Beta power from high-gamma-selected SEEG channels added 0.023 AUC, but the primary test was not significant

We compared beta power (13–30 Hz) from high-gamma-selected bipolar SEEG channels with high-gamma power alone for offline cued key-press versus rest detection, using k = 4, 8, and 16 channels in 8 subjects. With k = 8, adding beta from the same channels increased mean AUC by 0.023 (median 0.021; 5/8 subjects; exact two-sided sign-flip p = 0.0625), which was not significant at alpha = 0.05. Selecting channels separately for high gamma and beta did not give a reliable gain over using the same channels (best difference +0.012 AUC, p = 0.24). The top-8 channel sets overlapped by 0–3 channels (median 1), but this did not translate into a reliable benefit from separate selection. With all channels, beta added 0.036 AUC (7/8; p = 0.023). The analysis used archived features and shrinkage LDA. It included 8 subjects, one session per subject, fixed offline task windows, and move-versus-rest decoding; secondary tests were not corrected for multiple comparisons.

stereo-EEG · brain-computer interface · beta power · high-gamma power · channel selection · movement decoding · shrinkage LDA · AUC