Brain-computer implant devices have gained traction in the past few years for their indication in helping paralyzed people. The research has even attracted tech companies like Elon Musk’s Neuralink, who are currently testing their device on a monkey.
However, few of these devices have actually reached human trials, and even fewer successfully show satisfactory results.
Recently, a team of researchers attempted to decode the speech of a patient with anarthria. The 36-year-old male participant had lost his ability to speak (anarthria) or move his limbs (quadriparesis) due to a brainstem stroke 16 years ago. Before this study, he communicated his thoughts using little movements of his head to select words on a computer screen. This was a very time-consuming approach, producing only 5 words per minute.
In an effort to improve the patient’s communicative ability, a neurosurgeon from the University of California San Francisco, Edward Chang, and his team took a piece of the patient’s skull out and placed a thin sheet of electrodes on the subdural mater above his sensorimotor cortex, the area of the brain that controls speech.
The team used brainwaves from this area to train a deep learning algorithm over 22 hours. They asked the man to repeatedly say each word off a setlist of 50 so that the algorithm can relate the word with the type of brainwaves producing it in real time. They then progressed to full sentences using the same 50 words. The team also added a natural-language model that helped the algorithm predict the next word of the sentence.
According to the study, published in the New England Journal of Medicine, this technique allowed the participant to communicate almost 15 words per minute, a major improvement from his previous ability. The model only had an error rate of 25%, meaning that it got 3 out of 4 words right.
Previously, the researchers have tested this system on patients who did not have anarthria. This is the first time they have applied it to a person who really needed it. However, the current set-up is quite bulky, as the participant is constantly tethered to a large computer. In the future, they plan on making the device wireless so that patients can have more autonomy of movement.