Movement-locked SEEG high-gamma activity predicted movement vigor in 6 of 7 patients

We tested whether 70–150 Hz stereo-EEG (SEEG) activity predicts trial-by-trial peak mouse velocity (vmax) in the OpenNeuro ds006253 random-dot motion task. We analyzed 7 of 22 patients, selected by the smallest high-gamma pickle file size, using 127–270 trials per patient. A ridge decoder with time-ordered 5-fold cross-validation predicted log(vmax) from mean bipolar-channel activity in 0.5 s windows. We removed trial number, log reaction time, coherence, chosen side, and accuracy from the target within each training fold. In the primary window, 0–0.5 s after movement onset, held-out r was positive in 6 of 7 patients (mean r = 0.116; exact one-sided sign-flip p = 0.031). Two patients had per-patient permutation p < 0.01. The pre-stimulus control showed no prediction (mean r = −0.037, p = 0.74), and the post-stimulus, pre-movement, and pre-click windows did not show significant group prediction. Exploratory 200 ms bins had positive r in all 7 patients from −0.2 to 0 s and from 0.2 to 0.4 s (mean r = 0.107 and 0.142; uncorrected p = 0.008 for each); the 0 to 0.2 s bin was near zero. Univariate effects were sparse (0–2 FDR-significant channels per patient). The effect was small (mean held-out r about 0.12, or about 1–2% of variance) and distributed across channels. This offline analysis used only 7 patients chosen by file size, did not use electrode positions, and retained one sub-09 trial with an extreme residual (log vmax about −32). The group test had 7 units and a minimum possible p of 1/128; replication in the other 15 patients is necessary. The original movement-onset times may blur the time course, and univariate p values use a normal approximation.

stereo-EEG · high gamma · movement vigor · peak velocity · ridge regression · motor decoding · random-dot motion