A Methodological Improvement to Brain-Computer Interfaces

I-LABS researchers are discovering ways to make BCIs a practical treatment tool for brain injuries.

Photo caption: Mark Wronkiewicz, a UW neuroscience graduate student, in his office at I-LABS.

Imagine sitting in your chair and raising the window blinds by simply thinking about it. Or speaking by visualizing the words but not actually saying them. It would happen by a computer interpreting your thoughts and then acting on them.

"The computer interprets what the brain wants to do, but can't actually do," said Mark Wronkiewicz, a UW neuroscience graduate student working with I-LABS' Adrian KC Lee. "If the brain can't cause the hand to move, for instance, we could instead reroute those signals to cause a computer to move the hand."

Brain-computer interfaces, known as BCIs, could one day be used as a treatment for a range of brain-related disorders, including spinal cord injuries, stroke and degenerative diseases such as ALS (also known as Lou Gehrig's disease).

Right now they're seldom used as a medical treatment because they're impractical, but Wronkiewicz studies how to improve BCIs in hopes of making them useful in the clinic.

In a study published over the summer in the Journal of Neural Engineering, Wronkiewicz and co-authors Lee and I-LABS' Eric Larson identify a way to make BCIs more efficient. The work was funded by the National Science Foundation, the U.S. Department of Defense, and the National Institutes of Health.

In the Q&A below, Wronkiewicz explained his research:

Q. What got you interested in BCIs?

A. The story behind this is simple (and a little serendipitous). During my very first engineering class as an undergraduate student at Washington University in St. Louis, I decided to write our research report assignment on the topic of BCIs because of a cool diagram I saw in a textbook. The illustration elegantly showed the primary methods of recording brain signals, which were then processed on a computer and used to generate signals to control various output devices like a robotic hand or brain-controlled wheelchair. From that moment, I was hooked on the subject and was fortunate enough to be accepted into Dr. Eric Leuthardt’s BCI-focused lab the next year, and eventually receive an NSF Graduate Fellowship Research Program award to pursue this work here at UW.

Q. What is the goal of your research?

A. I’m working to make BCIs a useful treatment tool in clinical settings. One major issue is that these systems are time consuming to calibrate. Typically, 20-30 minutes is required each and every time a person wants to use a BCI. Time and money are both important, especially in the clinic, so this calibration issue represents a real roadblock for the field.

Imagine you had to wait 20-30 minutes for your computer to boot and you could only use it for 2 hours at a time. That’s essentially the practicality issue we’re addressing with this study.

In our study, we developed a method to almost eliminate that calibration time and overcome this obstacle.

Q. What did your study do to improve BCIs?

A. We used an approach called "transfer learning," which aims to speed up training by recycling brain data recorded from other subjects in previous recording sessions. But since everyone's head anatomy is slightly different, this approach isn't precise enough to be useful for a brain prosthetic.

To overcome this problem, we explored whether a "source imaging" tool would help. Using this method, we took brain recordings from multiple people, combined them with a structural map of each person's brain that we got from MRI scans, and then through a series of calculations we were able to estimate how any one person’s brain activity will manifest on a different person’s brain.

Q. That's kind of sci-fi – why is this significant?

A. It's exciting because instead of needing to collect calibration measures for each person, we can recycle existing training trials. Our study showed that transferring calibration trials from 7 or 8 subjects was enough to completely replace the trials currently collected during that 20-30 minute calibration session.

Q. How are your BCI methods different from others?

A. We’re using neuroscience to push the BCI field forward, which is in contrast to many groups focused exclusively on engineering concepts. In the source imaging method that we used, we take brain recordings from electrodes on the surface of the scalp (Figure 1a, black dots) to estimate what’s happening at many points on the surface of the brain (Figure 1b, yellow dots). Importantly, this avoids the need to undergo brain surgery to physically place electrodes on the brain.

Q. What is the advantage of source imaging?

A. Source imaging allows us to leverage scientific findings from many other areas of neuroscience (e.g., MRI studies, neuroanatomy, and even other BCI studies) in BCIs. This permits us to take advantage of knowledge developed by the neuroscience community as a whole and should provide a valuable complement to the advancements in signal processing and machine learning made by engineers.

For example, previous research in our lab has found a specific area in the brain involved in switching of auditory attention, as you would do if you were listening to the radio and then realize your friend has something important to say. In this study, we used that neuroscience knowledge to focus on brain signals coming from this area (called the right temporoparietal junction) and predicted based only on brain activity when subjects switched their attention to listen to another sound. In the future, this might be useful for dynamically tuning hearing aids to track a specific sound or voice instead of amplifying everything.

Q. When do you think BCIs will be available as a treatment?

A. This is the biggest question facing the field. Damage to the nervous system can be traumatic as in stroke or spinal cord injury, but it can also be the result of a slow degenerative disease like ALS. Because of this variability, it’s unlikely that we’ll have a major breakthrough and be able to treat all neural problems at once.

Instead, it’s more likely that the field will make a breakthrough in rehabilitating one specific type of injury or disorder, and we’re still searching for that first big breakthrough. If I had to guess, I would say we’re still 10-15 years away from BCIs being a standard clinical rehabilitation technique from any major form of neural damage. The silver lining is that it’s still a relatively untouched frontier with plenty of room for us to develop new theories and techniques.