detected, and words were classified with 47.1 percent accuracy using cortical signals which remained stable throughout the study period, they continued.

“[The findings suggest] that high-density recordings of cortical activity in the speech-production area of the sensorimotor cortex of an anarthric and paralyzed person can be used to decode full words and sentences in real time,” said the researchers. Together with language-modelling techniques, the deep-learning models used in the study may eventually be utilized to decode a range of sentences.

A neuro-engineering feat

“For persons with anarthria, it would be an extraordinary accomplishment to decode intended speech from brain signals alone … The [study results were] a feat of neuro-engineering,” commented Drs Leigh Hochberg and Sydney Cash from Massachusetts General Hospital and Harvard Medical School, Boston, Massachusetts, US, in a separate editorial. [N Engl J Med 2021;385:278-279]

In tetraplegic and anarthric individuals who are cognitively intact, it’s not a question of what message to deliver but rather, how. These individuals do know what they want to convey, and their brains can formulate messages; however, delivery is trapped by their inabilities, pointed out Hochberg and Cash. “[As such,] for persons with severe speech and motor impairments, restoration of the ability to communicate … is an important goal.”

The findings imply that efforts to restore neurologic function in individuals with stroke, ALS, cerebral palsy, or other neurologic disorders are inching towards clinical benefit. “Ultimately, success will be marked by how readily our patients can share their thoughts with all of us,” they said.