Beta and high-gamma channels differ, but separate beta channels did not significantly improve decoding in 8 SEEG subjects

In an offline cued key-press task with eight SEEG subjects, we tested whether beta power (13–30 Hz) should be read from its own best bipolar channels or from channels selected for high-gamma power (70–150 Hz). We ranked channels by the paired absolute t statistic for movement versus rest on trials 1–100, then compared shrinkage LDA decoders on later trials. The HG and beta top-eight channel lists shared a mean of 1.6 channels (range 0–3; chance 0.6), and their across-channel rank correlation was low (mean Spearman rho 0.14, p = 0.094; Holm p = 0.19). The primary comparison found no significant AUC improvement when beta was read from its own top-eight channels rather than the HG top-eight channels: mean difference +0.014 (median +0.011; 6/8 subjects; exact two-sided sign-flip p = 0.125). Beta on the HG top-eight channels increased AUC by +0.031 over HG alone (7/8); beta on its own channels increased it by +0.046 (8/8). At 16 channels, the own-channel decoder exceeded HG on the HG top-16 by +0.045 AUC, but this did not pass the Holm correction (p = 0.094). These offline results support reading both bands from HG-selected channels for this task, while not showing that separate beta channels are unnecessary. With one session and eight subjects, the null primary result does not establish equivalence; two subjects had large gains from separate beta channels. The top-eight rankings are noisy, B used 13–16 channels while A used 8, and the analysis did not use electrode positions.

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