In eight subjects, eight selected high-gamma channels did not improve 10-trial offline SEEG decoding

This offline analysis tested whether selecting eight high-gamma (70–150 Hz) channels during calibration reduced the number of cued key-press trials needed to decode movement from rest, and whether chronologically first trials performed worse than trials sampled across the session. It reused archived features from OpenNeuro ds004473 v1.0.2 for eight subjects and used trials 61–100 as a common test set. At 10 calibration trials, the mean AUC difference for eight selected channels versus all high-gamma channels was −0.010 (median −0.009); the selected decoder was higher in 4 of 8 subjects (exact two-sided sign-flip p = 0.35). The data therefore did not show that channel selection reduced calibration needs, and do not exclude differences of a few hundredths of AUC in either direction. The first 10 trials were not worse than 10 random trials from trials 1–60 (all-channel difference +0.008, p = 0.61; selected-channel difference +0.012, p = 0.47). In a post hoc comparison, adding beta power gave higher AUC than high-gamma alone in 6–7 of 8 subjects over most calibration sizes; this exploratory result has no multiplicity correction and requires preregistered confirmation on another dataset. The analysis is limited to eight subjects, one session each, known offline event windows, and a fixed late-session test set; it does not establish real-time performance or cross-day calibration needs.

stereo-EEG · brain-computer interface · motor decoding · calibration · high-gamma · beta power · shrinkage LDA · ROC AUC