Brain Network Dynamics from Neuronal Avalanches for Longitudinal Motor-Imagery Decoding

Neuronal-avalanche connectivity features from source-reconstructed EEG in 20 BCI-naive participants enable prospective motor-imagery decoding, reaching 73.5 +/- 7.9% AUC in an unseen fourth session.

• Why it matters: Inter-session variability and repeated calibration limit BCI use outside expert-supervised settings, motivating methods that can anticipate future brain states before new calibration.
• What they did: Across four training sessions, right-hand motor imagery and rest EEG were converted into directed weighted networks from neuronal avalanches, then graph measures and PageRank-based ROI selection modeled learning trajectories to predict class-conditional distributions.
• The result: PageRank ROI selection improved AUC versus using all ROIs (73.5 +/- 7.9% vs 64.5 +/- 10.6%) and cut computational time by about 39%, supporting more autonomous prospective decoding.

AI summary of the abstract · may contain errors · not medical advice · read the paper.