Equal-count SEEG decoding found no medial temporal advantage over lateral temporal pairs in 18 patients

We compared offline decoding of later word recall from medial temporal lobe (MTL), lateral temporal cortex (LTC), frontal cortex (FRO), other labeled sites (OTH), and all implanted sites in 18 patients from OpenNeuro ds004789 FR1, session 0. Each within-patient L2 logistic decoder used four bipolar contact pairs, selected in 20 random draws per group, and power features from 0–1.6 s after word onset. The pre-planned MTL–LTC comparison was not significant: median AUC difference (MTL minus LTC) was −0.015, MTL was higher in 7/18 patients, and the exact two-sided Wilcoxon p value was 0.39. AUC within serial position gave a similar result (−0.006; 8/18; p = 0.55). Each region decoded above chance (median AUC 0.553–0.576; Holm-adjusted p ≤ 0.02), but no region outperformed four random pairs from all sites (median AUC 0.578); all regional median differences versus ALL were negative and not significant. The all-pairs decoder had median AUC 0.682. An exploratory eight-pair analysis had the same ranking; only 15 patients had at least eight MTL pairs. These results do not show a per-pair MTL advantage in this dataset and suggest that contact count and spatial spread may matter for this offline decoder. The study covers one task and session, uses atlas labels not checked against images, and cannot exclude an MTL difference smaller than about 0.02–0.03 AUC with 18 patients. It does not assess stimulation targets or establish performance for a causal online decoder.

SEEG · word recall · memory decoding · bipolar electrodes · medial temporal lobe · AUC · OpenNeuro