Recall prediction from SEEG encoding power transferred between sessions in three patients

In three patients with depth-electrode recordings during free recall, logistic regression on five bands of SEEG power from 0–1.6 s after word onset predicted later recall across two sessions. Cross-session AUCs, averaged over both training and test directions, were 0.58, 0.67, and 0.66; each exceeded its permutation null (p = 0.010, 0.010, and 0.020). Within-session AUCs averaged 0.61, 0.60, and 0.65, with p = 0.038 in each patient, the minimum attainable with 25 permutations. Per-session z-scoring changed cross-session AUC by no more than 0.02. High-gamma subsequent-memory-effect maps correlated across sessions at r = 0.29, 0.04, and 0.19. The analysis used three subjects selected for file size and bandwidth, two sessions each, one feature set and fixed penalty, and few permutations; it did not assess anatomy or session gap. The cross-session and within-session training sets were not exactly matched. These results show transfer in this small sample and do not establish generalization to other patients or settings.

SEEG · free recall · memory encoding · cross-session transfer · band power · logistic regression · AUC · permutation test · subsequent memory effect