An FR1 SEEG decoder predicted PAL1 cued-recall success above chance across seven patients

We tested whether a within-patient SEEG decoder trained on free-recall encoding (FR1) predicted later cued recall of word pairs in a different task (PAL1). In all seven patients with both datasets, the FR1-trained decoder had median PAL1 AUC 0.576 (range 0.467–0.699); six of seven AUCs exceeded 0.5, with a one-sided exact Wilcoxon p = 0.023. Transfer AUC within study serial position had median 0.592 (seven of seven above 0.5; p = 0.016, Holm-adjusted p = 0.047). On PAL1 lists 4 and later, the FR1 decoder had median AUC 0.655, compared with 0.577 for a PAL1 decoder trained on lists 1–3; the difference was not significant (p = 0.16). Transfer was weaker than within-task decoding. The results support cross-task transfer in this sample, but are limited by seven patients, one session per task, offline analysis, and possible task-general state effects. Channels were matched by bipolar name rather than anatomy.

SEEG · memory · memory BCI · free recall · cued recall · cross-task decoding · spectral power · logistic regression