No one, not even José del R. Millán, was expecting Subject 1 to have a breakthrough when he did. The 26-year-old, who has tetraplegia with no mobility below the neck, had to be hospitalized part-way through Millán’s study for an unrelated complication, and he had gone through more than 20 training sessions to operate a brain-controlled wheelchair without much to show for it.
Yet, he surprised everyone including Millán when something clicked into place and he was able to control the wheelchair during his last training sessions. He even navigated it through a cluttered room in a German clinic with near-perfect accuracy.
“We would have never been able to predict the breakthrough for Subject 1,” Millán, a computer engineering and neurology researcher at The University of Texas at Austin, told The Daily Beast. “Everybody was super excited when we observed that kind of performance.”
Subject 1’s breakthrough is one of the major findings of a new study published on Friday in the journal iScience, that sheds new insight into the learning curve associated with technology that connects the human mind with machines—also known as brain-computer interfaces (BCI).
The study details the training trajectories and performance of three people with tetraplegia as they used a BCI wheelchair controlled with their minds. They wore a cap with electrodes on their head that allowed a computer to translate certain brain waves into commands for the wheelchair—the device would then move or turn based on the translated commands.
Subjects 1 and 3 got the hang of the wheelchair over time and navigated through the corridor handily. Meanwhile, Subject 2 did not improve over the course of training and could not steer the wheelchair as easily.
These differences extended beyond just driving skills. By the end of the training, the people who improved had produced brain signals that an algorithm could more accurately translate into commands for the wheelchair. They also more quickly produced a command over time. Additionally, connectivity—a measurement that reflects how well different regions of the brain can communicate—changed significantly over the course of the training for Subjects 1 and 3, while it did not change for Subject 2.
This variability is important for researchers to understand and to share, so as not to cherry-pick positive findings, Millán said.
“We wanted to record the case of Subject 2 to show that there is no magic bullet,” he said. “We need to have several options, and we also need to understand that the same intervention given to two people will not have the same effect.”
Most BCI studies thus far have looked at abled populations who are not the intended users of the assistive technologies in question, which is why it was critical to enroll participants who had disabilities into the study, Millán said.
“I think a major contribution is looking at people with tetraplegia, because it’s not easy to work with them,” W. Hong Yeo, a biomechanical engineering researcher at Georgia Tech who was not involved with the research, told The Daily Beast. In general, he added that recruiting participants for a BCI study can be difficult due to the time-intensive nature of the training process and people’s squeamishness with electrode caps and sticky adhesive gel. People with tetraplegia often have complex health conditions that can require hospital stays, and may not be interested in participating in exploratory or taxing research.
Millán hopes to study the root cause behind BCI learning curves to eventually speed up the learning process. By figuring out, for instance, what factors led to Subject 1’s breakthrough, he hopes that his research will lead to better assistive technologies for people with restrictive disabilities.