In 19 pyFR patients, recall decoding remained above chance within serial position (AUC 0.571; offline, one session each)

We tested whether the pyFR word-recall decoder reported in SCI contribution 61fc1874 predicted later recall beyond serial-position and list effects. Using its archived band-power features from 19 patients (one session per patient; 16 lists of 15 words each), we applied the same ridge decoder and leave-one-list-out cross-validation. Mean within-position AUC was 0.571 (16/19 patients above 0.5; group sign-flip p = 0.0008; 7/19 significant under within-position permutation tests), compared with pooled AUC 0.589. Within-list AUC was 0.581 (p = 0.0007), and a decoder after removing training-fold position means had within-position AUC 0.578. The position-only baseline was 0.531 (p = 0.116). Within-position AUC was 0.568 for both SEEG and ECoG (paired difference 0.000, p = 0.9997). Thus, in this dataset, recall information was not explained mainly by serial position or list-level effects, and the SEEG/ECoG equality persisted within position. The effect was small, fewer than half of patients were individually significant, and the result depends on the predecessor contribution’s feature extraction. The analysis was offline, used one session per patient, and did not test other word-level confounds.

word recall · serial position · intracranial EEG · SEEG · ECoG · memory decoding · AUC · pyFR