Wrist vibration and software that learns alongside the user helped beginners control a cursor with brain signals sooner
Wrist vibration and software that learns alongside the user helped beginners control a cursor with brain signals sooner
In a study of 31 people, the system trained users to imagine movement while simultaneously recalibrating the software that interpreted their EEG signals. Over several sessions, this combined approach produced a larger improvement in accuracy than either conventional training or vibration alone.
On July 15, a Carnegie Mellon University team published a study in Nature Communications on how to train a brain-computer interface, a system that reads electrical activity in the brain and converts it into a computer command.
During the experiment, each participant watched a cursor and imagined moving either the left or the right hand. An EEG headset recorded weak electrical signals from the scalp that reflected activity in the sensorimotor cortex, and the software used those signals to move the cursor. A beginner cannot easily produce a stable signal that the software can distinguish from noise. The person and the software must learn at the same time: the algorithm adjusts its classification rule while the user searches for a reliable way to control it.
In April, a bidirectional brain-computer interface restored an artificial sensation of stepping in one participant by stimulating the sensory cortex. In that case, the feedback was designed to accompany walking. In this study, wrist vibration served as a training cue while the participant learned to control a cursor.
The researchers gave the user and the algorithm a shared cue. A small motor on the wrist produced a brief vibration during training trials. After each block, the algorithm was updated to give more weight to trials in which the brain signal already distinguished the intended direction well. The vibration helped the user find a reproducible signal, while the algorithm adapted specifically to that signal.
The main group included 15 people. Another eight received vibration, but the software did not prioritize signals that were easier for the user to learn. Eight others used a conventional system without vibration. Between the first trial and the second session, control accuracy in the main group increased by 19%. The group that received vibration without this signal selection improved by 6,3%, while the conventional group improved by 3,0%. In a two-dimensional task, in which the cursor moved across the screen in different directions, mean continuous-control accuracy reached 66,9%.
A control condition with continuous vibration and another condition in which the software was updated without the combined approach showed that the result depended on the interaction between its two components. In six participants from the main group, the advantage persisted for more than two months after the vibration was removed.
For now, this is a laboratory experiment in healthy young people, not a demonstration of restored movement in people who have had a stroke or spinal cord injury. In these patients, the sensorimotor cortex is often reorganized, and sensory pathways may be impaired.
How quickly a person learns to use an EEG-based brain-computer interface depends on whether the user’s learning process and the algorithm’s updates are aligned.