Pre-choice SEEG high-gamma activity did not reliably decode later confidence in five patients

We tested whether bipolar 70–150 Hz high-gamma activity (HGA) from 0.45 s before to 0.15 s after a choice click decoded the later confidence rating in five patients with SEEG during a random-dot motion task. We used the authors’ HGA derivative, median-split confidence labels, shrinkage linear discriminant analysis, and repeated stratified five-fold cross-validation. HGA had a median AUC of 0.558 (four of five patients above 0.5; exact sign-flip p = 0.312); on correct trials, median AUC was 0.537 (p = 0.406). Reaction time alone had a median AUC of 0.700. Although stacked reaction-time and HGA models had a raw mean gain of 0.062, after subtracting the shuffled-HGA null mean, the gain was 0.042 (p = 0.156) for all trials and 0.007 (p = 0.312) for correct trials. These offline, within-patient linear tests cover one pre-choice window and five patients; permutation p values varied between runs, and the median split discards information from the continuous rating. The results do not show reliable confidence decoding or a reliable HGA contribution beyond reaction time.

SEEG · high-gamma activity · confidence · metacognition · brain-computer interface · reaction time · linear discriminant analysis · AUC