Researchers at UCSF's Weill Institute for Neurosciences have pioneered a "plug-and-play" brain-machine interface (BMI) powered by machine learning, aimed at restoring independence for individuals with paralysis.
Innovative solutions are emerging to enhance autonomy for people living with paralysis. For instance, the U.S. National Institutes of Health (NIH) supports exoskeleton technologies to aid children with cerebral palsy, easing physical limitations.
Brain-computer interfaces (BCIs) offer even greater potential through direct brain-to-machine communication. These systems are advancing rapidly, especially with machine learning integration. A landmark study published in Nature Biotechnology on September 7, 2020, by UCSF researchers details their latest progress.
Lead author Karunesh Ganguly, MD, PhD, notes the undeniable progress in BCIs, but highlights a major hurdle: daily recalibration requirements. This prevents leveraging the brain's natural adaptation, undermining long-term interface stability.
Current systems demand intensive daily training to enable brain-machine dialogue, yet they fail to learn from these sessions, limiting efficiency.
The new plug-and-play BMI addresses this by combining machine learning with electrocorticography (ECoG). ECoG uses electrodes placed on the brain's surface to capture stable, long-term neural signals—less precise than deep-brain electrodes but far more reliable.
Tested on quadriplegic patients, the interface detects imagined arm and wrist movements via ECoG signals, translating them into cursor control on a screen. Over days, machine learning enabled the system to adapt to user patterns, while the brain developed a parallel mental model for seamless interaction—mirroring a collaborative learning process.
Ongoing research at UCSF positions this BMI as a potential prosthetic extension, such as for hand or arm function, drawing from established neuroscience expertise.
