Bipolar high-gamma activity did not reliably decode confidence in five SEEG patients

We tested whether bipolar 70–150 Hz high-gamma activity (HGA), recorded around the choice click and before the confidence scale appeared, decoded confidence on correct trials in five stereoelectroencephalography (SEEG) patients from OpenNeuro ds006253. A within-patient shrinkage LDA used stratified five-fold cross-validation with five repeats. The median AUC for high versus low confidence was 0.494 (two of five patients above 0.5; exact group sign-flip p = 0.44); only sub-12 was individually significant (AUC 0.62, permutation p = 0.01). A longer pre-click window gave median AUC 0.513 (p = 0.41). The HGA score had median Spearman correlation −0.004 with continuous confidence (p = 0.25). Response time alone performed better (median AUC 0.652). Adding HGA to response time, movement time, and stimulus coherence gave a mean AUC gain of 0.042 (p = 0.031), but a shuffled-score null check showed gains beyond the null only in sub-12 and sub-13; the sub-13 gain used a reversed HGA score (AUC 0.41). We therefore do not interpret the combined-model gain as evidence of reliable confidence information. The result applies to these five of 22 patients, selected by file size in the source work, and to offline within-patient linear decoding without anatomical channel selection. It does not show that the brain has no confidence signal.

stereoelectroencephalography · high-gamma activity · confidence · decoding · decision-making · brain-computer interfaces · cross-validation