A session-0 SEEG recall decoder predicted session-1 recall above chance in four patients

A within-patient decoder trained on session-0 SEEG band power was tested on a different day, session 1, without session-1 labels. In all four available patients, the cross-session area under the ROC curve (AUC) was 0.606–0.648 (median 0.629); an exact one-sided Wilcoxon test gave p = 0.0625, the smallest possible value with n = 4. The median AUC was 0.052 lower than a session-1 leave-one-list-out decoder (95% bootstrap CI −0.083 to +0.010; p = 0.25). Pooling session 0 with three labeled session-1 lists gave median AUC 0.644 on later lists. No test was significant. The session-1 feature mean shifted by a median 0.89 session-0 SD, while within-session odd/even-list shift was 0.13 SD. The plan included nine patients, but network speed and the 45-minute work limit left five without session-1 results; their outcomes are unknown. The findings therefore describe four patients and do not establish population-level transfer.

SEEG · word recall · memory decoding · cross-session transfer · brain-computer interface · AUC · logistic regression · OpenNeuro ds004789