SEEG high-gamma activity followed the chosen side during movement, while pre-movement decoding was not significant in five new patients

A pre-registered analysis tested whether 70–150 Hz high-gamma activity (HGA) in stereo-electroencephalography (SEEG) could decode left/right choice in a random-dot motion task with mouse responses. A Ledoit–Wolf shrinkage linear discriminant was trained on the first 60% of trials and tested on the last 40% in five patients not previously used for choice decoding and seven patients with archived features. The primary test was not supported: in the five new patients, mean area under the ROC curve (AUC) for choice in the 0.5 s before movement onset was 0.552 (4/5 above 0.5; one-sided sign-flip p = 0.156). Across all 12 patients, pre-movement AUC was 0.542 (10/12; p = 0.020), which did not survive Bonferroni correction across six group tests (adjusted p = 0.12). A decoder trained on correct trials followed the chosen side on held-out error trials in the first 0.5 s of movement (mean AUC 0.611, 10/12; two-sided p = 0.008); this window contains the mouse movement, so the signal may reflect movement direction. Limits include five new patients, 40–90 test trials and 15–40 test error trials per patient, and no correction across the six group tests. The result supports a weak choice signal during movement in this task, but not a reliable pre-movement intent signal.

stereo-electroencephalography · high-gamma activity · choice decoding · movement direction · area under the ROC curve · pre-registration · brain-computer interface