A silicon chip implanted into the brain can measure signals and pass them through an algorithm designed by Stanford researchers with more speed and accuracy than ever before to control objects outside the body.
For example, a brain-implanted rhesus monkey can control a computer cursor with its thoughts. The old algorithm system allowed a monkey to move the cursor to contact a colored ball target 10 times in 21 seconds. With ReFIT, the new, Stanford-developed mathematical algorithm, the monkey’s accuracy more than doubled to 21 targets in 21 seconds.
Electrical engineering, bioengineering and neurobiology professor Krishna Shenoy, research associate Vikash Gilja M.S. ’10 Ph.D. ’10 and bioengineering doctoral candidate Paul Nuyujukian M.D. ’10 M.S. ’11 Ph.D. ’15 developed the new algorithm, which has drastically increased the performance of a system designed to translate the brain’s neural firings into cursor movements.
The system’s eventual goal is the restoration of lost functions to disabled patients through what is known as neural prosthetics. The team’s findings were published in a Nov. 18 article in the journal Nature Neuroscience.
While work in the subject has been ongoing for decades at other universities and hospitals, the Stanford team’s algorithm is a major breakthrough.
ReFIT was developed by Gilja and enables the cursor to be moved with more control and to stop at its destination rather than jerking erratically.
Like previous studies, Gilja’s approach uses a Kalman filter algorithm to interpret the data from the miniscule array of probes implanted into the subject’s brain. A model derived straight from that data produces a path that sees the cursor eventually reach its goal, but not without detours along the way.
According to Gilja, the “noise” in our brains is principally to blame for these detours. During different attempts at the same task, neurons will react in different patterns as determined by the direction and intensity of thought. In a hundred trials of directing the cursor, our thoughts would send it along a hundred slightly different paths.
The new algorithm removes the position-contingent variants, cleaning up the Kalman filter’s estimates by making assumptions of the subject’s intention. Because the researchers know the intended direction of the cursor, they correct each registered point to aim in that direction.
This strategy, which Gilja described as a “nuts and bolts engineering approach,” was able to markedly increase the system’s performance.
He emphasized that an interdisciplinary influence was crucial.
“It’s a rare Stanford opportunity to sit between engineering and more basic science research,” Gilja said. “I got exposed to both.”
Stanford has joined Brown University and Massachusetts General Hospital in an upcoming clinical trial of the system, called BrainGate2. The original BrainGate trial was published in 2006.
On a broader scientific level, the experiment is a key step in discovering more about movement and its neural basis.
“Neural prosthetics are a powerful tool to probe these neural dynamics that control movement, because they can act as a more direct proxy to the state of the brain than the arm,” Nuyujukian said.
By working in the field of neural prosthetics, the Stanford researchers can simultaneously examine the neurological basis of movement while pursuing their practical goals of restoring certain abilities to paralyzed patients.
“All of this is part of this huge enterprise in neuroscience that is blossoming everywhere,” Gilja said.