The Press Stories
In an exciting new study, scientists have succeeded in transforming the function of the brain into written text.
It can have a variety of causes, such as anarchy, or loss of speech, stroke, or amyotrophic lateral sclerosis. People with anorexia often retain their language ability, but stroke can prevent them from using technical assistance to communicate. Neuroscientists have used machine learning to translate the electrical activity of a part of the brain into words and sentences. For the first time, scientists have succeeded in translating into text the function of the brain of a paralyzed person.
Neuroscientists matched a sequence of electrodes the size of a credit card to the 36-year-old’s sensorimotor cortex, the part of the brain that controls speech pronunciation. Scientists led by the University of San Francisco, California, used deep learning methods to train participants’ computer models that could identify and classify words from patterns of brain activity. They asked the person to think about the words they had given him on a screen.
In addition to the deep learning models, they used a “natural language model”, a program that predicts the next word in a sentence based on the previous words.
Loss of comprehensible speech
The 36-year-old participant at the beginning of the study suffered a stroke in his brain when he was 20, resulting in a severe stroke and loss of intelligent speech known as anarthria. Other causes of anorexia include amyotrophic lateral sclerosis, a rare neurological disease that primarily affects the nerves that cause voluntary movement.
The person in the study could murmur and lament, but was unable to express words despite healthy cognitive functions. With small head movements, he can control the computer-based typing device to communicate. However, his typing speed on this device is only five perfect words per minute.
After 48 training sessions of the new “mind-reading” technology over a period of 81 weeks, he was able to generate approximately 15 words per minute, with an error rate of 26%.
Scientists who have published their findings in the New England Journal of Medicine say that speech decoding technologies are generally considered useful if their error rate is less than 30%. This study reflects a revolutionary breakthrough in the field of brain-computer interfaces. Although the process is very aggressive, brain surgery is required to attach a recording tape to the surface of the brain, and the accuracy of speech transition from thought to normal, the paradigm is revolutionary.
Brain function data
During the 48 training sessions, the participant attempted to generate specific words from a set of 50 words. In each attempt, one of these words was given to him on the screen. When the word turned green after a 2 second delay, he tried to speak it. During these sessions, the researchers collected 22 hours of brain activity data, which were fed into an in-depth learning process. In two final extra sessions, the man tried to use previously trained words to form phrases such as “I’m thirsty” and “I need my glasses”. In each test, the participant was presented with a phrase and tried to form words quickly while thinking about what to say about them.
The deep learning algorithm and language model decoded his thoughts in more than half of the sentence tests.
proof’s
Neuroprosthesis for decoding the speech of a person with paralysis
* The information and services available at pressesante.com do not in any way alter the advice of competent health professionals.
Do you like our content?
Receive our latest publications for free and directly in your mailbox daily
[ADINSERTER AMP]
