A new algorithm automatically figures out which brain and muscle signals actually talk to each other during rehab exercises.
Researchers built a framework that treats EEG-EMG channel pairing as a constrained bi-objective optimisation problem, solved with the genetic algorithm NSGA-II to find channel pairs that both target motor cortex regions and show strong corticomuscular coupling. They combined correlation features between EEG-EMG band power with ERD-based EEG features, using sliding-window analysis to track how motor imagery signals change over time. Tested on motor imagery data from eight stroke patients, the system reached an average classification accuracy of 89.6%.
Hybrid EEG-EMG brain-computer interfaces get pitched for stroke rehab because EEG alone is noisy and inconsistent from person to person. But manually choosing which channels to pair has never generalized well across patients. Automating that selection with a proper multi-objective search, instead of a fixed template, is the unglamorous plumbing work that decides whether a system works on patient nine, not just patients one through eight.
Eight patients is a proof of concept, not a clinical result, and a 89.6% accuracy figure from a lab dataset says little about how the same setup holds up on a hospital rehab floor with electrode drift and less cooperative subjects.