Researchers at the University of California, San Francisco, have given a paralyzed patient control of a robotic arm through a device that relays brain signals to a computer. He can grab, move, and put down objects simply by imagining himself making the movements. The device, known as a brain-computer interface (BCI), worked for a record seven months without requiring adjustments. Previously, such equipment typically only worked for a day or two.
BCI relies on an artificial intelligence model that can adapt to the tiny changes that occur in the brain when a movement is repeated - or in this case imagined movement - and learn to complete the movement in a more refined way. "This fusion of learning between humans and artificial intelligence is the next stage in brain-computer interfaces," said Karunesh Ganguly, Ph.D., professor of neurology and a member of the Weill Neuroscience Institute at UCSF.
"This is what we need to achieve complex, human-like functionality."
The research, funded by the National Institutes of Health, was published March 6 in the journal Cell. The key was finding that brain activity changed day-to-day as participants repeatedly imagined specific actions. Once the AI is programmed to take these changes into account, it can continue to work for months.
Ganguly studied how patterns of brain activity in animals represent specific actions and found that these representations changed day by day as the animals learned. He suspects the same is true for humans, which is why their BCIs quickly lose the ability to recognize these patterns.
Ganguly, along with neurology researcher Dr. Nikhilesh Natraj, worked with a study participant who had been paralyzed by a stroke many years ago. He was unable to speak or move. Researchers implanted tiny sensors on the surface of his brain that captured brain activity as he imagined movements.
To see if his brain patterns changed over time, Ganguly asked participants to imagine moving different parts of the body, such as hands, feet or head. Even though he couldn't actually move, the participant's brain could still generate movement signals when he imagined himself making the movements. The BCI records representations of these movements via sensors on the surface of the brain.
Ganguly's team found that the shape of these representatives remained the same, but their positions changed slightly from day to day.
Ganguly then asked the participant to imagine making simple movements with his fingers, hand or thumb for two weeks while sensors recorded his brain activity to train the artificial intelligence. The participants then tried controlling the robotic arms and hands, but the movements were still imprecise.
So Ganguly had the participant practice on a virtual robotic arm that provided him with feedback on the accuracy of his imagination. Eventually, he succeeded in getting the virtual robotic arm to move as he wished.
When the participant started practicing with a real robotic arm, it only took him a few practice sessions to transfer the skills to the real world. He can make a robotic arm pick up blocks, turn them and move them to new positions. He was even able to open a cupboard, remove a cup and hold it up to the water fountain.
Months later, the participant was still able to control the robotic arm after 15 minutes of "adjustment" to changes in his movement reps since he started using the device.
Ganguly is currently improving the artificial intelligence model to make the robotic arm's movements faster and smoother, and plans to test BCI in a home environment. For people with paralysis, being able to feed or drink on their own can be life-changing. Ganguly believes this is achievable.
"I'm very confident that we've learned how to build this system and we can make it work," he said.
Other authors include Sarah Seko and Adelyn Tu-Chan of the University of California, San Francisco, and Reza Abiri of the University of Rhode Island. The research was funded by the National Institutes of Health (1DP2HD087955) and the UCSF Weill Neuroscience Institute.