SEEG band power was near chance for repetition decoding after serial-position control in four patients

We tested whether stereo-electroencephalography (SEEG) band power distinguishes repeated from first word presentations in the RepFR1 free-recall task, using session 0 from four patients in OpenNeuro ds005411. Repeats occur later in each list, so serial position can predict the label. With shrinkage LDA and list-grouped five-fold cross-validation, serial position alone had mean held-out AUC 0.880. Five-band SEEG features had mean AUC 0.620 before control (all four patients above 0.5, but none above the list-permutation null at p < 0.05) and 0.491 after training-fold linear and quadratic serial-position removal (one of four above 0.5; one-sided sign-flip p = 0.750). High-gamma onset responses occurred in 97–100% of channels. These results indicate that, in this small sample and with these features, the apparent repetition signal was mostly associated with time in list. The analysis included 4 patients, 8–10 lists per patient, and 10 permutations per decoder; residualization could also remove a repetition effect with the same trend, and the study was offline. R1531T was excluded because two lists had unequal word counts.

stereo-electroencephalography · free recall · repetition decoding · serial position · memory BCI · band power · shrinkage LDA