Yale BCI works with the brain's natural patterns, cutting training to under an hour
Using real-time fMRI, researchers showed people learn a brain-computer interface much faster when it is matched to each person's existing patterns of brain activity.
Researchers at Yale have shown that a brain-computer interface is much easier to learn when it is designed around the brain’s existing activity patterns rather than against them. The study, by Erica Busch, Smita Krishnaswamy and Nick Turk-Browne, was published in Nature Neuroscience.
The experiment
Healthy young adults lay in an MRI scanner while a real-time fMRI system read their brain activity every two seconds and turned it into movements of a video-game avatar.
The key idea is the neural manifold: each person’s brain tends to move through a limited set of activity patterns. When the BCI was mapped onto those natural patterns, participants learned to control the avatar in under an hour. The researchers note that traditional approaches can need up to ten training sessions, and roughly a third of users never gain control. Brain activity also reorganized during learning to better support the interface.
Why it matters
fMRI itself is too large for everyday BCIs. The principle, however, could apply to any decoder, including implanted and EEG systems. Personalizing the mapping to each brain could shorten calibration and help the many people who struggle to learn current non-invasive BCIs.
Learn more: Neural decoding · fMRI · Calibration