In eight SEEG subjects, white matter pairs decoded key presses about as well as equal-count gray matter pairs

We tested whether white-matter (WM) bipolar pairs carry key-press information and whether excluding them reduces decoding in stereo-electroencephalography (SEEG). We used archived features for eight subjects (60 trials each) from OpenNeuro ds004473 v1.0.2: log power at 70–150 Hz and 13–30 Hz, labels at each pair midpoint, and distance to cortex. Features were copied unchanged from an earlier SCI contribution and retained with SHA-256 checksums. A Ledoit–Wolf shrinkage linear discriminant decoder used five-fold cross-validation, with both windows from each trial kept in the same fold, and ten repeats. At equal pair counts, WM-only and gray-matter-only sets had no significant group AUC difference in any feature set (p ≥ 0.31); this does not establish equivalence. WM-only decoding exceeded a label-swap null in six of eight subjects with both bands. Adding WM pairs to gray-matter pairs gave a mean gain of 0.020–0.035 AUC, not statistically significant. With both bands, all-pair decoding exceeded WM-only decoding by 0.018 mean AUC in all eight subjects (p = 0.008); this added analysis was not in the plan and was not corrected for multiple tests. Single-pair effects did not decrease with distance to cortex. The results do not support a fixed rule that removes WM pairs for this task, but are limited by eight subjects, 60 trials each, midpoint tissue labels, possible volume conduction, and offline windows in which the press interval also contains the visual cue.

stereo-electroencephalography · SEEG · white matter · gray matter · brain-computer interface · key-press decoding · linear discriminant analysis · cross-validation · OpenNeuro