Late-maintenance SEEG decoded verbal working-memory load in one of five patients; scalp EEG did so in one other patient

This replaces the earlier two-subject version 47904607. We asked whether temporal-lobe SEEG contacts or simultaneous scalp EEG carry information about verbal working-memory load. We streamed late-maintenance windows (trial seconds 4–6) from OpenNeuro ds004752 and decoded set size 8 versus 4 from log band power with cross-validated logistic regression and a 50-shuffle permutation null. Decoding was above the null in 4 of 20 subject-by-feature-set tests (uncorrected p<0.05). SEEG passed only in sub-04 (AUC 0.87–0.91, p=0.02–0.04). In sub-04, left hippocampal and temporal theta power increased with load (best rho=+0.64), while right hippocampal gamma decreased (rho=−0.64). EEG passed only in sub-03 (AUC 0.74, p=0.04). No feature set passed in the other three patients. Each patient contributed one session and 25–32 trials; the permutation null was coarse and tests were not corrected for multiple comparisons. Load information was not consistent across patients, so a BCI load decoder will need per-patient calibration and more trials.

verbal working memory · cognitive load · SEEG · scalp EEG · band power · decoding · permutation test · brain-computer interface