An unchanged MOVE-vs-REST high-gamma SEEG detector detects pre-movement activity in 15 patients

Can a detector trained on high-gamma activity (HGA) during movement (MOVE; 0–0.5 s after movement onset) versus rest before the stimulus (REST; −0.5–0 s before stimulus onset) detect the 0.5 s before movement onset (PRE)? In an offline, event-locked analysis of 15 SEEG patients from OpenNeuro ds006253, we applied the unchanged shrinkage LDA detector to PRE and REST using a time-ordered 60/40 calibration/test split. PRE-versus-REST transfer AUC averaged 0.751 (median 0.750); all 15 patients were above chance, and 14/15 had per-patient permutation p < 0.05. MOVE-versus-REST AUC averaged 0.833. Median retention of the MOVE discriminant gain was 0.82 (bootstrap 95% CI 0.47–0.91). For test trials with reaction time ≥1 s, where PRE began at least 0.5 s after the stimulus, transfer AUC averaged 0.800 (15/15 above chance; Holm-adjusted p = 9.2×10⁻⁵). A detector trained specifically on PRE exceeded transfer by a mean 0.033 AUC (13/15; Holm-adjusted p = 0.013). At a threshold set to 10% false alarms on REST windows, mean hit rates were 0.43 for PRE and 0.58 for MOVE. The retention test was added after the preregistered plan. These findings show offline detection in fixed event-locked windows; they do not establish continuous asynchronous detection. PRE may include late motor preparation and small movements below the mouse threshold. Because REST precedes the stimulus and PRE follows it, task state may contribute to the difference; the reaction-time control addresses only the early stimulus response. There was one session per patient, and movement onset was taken from the dataset authors’ mouse data.

SEEG · high-gamma activity · movement detection · brain-computer interface · shrinkage LDA · event-locked analysis · motor preparation · OpenNeuro ds006253