When faced with a debilitating limb or spinal injury, many patients can use prosthetics to help recover some of their normal function. Dr. Jaimie Henderson is applying that same principle to the human brain. Alongside his brilliant team, Dr. Henderson is using Brain Computer Interface (BCI) technology to help patients who have suffered spinal cord injuries regain the ability to communicate. That alone is an impressive feat. However, Dr. Henderson isn’t just changing lives through the BrainGate study, he’s changing the way we understand neurobiology.
Assembling a neuroscience dream team
Jordan and Pippen. Jobs and Wozniak. Peanut butter and jelly. You can’t talk about one without the other. Now, the collaboration of Henderson and Shenoy may be joining that list (at least in the neuroscience community). Dr. Jaimie Henderson is the co-director of the Stanford Neural Prosthetics Translational Laboratory (NPTL) alongside his counterpart Prof. Krishna Shenoy, PhD. Together with the BrainGate Consortium, they are redefining what is possible for the aftercare and rehabilitation of patients with spinal cord injuries. Specifically, those who have lost the ability to communicate. Using Brain Computer Interface technology, their team can study handwriting representations in the motor cortex in order to accurately predict letters and display them on a screen.
What was needed to make this kind of technology work? First, Dr. Henderson implants two 100 by 100 arrays (a total of 200 electrodes with 196 active) into the hand motor cortex. Participants are then asked to write individual letters as the brain activity is recorded. From there they were able to determine that the velocity tuning of those neurons represented the velocity (instantaneously) of the imagined pen tip and that the letters could be reconstructed from those velocities integrated over time. Second, they needed to account for the fact that they have no idea how fast or slow a participant was writing the letters. That’s when staff research scientist Frank Willett came up with “time-warping”. He figured out how to stretch and shrink the neural data to fit the patterns that were expected so that independent of any behavior the participant was exhibiting, they would be able to read out the pattern of activity that represented an individual letter and display it on the screen.
The final obstacle the team had to overcome was being able to rapidly and accurately identify these large-scale neural patterns that represent an individual letter and display it on the screen in real-time. Thanks to Frank, they developed a multi-layered gated recurrent neural network (RNN) to solve this problem. The RNN takes in neural activity, processes it, learns the patterns of neural activity, and then outputs the highest probability character which is then displayed on the screen. We use this technology every day with things like Siri and talk-to-text. The problem is that those kinds of recurrent neural networks need tons of data and they receive it from millions of people every second of the day with cell phone use. The BrainGate team only has access to their participants for a few hours a session. In order to bypass this issue, Willett was able to augment the data so that they could use the small amount of information they had, but give it enough richness that the neural network would not stick to particular patterns and be more general. This breakthrough allowed for the RNN to predict known characters with 90% accuracy and original ones with 84% accuracy.
The real question is, is there anything this team CAN’T do?
The brain turned upside down
Well, not literally, but perhaps our understanding of how it works. We typically understand brain function through a topographic representation of different body parts and their correspondents along the precentral gyrus of the frontal lobe called the motor homunculus. Essentially, it’s an orderly progression as you move laterally down the motor strip. But Dr. Henderson and his team have made a most surprising discovery in their research with neural prosthetics and BCI technology. It turns out that the motor cortex is a very intermixed and non-somatotopic representation and that nearly every body part is represented in it. The implications of these findings suggest that by studying activity from this part of the brain, the data can be used to help with everything from exoskeletons to speech therapy for spinal injured patients.
Dr. Henderson shared with me his great excitement over the direction his field of research is heading. A diagnosis of paralysis is no longer a guarantee for low quality of life and care. Henderson and his peers are paving the way for patients to be able to regain control of their lives and the function of their bodies after injury. With growing corporate interest from companies like Elon Musk’s Neuralink, he believes that the possibility of in-home BCI solutions is only five to ten years away. I admit I share in his excitement and I can’t wait to see what happens next in the world of neural prosthetics.