Equal-size white-matter and gray-matter SEEG pair sets had similar key-press AUC in eight subjects

This analysis repeated a white-matter (WM) versus gray-matter (GM) SEEG test with independent code and a larger matched set size (k = min(nWM, nGM) = 24–48 pairs; the earlier internal study used 8 pairs). It used 8 subjects from OpenNeuro ds004473, offline key press versus rest, log 70–150 Hz power from adjacent bipolar pairs, shrinkage LDA, and 5-fold cross-validation. Tissue was assigned from the FreeSurfer aparc+aseg label at each pair midpoint. The primary test, fixed before decoding, compared the mean AUC over 20 random k-pair subsets per class. The median WM-minus-GM AUC was −0.015 (mean −0.009), with 3 of 8 subjects above zero and exact two-sided sign-flip p = 0.67. This agrees with the earlier result (+0.004, p = 0.80), but the per-subject differences ranged from −0.098 to +0.096 and do not prove equivalence. Secondary tests were not corrected for multiple tests: WM-only pairs exceeded the per-subject 200-permutation null in 4 of 8 subjects (median AUC 0.625, versus 0.676 for GM-only and 0.707 for all pairs); deep WM pairs (at least 3 mm from cortex) exceeded the null in 4 of 8; adding WM pairs to GM pairs increased AUC in 7 of 8 (median +0.033, p = 0.070); WM single-pair effects did not decrease with distance to cortex; and 13–30 Hz power showed the same pattern. These offline data cover 60 trials per subject and 8 subjects, have low power for a group difference, and the press window includes a visual cue response. Pair tissue uses one midpoint voxel, a WM pair can record nearby cortex, and only a bipolar montage was tested. The results support checking WM pairs per subject instead of removing them by a fixed rule for SEEG BCI channel selection.

SEEG · white matter · gray matter · brain-computer interface · key-press detection · bipolar pairs · AUC · shrinkage LDA · OpenNeuro ds004473