At matched pair counts, SEEG word-recall decoding showed no significant MTL advantage over lateral temporal cortex in 18 patients

A pre-registered analysis compared word-recall decoding from medial temporal lobe (MTL), lateral temporal cortex (LTC), frontal cortex (FRONT), other cortex (OTHER), and white matter (WM) in 18 patients with stereo-EEG recordings during session 0 of the OpenNeuro ds004789 FR1 task. We used archived per-word log-power features from 3 to 140 Hz in the 0–1.6 s after word onset, and leave-one-list-out L2 logistic regression. All regions decoded recall above chance: median AUCs were 0.566 (MTL, n=18), 0.621 (LTC, n=18), 0.620 (FRONT, n=18), 0.641 (OTHER, n=16), and 0.575 (WM, n=9); Holm-adjusted p values were 6.7e-4, 5.7e-5, 7.6e-5, 6.4e-4, and 5.9e-3, respectively. The all-channel median AUC was 0.669. In the primary matched-pair comparison, MTL was not significantly better than LTC: median AUCs were 0.566 and 0.601, respectively; the median paired difference was -0.022, MTL was higher in 7/18 patients, and the exact two-sided Wilcoxon p value was 0.26. Comparisons of FRONT with LTC, FRONT with MTL, and WM with LTC were also not significant after Holm adjustment (p >= 0.59); results within serial position agreed. Single-pair decoding was weak (median patient-level HFA AUC 0.477–0.523 and theta AUC 0.466–0.494 across regions). The results indicate distributed recall information and do not show an MTL advantage at matched pair count. Limits include automatic atlas labels that differed by patient, exclusion of mixed pairs, small clinical pair counts, 18 patients, offline AUC from one session and one time window with one fixed regularization value, and no causal stimulation test. The study cannot exclude a small MTL–LTC difference.

stereo-EEG · word recall · memory decoding · medial temporal lobe · lateral temporal cortex · brain-computer interface · AUC · matched channel count