Movement detection reached AUC 0.953 in one new ds006253 SEEG patient, and post-stimulus scores exceeded rest in 10 of 11 patients

We applied an unchanged 70–150 Hz movement-versus-rest detector to sub-21, a patient from OpenNeuro ds006253 not used by earlier SCI publications. In a time-ordered held-out test, the MOVE-versus-REST AUC was 0.9525 (permutation p = 0.0001), above the range of seven archived patients (0.5688–0.9447). One of eight planned patients finished this run; the new-patient group test therefore had n = 1 and exact sign-flip p = 0.5. Combined with three new patients from earlier work, 4/4 were above chance (mean AUC 0.8968; exact p = 0.0625, the minimum possible for n = 4). The same detector scored the 0–0.5 s post-stimulus window above pre-stimulus rest in 10/11 patients (mean AUC 0.6857; p = 0.0009766). In an exploratory subset with reaction time ≥0.5 s, where that window contained no movement, 11/11 were above chance (mean AUC 0.6810; p = 0.0004883). At the threshold detecting 80% of movements, the median fraction of post-stimulus windows above threshold was 0.625 versus 0.2533 for rest. These offline, single-session results use the authors’ HGA derivative; the descriptive 11-patient result includes previously used patients and is not fully out of sample. Only one new patient finished, and post-stimulus windows with reaction time <0.5 s can include movement.

SEEG · movement detection · high-gamma activity · brain-computer interface · false alarms · post-stimulus activity · time-ordered validation · OpenNeuro ds006253