Decoding brain signals to build a better brain-computer interface

Mary Guiden

QWOP, an online game that challenges you to help a man runJeremiah Wander likes to use the online running game known as QWOP as an analogy for his research. 

In QWOP, you control an athlete using the Q, W, O and P keys on a keyboard to move a man’s calves and thighs. The game is incredibly challenging; when most people run or walk, it just happens without thinking about it.

Wander studies signals that people don’t normally control, with the aim to use those signals to eventually create an action, like making a hand move.

Earlier this year, Wander received his PhD from the University of Washington’s Department of Bioengineering. He is now employed by Microsoft Research, where he is working on new devices, algorithms and software for recording physiological signals—things like heart rate—and using those signals to help people get healthier.

While at the UW, Wander worked on decoding brain signals to design a better brain-computer interface (BCI), a device that can potentially be used to replace or restore function in people with severe motor disorders. His work is integral to and was supported by the Center for Sensorimotor Neural Engineering (CSNE).

Studying brain signals via ECoG

While a graduate student, Wander worked in the labs of CSNE Director Rajesh Rao, PhD, and Jeff Ojemann, MD, a neurosurgeon. Through the Ojemann Lab, Wander studied the brain signals of people with epilepsy who are implanted with electrocorticography (ECoG) electrodes. The ECoG device or grid is implanted in preparation for surgery to better control seizures. People with epilepsy volunteer to help with this research during their hospital stay.

Wander’s research included looking at how the brain responds when someone is learning how to control a BCI. He said that when people perform a particular task—like controlling a cursor on a screen—the brain seems to tap into a network of areas across the brain akin to the areas activated when someone learns how to type or ride a bicycle.

At first, the person who learns the skills is deliberately focused on controlling the BCI. But after some time, the skill becomes as easy as jumping on a bike they’ve known how to ride for years. Findings from this research, “Distributed cortical adaptation during learning of a brain-computer interface task,” were published in PNAS in April 2013.

Populations of neurons, how do they connect?

From there, Wander and a team of researchers then decided to take a closer look at the areas that were active during BCI use to better understand how the brain works as a system.  “We looked at the way all areas of the brain were reacting with one area controlling the BCI,” Wander said. “We were using a BCI to try to better understand the models of connectivity among big populations of neurons.”

Hierarchical design approach for BCIs

In research that stems from work in the Neural Systems Lab (led by Professor Rao), Wander also talked about the use of a new adaptive hierarchical structure for a BCI, which allows a user to teach the BCI or system new and useful skills on an ongoing basis. (He was not involved in this research.)

Low-level actions—move left, right and stop—are first learned and later carried out using a higher level or broader goal, such as “go to the kitchen” when controlling a robot. This makes it easier for the user, who no longer has to repeat low-level commands. Findings related to this new architecture (“A hierarchical architecture for adaptive brain-computer interfacing”) were published by Rao and others in the Proceedings of the International Joint Conference on Artificial Intelligence in 2011.

Wander said the research team continues to study several hierarchical versions for BCIs. “They now have a much better idea of where and when there are strong representations of the higher-level goal signals,” he said.

He’s been surprised by the wide variation researchers have uncovered, which also creates challenges for ongoing research. “There are differences for everyone that uses the BCI. It’s not just in the activity patterns in brains, but also in the mental approach to the task.” 

The young scientist remains fascinated by BCIs in general. “They provide a neat opportunity, not just to try and build something to improve the lives of patients, but also the opportunity to make the nervous system jump through some hoops, and record how the brain solves that problem. It has been so interesting to see how the brain adapts and adjusts to almost any problem we throw at it.” 

Next steps for BCI research

What’s next? Wander said the team will start building systems that try to harness the information they’ve uncovered to date. “When we tried to determine what someone was going to do before they did it, there was a strong indication that they were using signals outside of the primary motor cortex, and that means different things,” he said. The motor cortex is the part of the brain that is most involved in voluntary movements.

Wander found in one of his experiments, for example, a response in the secondary motor cortices when someone tries to learn a BCI skill.

Next-generation BCI architectures will include signals from multiple cortical regions to allow for more robust device control strategies.

Researchers will also focus on building systems that can adapt to each person’s response to BCI tasks, leading to co-adaptive BCIs. “Your strategy will be different from my strategy, so we need to address that in constructing these systems,” Wander said.

For a more detailed graphic of the effects of motor, sensory and bidirectional BCIs on brain function, see "Brain-computer interfaces: A powerful tool for scientific inquiry," published in Current Opinion in Neurobiology by Wander and Rao in 2014.