Adding beta power improved offline key press detection in an eight-subject SEEG replication

In eight subjects performing an offline cued key-press versus rest task, adding 13–30 Hz beta power to 70–150 Hz high-gamma power increased shrinkage LDA AUC when trained on trials 1–100 and tested on trials 101+: mean subject-level gain +0.033, positive in 6 of 8 subjects, exact one-sided sign-flip p = 0.020. The gain also appeared with the reversed time split (train on the last 100 trials, test on earlier trials): +0.029, positive in 8 of 8, p = 0.0039. Forward and reversed gains did not differ (mean difference +0.004, two-sided p = 0.88). The original analysis reported a forward gain of +0.040 (6 of 8, p = 0.0156); a scikit-learn cross-check here reproduced its per-subject values, while the new pooled-shrinkage implementation gave a mean gain 0.007 lower with the same conclusion. Subject 12 had the largest forward gain (+0.098) while high-gamma alone was at chance (AUC 0.496); its reversed-split gain was +0.005. These findings are limited to eight subjects, one session, archived features, and overlapping forward and reversed test sets; no new signal processing from raw data was performed.

stereo-EEG · SEEG · beta power · high-gamma power · key-press detection · AUC · shrinkage LDA · replication · offline classification · OpenNeuro ds004473