Imagine playing Call of Duty with your friends using only your thoughts. Thanks to brain-computer interface (BCI) technology developed by researchers at the University of Texas at Austin, that could soon become a reality. A BCI reads brain signals and translates them into commands to a device. However, learning how to control a BCI is difficult, and users typically have to go through a lengthy tuning process to train the system to recognize their unique brain patterns.
Now, researchers have created a new approach that allows novice users to skip this tedious training and start using BCI right away. What is their secret? Borrow brain signal patterns from expert users.
The research results were published in a magazine PNAS Nexus.
In a typical BCI setting, users are asked to imagine a specific movement, such as moving their left or right hand. In doing this, BCI systems record brain activity using electroencephalography (EEG), which uses sensors placed on the scalp to measure electrical signals from the brain. The system then learns to recognize the patterns associated with each imagined movement.
The UT Austin team realized that this tuning process could be avoided if the BCI system was pre-trained on data from expert users who had already mastered the skills to mentally control the device. .
“If you think about this in a clinical setting, this technology eliminates the need for a dedicated team to go through this long and tedious calibration process,” said study author Satyam Kumar, a graduate student in the lab. he says. Jose del R. Millan, professor in the Chandra Family Department of Electrical and Computer Engineering at the Cockrell School of Engineering and the Department of Neurology at the Dell School of Medicine, said in a media release. “It’s a much faster transition from patient to patient.”
But there's a catch. Brain signals can vary widely from person to person, so a system trained on one person's data may not work well for another person. To overcome this hurdle, researchers have developed two techniques called Generic Recentering (GR) and Personal Assisted Recentering (PAR), which adapt an expert's brain patterns to match a new user's unique neural signature. We have developed two machine learning frameworks.
The GR framework continuously adjusts expert data to match novice users' brain signals without changing the underlying pattern recognition model. The PAR framework goes one step further and fine-tunes the model parameters based on a small sample of new user data.
To test these frameworks, the team hired 18 BCI novices and had them practice using the system over five sessions. Half used the GR approach and the other half used PAR. Participants operated a simple bar on the screen and played a car racing game using only imagined hand movements.
Novices were able to achieve significant control from the first session, demonstrating that the transplanted brain patterns provided a viable starting point. Over the course of their training, they continued to improve and their neural patterns evolved to the point that they were distinguishable from those of the original experts.
The neural patterns that participants learned to control the bar task were different from those developed for racing games, suggesting that the BCI can flexibly adapt to different situations. The researchers also found that the simpler GR framework performed similarly to the more complex PAR approach, indicating that extensive system retuning may not be necessary.
This study represents an important step toward making BCI more user-friendly and accessible. By leveraging the neural knowledge of experts, these transfer learning technologies lower the barrier to entry for new users, whether they are patients looking to regain lost motor function or gamers looking to improve their play. There is a possibility.
“On the one hand, we want to apply BCI in the clinical realm to help people with disabilities. On the other hand, we want to improve the technology to have a stronger impact on people with disabilities. “We need to make it easier to use,” says study author Millan.
As BCI technology continues to advance, it's tempting, especially for video gamers, to imagine a future where mind-controlled devices become part of everyday life. Through the power of shared neural patterns, that future may be closer than we think.
