Hand-knob coverage did not predict offline movement-detector AUC in 15 SEEG patients

This pre-registered analysis asked whether stereoelectroencephalography (SEEG) implant geometry, known before calibration, predicts offline movement detection from 70–150 Hz high-gamma activity. In 15 patients from OpenNeuro ds006253, the held-out area under the ROC curve (AUC) came from a Ledoit–Wolf shrinkage linear discriminant analysis using all bipolar pairs with coordinates to distinguish movement (0–0.5 s after onset) from pre-stimulus rest. Distance from the nearest pair midpoint to a fixed MNI hand-knob point was not associated with AUC (Spearman rho = +0.05, one-sided permutation p = 0.58). The number of pairs within 20 mm (rho = +0.02, Holm-adjusted p = 0.69) and the total number of pairs (rho = +0.11, Holm-adjusted p = 0.69) also showed no association. Distance-based leave-one-patient-out prediction had a mean absolute error of 0.090, compared with 0.076 for the leave-one-out mean. Two patients whose nearest pair was 36–53 mm from the hand knob had AUCs of 0.95–0.96; the patient with the most pairs within 20 mm had the lowest AUC (0.57). The results indicate that hand-knob coverage was not a useful pre-calibration predictor for this offline task and cohort. The sample was small (n = 15), so a moderate effect cannot be excluded. Other limits include one session, offline event-locked movement versus rest windows that also include visual and decision activity, a fixed MNI hand-knob point, fsaverage coordinates treated as MNI, and no individual anatomy.

stereoelectroencephalography · SEEG · brain-computer interface · high-gamma activity · movement detection · electrode geometry · hand knob · area under the ROC curve · pre-registration