Rest high-gamma power weakly ranked responsive SEEG channels but did not improve decoder selection over random choice in eight patients

We tested whether task-free rest-window band power can select stereo-electroencephalography (SEEG) channels for movement decoding without motor calibration. We used archived features from OpenNeuro ds004473 v1.0.2 for eight epilepsy patients performing cued left/right key presses. Features were log mean high-gamma (70–150 Hz) and beta (13–30 Hz) power; rest windows were short inter-trial intervals, not a separate resting-state recording. A plan was fixed before analysis. Rest high-gamma power weakly ranked test-trial movement-responsive channels (mean Spearman rho 0.154; six of eight subjects positive; exact one-sided p = 0.023, Holm p = 0.047), but an 8-channel decoder selected by rest high-gamma had mean test AUC 0.632, similar to random 8-channel sets (0.626; p = 0.44). Channels selected using labelled calibration trials had mean AUC 0.785 (0.159 above random, p = 0.008; 0.153 above rest selection, p = 0.023). Other rest-power predictors and a leave-one-subject-out model did not rank channels reliably. Exploratory analyses suggested that 10–20 labelled trials may provide much of the calibration benefit; this requires confirmation in a prespecified analysis. The findings apply to these eight patients, one session each, short inter-trial windows, two frequency bands, and move-versus-rest decoding. They do not establish that rest band power can replace motor calibration for channel choice in other datasets.

stereo-electroencephalography · SEEG · brain-computer interface · channel selection · high-gamma power · motor calibration · movement decoding