Left and right hemisphere SEEG channels decode verbal working-memory load equally well (ds004752)

Verbal working memory is frequently described as left-lateralized. We tested this for SEEG load decoding, which is important for selecting implant sites for a brain-computer interface. We used archived maintenance-period (3–6 s) log10 band power (5 bands) from OpenNeuro ds004752, session 1. In each patient, shrinkage LDA decoded set size 8 versus 4 from 6 random channels in each hemisphere (30 subsets, stratified 5-fold cross-validation with 3 repeats). The test plan was written before analysis. In 14 patients with channels in both hemispheres, mean AUC was 0.641 on the left and 0.647 on the right. The mean difference (left minus right) was −0.006 (95% patient-bootstrap CI −0.066 to 0.056; exact sign-flip p = 0.86; left was better in 6 of 14 patients). Each side decoded load above chance (p = 0.0006 left; p = 0.0007 right). Three sensitivity analyses gave the same conclusion: all channels per side (difference −0.044, p = 0.25), homologous shafts only (−0.020, p = 0.63, n = 13), and hemisphere from the shaft-name suffix (+0.013, p = 0.70). The primary confidence interval excludes a mean left advantage larger than 0.056 AUC; the all-channel point estimate favored the right, without a significant difference. Limits: participants had epilepsy; implants sampled mostly medial temporal areas, not lateral language cortex; seizure side was not controlled; there were 9–29 trials per class; and the offline analysis used one session. The package includes analysis code, archived features with SHA-256 hashes and raw-file provenance, results, and an SVG figure.

verbal working memory · stereo-electroencephalography · SEEG · hemispheric lateralization · brain-computer interface · memory load decoding · intracranial EEG