In 15 SEEG patients, a movement detector responded after stimulus onset; post-stimulus negatives improved specificity but reduced rest AUC

We assessed whether a published movement-versus-rest SEEG detector responds to the post-stimulus period before movement, and whether post-stimulus calibration negatives improve specificity. Using archived 70–150 Hz HGA window features from 15 patients, we reproduced the held-out MOVE-versus-REST mean AUC of 0.833 (maximum per-patient difference 0.0005 from rounding). The standard detector separated POST from REST with AUC 0.661 (95% CI 0.601–0.724); 14/15 patient AUCs exceeded 0.5 (two-sided sign-flip p=0.0003). MOVE-versus-POST AUC was 0.726, 0.107 below MOVE-versus-REST (95% CI 0.058–0.161; 13/15; Holm p=0.0005). At a threshold set for 10% false positives on REST, 27% of POST windows were positive. Adding eligible training POST windows to the negative class increased MOVE-versus-POST AUC by 0.035 (95% CI 0.014–0.065; 14/15; Holm p=0.0004), while POST false positives fell to 19%. MOVE-versus-REST AUC decreased by 0.017 (95% CI 0.004–0.033); non-inferiority at a 0.02 AUC margin was not shown (Holm p=0.35). The exploratory 1.5 s latency rule showed the same pattern: H1 +0.030 (13/15; Holm p=0.0020), H2 0.105 (Holm p=0.0016), and H3 −0.010 (p=0.080, not shown). About one third of the published MOVE-versus-REST separation was not specific to movement. The result supports calibrating with no-movement windows from all task phases, with a small measured sensitivity cost against pre-stimulus rest. This was an offline analysis using event-aligned windows, one session and one time-ordered split per patient. POST trials may differ from other trials; detector B had more training windows and no size-matched control was tested; mouse-trace onset may miss small earlier movements; and some patients had few eligible POST test windows.

SEEG · brain-computer interface · movement detection · high-gamma activity · specificity · shrinkage LDA · post-stimulus state · calibration