Left and right SEEG channels decoded later word recall similarly in eight patients

We tested whether left-hemisphere stereo-electroencephalography (SEEG) channels decode later word recall better than an equal number of right-hemisphere channels. We used archived encoding features from 19 patients in the OpenNeuro ds004789 RAM FR1 delayed free-recall dataset. Eight patients had at least 20 same-hemisphere bipolar pairs on each side and at least 15 recalled and 15 unrecalled words. For each patient and hemisphere, we repeatedly drew 20 pairs and used leave-one-list-out L2 logistic regression on log power at 3–140 Hz. Mean AUC was 0.615 for left channels and 0.576 for right channels. The mean left-minus-right difference was +0.039 (95% patient-bootstrap CI −0.017 to +0.110); 4 of 8 patients had a positive difference (exact sign-flip p = 0.35; exact Wilcoxon p = 0.31). Secondary group contrasts were also not significant. Combining both hemispheres did not improve AUC over the better single hemisphere (mean difference −0.005; 95% CI −0.020 to +0.011). In uncorrected patient-level list bootstraps, three patients showed left greater than right at p < 0.05 and none showed right greater than left; none passed the Bonferroni threshold of 0.05/8. With only eight patients, the interval does not exclude a left advantage of about 0.11 AUC. Language dominance and handedness were unavailable, electrode regions differed by hemisphere, and the result applies to one linear decoder of spectral power. The raw-data release is not recorded; S3 URLs, ETags, and feature-file MD5 hashes are preserved in the package.

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