A*STAR scientists have found a new way to give stroke patients instructive feedback during rehabilitation exercises aimed at restoring mobility.

For stroke patients, the long road to recovery starts with rehabilitation exercises, including visualizations in which people imagine enacting motions they are physically unable to perform. This ‘rewires’ their brains, and leads to real-life bodily improvements. The exercises only work if you’re performing them correctly, and since visualization takes place entirely in the mind’s eye, it has been difficult for clinicians and patients to know if the correct brain connections are being made.

To get a glimpse inside the brain, patients typically wear caps fitted with electrodes that track neural activity via electroencephalography (EEG) readings. Now scientists from the A*STAR Institute for Infocomm Research have validated a more accurate way of turning those EEG signals into clinically meaningful feedback for stroke sufferers.

“This feedback will help stroke patients in rehabilitation restore brain functions and improve motor recovery,” says study author, Kai Keng Ang, an A*STAR senior scientist who heads the Neural and Biomedical Technology Department.

Ang and his colleagues—including Cuntai Guan, formerly with A*STAR and now a professor at Nanyang Technological University in Singapore—previously ran a clinical trial in which stroke patients relearned hand grasping and knob manipulation through combined mental and physical training. Traditionally, patients have had to teach themselves over the course of several sessions to produce predefined EEG rhythms in their brains while performing a motor imagery exercise, but Ang’s team used a newer machine-learning strategy that automatically calibrates the EEG patterns associated with a particular mental exercise to an individual’s unique brain waves while engaged in that task.

This strategy helps correct for brain differences between individuals and avoids the hassle of multiple preparative sessions. However, it doesn’t account for the fact that individuals themselves can differ in their brain activity associated with a motor imagery exercise from one rehab appointment to the next.

It is this gap that is filled by the newest adaptive strategy. It works much the same as the machine-learning model, but it doesn’t stop at just one calibration session. It keeps updating its model of how the patient’s EEG patterns match motor imagery exercises as new information comes in after each and every session, thereby correcting for day-to-day changes in neural connectivity.

Ang and Guan have retrospectively applied the adaptive model to the clinical data from their earlier trial. As they report in their new paper, this would have significantly improved motor imagery detection in the patients. The researchers have also tested the adaptive strategy on 11 additional stroke patients in another clinical trial last year. “We are now analyzing the results and will be reporting the results soon,” Ang says.