In eight SEEG subjects, paired scores did not improve discrimination and baseline subtraction reduced AUC

We tested whether decoder scores from rest and movement windows in the same trial share an offset that could make paired comparisons or baseline subtraction useful for an offline, cued motor switch. We used archived features from eight subjects in OpenNeuro ds004473 v1.0.2: log high-gamma (70–150 Hz) and beta (13–30 Hz) power from adjacent-bipolar channels. A shrinkage LDA trained on trials 1–100 was tested on trials 101 onward. The primary contrast, paired accuracy minus unpaired AUC, had a mean of −0.012 (median −0.013; two of eight subjects above zero; one-sided exact sign-flip p = 0.95). Rest and movement scores were slightly negatively correlated (median Spearman rho = −0.071, range −0.151 to +0.005; p = 0.016), while consecutive-trial rest scores were uncorrelated (median rho = +0.010; p = 0.75). No subject exceeded the model’s baseline-subtraction break-even correlation of 0.5. A differenced detector reduced AUC in all eight subjects (mean change −0.068; median −0.071; p = 0.0078). These results do not support trial pairing or subtraction of a rest baseline about 2–7 seconds earlier for this task. The study was offline and limited to eight epilepsy patients, one session each, 0.5-second windows, and one shrinkage LDA decoder; it does not test slower drift over minutes.

stereo-EEG · SEEG · motor brain-computer interface · high-gamma power · beta power · shrinkage LDA · baseline subtraction · paired accuracy · AUC