In 18 patients, single-session SEEG memory maps agree weakly and the top channel retains 30% of its t value

We assessed whether a per-channel subsequent-memory-effect (SME) map from one session can identify a recording or stimulation target. We used archived word-level SEEG power features for 18 patients in session 0 of a free-recall (FR1) task and compared channel-wise recalled-versus-not-recalled Welch t maps from odd and even lists. For high-frequency activity (HFA, 80 and 140 Hz), median split-half Spearman rho was 0.26 (16/18 positive); the exact Wilcoxon test against zero gave p = 2.5e-04. In a group permutation test, mean rho was 0.218, above the shuffle-null mean of 0.038 (95th percentile 0.098; p = 0.001, the minimum possible with 1000 shuffles). Six of 18 patients had per-patient shuffle p < 0.05. Median Spearman-Brown full-session reliability was 0.41. Theta and HFA-minus-theta tilt maps had median rho values of 0.28 and 0.21. The channel with the largest positive HFA t in one half had median t = 3.04 in the selection half and 0.86 in the held-out half (ratio 0.29); held-out t exceeded 2 in 3/18 patients. Map agreement was not significantly correlated with decoder AUC (Spearman r = 0.34, permutation p = 0.16). Thus, in this sample and task, one FR1 session did not reliably identify a single best channel. The study is limited to session 0, 18 patients, one 0-1.6 s window, fixed frequency bands, and a bipolar montage; reliability across sessions or days was not tested. The package retains the original analysis plan, code, per-patient results, maps, null distributions, figure, and input checksums.

SEEG · subsequent memory effect · split-half reliability · free recall · intracranial EEG · high-frequency activity · brain-computer interface · winner's curse