Matched-count left and right SEEG channels showed no detectable difference in word-recall decoding

We compared within-patient word-recall decoding from left- and right-hemisphere SEEG channels in eight patients with bilateral contacts, matching the number of channels between hemispheres. The pre-registered primary test found no detectable difference in held-out AUC: median left-minus-right difference +0.014, bootstrap 95% CI [-0.028, +0.080], with 4/8 positive differences and exact two-sided Wilcoxon p = 0.31. This result does not establish equivalence; with eight patients, differences of about 0.03-0.05 AUC remain possible. Secondary analyses gave similar results for AUC within serial position (median +0.021, p = 0.38) and single-channel spectral-tilt effect size (median +0.004, p = 0.55). In uncorrected analyses using all channels, right-only decoding had a median AUC loss of 0.051 relative to all channels (p = 0.039), while left-only decoding had a loss of 0.017 (p = 0.64); these comparisons do not isolate hemisphere because channel counts differ. The unequal-count all-left versus all-right comparison was not significant (median +0.022, p = 0.38). The matched left-only decoder was above chance in 8/8 patients (median AUC 0.637, p = 0.004), and the matched right-only decoder in 6/8 (median 0.612, p = 0.098). The data were archived per-word band-power features from OpenNeuro ds004789 v3.1.0, FR1 session 0. Limitations include n = 8 bilateral patients, wide confidence interval, no region matching, unknown language dominance, one session, one feature set, and one linear decoder. The results do not show that a right-hemisphere implant is equivalent or universally adequate for a memory BCI; they show no detectable disadvantage at equal channel count in this sample.

SEEG · word recall · memory decoding · hemisphere · AUC · intracranial EEG · brain-computer interface · logistic regression