NEJM: Let the paralyzed aphasia patient "speak" again?

Without words, using consciousness to communicate with the outside world and control the environment is a dream that human beings have pursued since ancient times

With the development of brain-computer interface technology, although it is still very far away from the consciousness control environment, the use of consciousness to speak has gradually become a reality

Unfortunately, like most brain-computer interface products, NeuroKey can only piece together words and sentences slowly by generating vowels and phonemes

On July 15, scientists from the University of California published a research report in the top international medical journal NEJM, stating that they have developed a system that allows patients who are paralyzed and unable to speak due to illness to restore language activity by relying on brain signals

Specifically, in February 2019, researchers implanted a high-density electrocortical electrode array made by PMT in the sensorimotor cortex area of ​​a patient who had lost mobility and language skills due to pontine infarction.

In a year and a half, the participants received a total of 50 sessions of task training, including "speaking" the vocabulary seen from 50 words and the English sentences seen in a quick report

Training diagram

In order to test the neural activity decoding performance of this model, the researchers analyzed the participants' neural activity data using two indicators: word error rate and the number of words decoded per minute

The accuracy of decoding various sentences in real time through neural signal processing and language modeling

"As far as we know, this is the first example in the scientific community that has successfully decoded complete words directly from the brain activity of a paralyzed, unable to speak patient," said Edward Chang, PhD in neurosurgery at the University of California, San Francisco, the lead author of the report.

At present, researchers are trying to include more types of participants to expand the scope of the study, and are also expanding the vocabulary list to help increase the number of words per minute that the natural language model can recognize

Original source:

David A.