In five SEEG patients, late high gamma improved response-time prediction in four, but the group test was not significant

We tested whether stimulus-locked 70–150 Hz high-gamma activity (HGA) predicted single-trial response time (RT) beyond motion coherence in five patients from OpenNeuro ds006253. Ridge regression used the mean HGA of each bipolar channel in pre-stimulus (−0.5 to 0 s), early (0 to 0.5 s) or late (0.5 to 1.0 s) windows. We used contiguous five-fold cross-validation and compared out-of-fold R² for a coherence baseline with a coherence-plus-HGA model. A 50-shift circular null preserved slow drift. In the late window, HGA added out-of-fold RT information in four of five patients (dR² +0.065 to +0.251; shift p ≤ 0.04); one patient showed no gain (dR² −0.060). The group mean dR² was +0.098, but the exact sign-flip test was not significant (p = 0.062). Coherence alone did not predict RT out of fold (R² ≤ 0 in all patients). Pre-stimulus and early windows gave smaller and less consistent gains. After excluding trials with movement before 1 s, only three patients had at least 60 trials; the late-window gain remained in two. This indicates that part of the late-window effect may reflect movement onset. The analysis is offline, within patient, and based on five patients with one session each.

SEEG · high-gamma activity · response time · motion coherence · ridge regression · cross-validation · single-trial prediction · OpenNeuro ds006253