The first pediatric exoskeleton, ready for use in children with paralysis
13 May 2021
Ten kilometers of strokes in the open sea to achieve hours of research against childhood cancer
14 May 2021
In the research, led by Stanford University in the USA, we worked with an individual who had two small sensors implanted in the area of the brain that controls the arms and hands, allowing him, for example, to move a cursor in a screen with mental commands launched at his own extremity.
scientists in United States have achieved, for the first time, decode the neural signals associated with writing, which has allowed an individual with multiple paralysis to use his mind to reproduce hand-drawn letters and phrases on a computer screen.
However, Nobody, until now, had focused on handwriting to explore the possibility of putting the neural signals evoked by the brain on paper, explains in a statement Frank Millet, the main author of this study, which is published this Wednesday in the magazine 'Nature'.
His team worked with a 65-year-old quadriplegic individual, called T5, who was implanted two small sensors, "the size of a baby aspirin," in the area of the brain that controls the arms and hands, allowing you, for example, to move a cursor on a screen with mental commands issued to your own limb.
Likewise, through the signals that the sensors captured from individual neurons when the subject thought that he was writing on a pad and with a pencil -both imaginary-, a machine learning algorithm He was able to recognize the patterns that his brain generated with each letter.
In this way, T5 managed to copy sentences and answer questions at a rate of 90 characters per minute, a speed similar to what a person of his age writes with a mobile phone.
This “brain-to-text” BCI device is so fast because each letter generates a highly recognizable pattern of activity, allowing the algorithm easily distinguish one letter from another, highlights.
During the multiple trials, the researchers presented T5 with sets of sentences and asked him to make a mental effort to "draw by hand" and without using capital letters compositions such as "I interrupted him, unable to remain silent" or "in thirty seconds the army had landed."
These exercises, repeated over time, improve the ability of the algorithms to differentiate the neural activation patterns that typify different characters, until achieving the interpretation of a letter "imagined" by T5 appeared on the screen after approximately half a second.
In subsequent sessions, they asked the subject to copy phrases unknown to the algorithms and he was able to generate 90 characters, about 18 words, per minute.
Later, T5 had to provide answers to open questions, which, the authors recall, force you to stop and think, and managed to generate 73,8 characters, almost 15 words, per minute, far surpassing the previous free composition record set in the 2017 study.
The authors also highlight that the error rate when copying sentences was at one error for every 18 or 19 characters attempted, while for free composition it was around one for every 11 or 12 characters.
However, when the experts resorted to an auto-correction function, similar to that offered by mobile phones, those error rates fell to 1% and 2%, respectively, quite low parameters compared to those of others. BCI, says Krishna Shenoy, co-author of the study.
"While handwriting can generate about 20 words per minute, we tend to speak at about 125 words per minute, and this is another interesting issue that complements writing. If we combine it, these systems could offer even more alternatives for patients to communicate effectively," adds Shenoy.
