Six held-out ds006253 SEEG patients had MOVE-vs-REST AUCs of 0.787–0.966; the exact group test reached its minimum p = 0.0156

We applied the unchanged author 70–150 Hz high-gamma amplitude (HGA) shrinkage-LDA decoder to six held-out patients in OpenNeuro ds006253. It distinguished MOVE (0–0.5 s after movement onset) from REST (0.5 s before stimulus onset), using the first 60% of trials for training and the last 40% for testing. The three newly analyzed patients had AUCs of 0.953 (sub-21), 0.851 (sub-07), and 0.787 (sub-06), each with permutation p = 0.0001. Across all six held-out patients (including sub-19, sub-14, and sub-03 from earlier work), mean AUC was 0.871 (range 0.787–0.966); the pre-registered exact one-sided sign-flip test gave p = 0.0156, the smallest possible p for n = 6, because all six AUCs exceeded 0.5. The development mean was 0.784 across seven patients; the held-out minus development difference was +0.087 (exact two-sided permutation p = 0.141; 1,716 partitions), providing no evidence of development-set optimism. Exploratory results were Spearman rho = 0.132 between AUC and channel count (n = 13), and PRE-MOVE-versus-REST mean AUC = 0.735, with 13/13 patients above 0.5 (sign-flip p = 0.00012). This was offline analysis, not a closed-loop test. MOVE and REST occur in different task phases, so task-phase signals may contribute to AUC; PRE-MOVE follows the stimulus and includes visual and decision activity. Patient order followed file size, and three planned patients were not analyzed within the run time. Only the author HGA derivative was used; HGA was not computed from raw SEEG.

stereo-EEG · high-gamma amplitude · shrinkage LDA · movement decoding · AUC · held-out evaluation · OpenNeuro ds006253