AI Brain Computer Interface transforms mental handwriting into text


Brain Computer Interface (BCI), also known as Brain Machine Interface (BMI), converts brain activity into performance, thus enabling the loss or impairment of functions such as movement and language. BCI is used to help people with severe paralysis or communication problems due to stroke, spinal cord injury, amyotrophic lateral sclerosis (ALS) (also known as Lugeric’s disease), and other medical conditions. I came. New research recently published nature Led by researchers from Stanford University demonstrates a new paradigm for brain-computer interfaces Artificial Intelligence (AI) decodes brain activity in real time into text by taking into account handwritten movements.

The lead author of this study is Dr. Frank Willett, a researcher at the Howard Hughes Medical Institute (HHMI) at Stanford University, and co-authored by Krishnachenoy, an HHMI researcher and Professor Lee R. at Stanford University. Including doctor. .. Donald T. Avansino, Senior Lecturer in Hochberg, MD, PhD and Neurology at Harvard Medical School and Researcher at Stanford University. The research was funded by the National Institute of Health’s Advancing Innovative Neurotechnologies® (BRAIN) initiative, the National Institute of Neurological Disorders and Stroke (NINDS), and the National Institute on Deafness and other Communication Disorders (NIDCD).

Mental-Handwriting-to-Text: BCI’s New Paradigm

“So far, the main focus of BCI research has been on restoring general motor skills such as grasping and grasping and pointing and clicking with a computer cursor,” the researchers write. I am. “But a quick succession of very nifty actions like handwriting and touch input can enable faster communication speeds. Here we are developing an intracortical BCI that will decipher attempted handwriting movements. Have. neural Converts to text in real time using activity in the motor cortex and a recurrent neural network decoding approach. “

In this study, a 65-year-old right-handed man was equipped with two 96-microelectrode cortex arrays to record neural signals. Specifically, a NeuroPort â„¢ array from Blackrock Microsystems with 1.5 mm electrodes was placed in the precentral gyrus region of the left hemisphere of the study participants’ brain. Participants had suffered a spinal cord injury nine years prior to inclusion in the study and had very limited voluntary limb movements.

The researchers used software developed by MATLAB and Simulink to manipulate recorded data and decode it in real time. Brain activity data was collected when participants were asked to try handwritten text over the course of multiple sessions. The decoder was trained with session data.

AI Deep Learning: Recurrent Neural Networks (RNN)

In this study, we used a two-tier Gated Recurrent Unit Recurrent Neural Network (RNN) to convert participants’ brain activity into time-series character probabilities and trained the decoder with Forced Alignment Labeling. RNNs were trained to predict character probabilities from a one-second delay, taking system processing time into account.

Artificial intelligence, recurrent neural networks are a class of artificial neural networks that are widely used when contextual data is needed to make decisions based on natural language processing, speech recognition and input data. Artificial neural networks are somewhat inspired by the biological brain, whose architecture consists of layers of interconnected nodes called artificial neurons. RNN’s deep learning algorithms can process sequences of inputs using internal states that work as follows: memory when the inputs are linked. Relationships are “remembered” while the recurrent neural network trains. RNNs are useful in scenarios where you need to model nonlinear temporal or sequential relationships.

“This BCI paralyzed our hands due to spinal cord injuries and achieved a typing speed of 90 characters per minute online with 94.1% raw accuracy and over 99% offline with the universal autocorrection. We have achieved accuracy, ”said the researchers. Report. “As far as we know, these input speeds exceed those reported by other BCIs and are comparable to the typical smartphone input speeds (115 characters / minute) for people in the age group of the participants. “

Copyright © 2021 Cami Rosso All rights reserved.


Leave A Reply