Pre-stimulus SEEG high-gamma activity weakly predicts decision errors in 15 patients

In an offline analysis of 15 patients performing random-dot motion decisions, we tested whether 70–150 Hz stereo-electroencephalography (SEEG) activity in the 0.5 s before stimulus onset predicts later errors. Using archived OpenNeuro ds006253 features, a pre-registered time-ordered 60/40 train/test split, and shrinkage linear discriminant analysis, pre-stimulus activity predicted errors above chance at group level (mean test AUC 0.541, 95% CI 0.509–0.576; 11/15 patients above 0.5; Holm-adjusted p = 0.040). Coherence-stratified AUC was 0.539 (p = 0.033); only 3/15 patients had a per-patient permutation p < 0.05. The upper confidence bound was below the pre-registered usefulness threshold of 0.60. Post-stimulus activity did not predict errors (AUC 0.507, p = 0.34). The coherence behavior baseline separated errors better than the neural score after a post hoc sign flip (AUC 0.595); higher `coh` values were associated with more errors in all 15 patients, opposite to the pre-registered sign. The analysis is offline, uses one session per patient, and may be affected by within-session drift. The direction of the dataset's `coh` scale was not checked against its documentation.

stereo-electroencephalography · high-gamma activity · decision errors · lapse monitoring · shrinkage linear discriminant analysis · OpenNeuro ds006253