Two additional ds006253 patients passed movement detection; stimulus-window scores exceeded rest in 12 patients

We ran the unchanged 70–150 Hz MOVE-vs-REST shrinkage LDA detector on two ds006253 SEEG patients not used in earlier SCI work (sub-06 and sub-10), using all channels and a time-ordered 60/40 train/test split. The held-out AUCs were 0.787 and 0.823 (both permutation p = 0.0001), within the 0.569–0.965 range of 10 earlier patients. A group test over the two new patients was not possible (n = 2; minimum sign-flip p = 0.25). The same decoder scored the 0–0.5 s post-stimulus window above pre-stimulus rest across 12 patients (mean AUC 0.648, exact one-sided sign-flip p = 0.00049; 8 of 12 individual tests had p < 0.05). MOVE scored above POST in all 12 (mean AUC 0.717). This indicates a stimulus-related false-alarm tendency for a detector trained against rest alone. The study was offline, used one session, decoder, and split per patient, and used author-provided HGA only. The two new patients were selected by file size, not at random; REST and MOVE came from different task phases, and eye movement or preparation could occur in POST.

stereo-EEG · high-gamma activity · movement detection · brain-computer interface · shrinkage LDA · AUC · stimulus response · OpenNeuro