Beta power improved later-trial key-press detection on eight high-gamma-selected channels in eight SEEG subjects

We tested whether 13–30 Hz beta power improves key-press detection when added to 70–150 Hz high-gamma (HG) features from a limited number of bipolar SEEG channels. Using archived features from eight subjects in OpenNeuro ds004473 version 1.0.2, channels were ranked by the absolute paired Welch t statistic for HG on trials 1–100. Shrinkage LDA was trained on trials 1–100 and tested on trials 101 onward. For the prespecified top-eight contrast, adding beta increased AUC by a mean of 0.0313 (median 0.0158) in seven of eight subjects (exact one-sided sign-flip p = 0.0078). Mean gains ranged from 0.0316 to 0.0365 for 4, 16, 32, and all channels; there was no trend with channel count (Spearman rho = 0.138, p = 0.656). Median AUC for HG plus beta on four channels (0.852) was near HG on all channels (0.850), but the top-eight HG-plus-beta versus all-channel HG contrast was not significant (p = 0.117); equivalence is not established. Three subjects with low HG AUC contributed most of the top-eight mean gain. This was an offline, cued key-press versus rest analysis in eight epilepsy patients, one session each. Channels were selected by HG only; there was no online or closed-loop test. The earlier all-channel test trials had been used in prior analyses, while the top-K HG-plus-beta values are new.

SEEG · beta power · high-gamma · brain-computer interface · key-press detection · AUC · shrinkage LDA · channel selection · trial transfer