Rest 70–150 Hz power distinguished gray and white matter SEEG pairs in eight subjects; a ds006519 threshold transferred with 0.68 balanced accuracy

In eight subjects from OpenNeuro ds004473, gray-matter (GM) stereo-EEG (SEEG) bipolar pairs had higher 70–150 Hz rest log power than white-matter (WM) pairs in all subjects (median AUC 0.75; exact one-sided sign-flip p = 0.0039). A threshold fixed on ds006519 (z = −0.387) transferred without refitting and gave median balanced accuracy 0.68 in ds004473 (8/8 above 0.5; p = 0.0039). The top 25% of pairs by power had a median 26 percentage-point increase in GM over the subject base rate (7/8; p = 0.0078). Within WM, power decreased with distance from cortex (median Spearman rho = −0.30; 8/8; p = 0.0039). A two-feature logistic model ranked pairs (median AUC 0.71) but its probability >0.5 rule did not transfer (median balanced accuracy 0.50), consistent with a GM base-rate change from 93% in ds006519 to 28–58% in ds004473. The result supports a rapid tissue check, not replacement of imaging: balanced accuracy was modest, labels came from a single midpoint voxel, and higher WM power near cortex may reflect volume conduction, contact impedance, or bipolar reference geometry. The analysis used 60 half-second windows (30 s) between trials, eight subjects, differing tissue-label methods and frequency bands across datasets, and no correction for secondary tests. The exact ds004473 release used for the archived features is not established; its source URL and file hashes are retained.

stereo-EEG · SEEG · gray matter · white matter · bipolar pairs · high-gamma power · tissue classification · brain-computer interfaces · transfer · OpenNeuro