Eight-subject SEEG comparison found similar key-press decoding AUC for white- and gray-matter pairs

In 8 subjects from OpenNeuro ds004473, random sets of 8 white-matter (WM) and 8 gray-matter (GM) SEEG bipolar pairs detected offline key press versus rest using 70–150 Hz log power and shrinkage LDA. The pre-specified primary test found similar group AUC: median WM minus GM difference +0.004, 5 of 8 subjects positive, exact two-sided sign-flip p = 0.80. It did not find a difference, but does not establish equivalence: power was low and subject differences ranged from −0.089 to +0.134. In descriptive secondary analyses, all WM pairs gave AUC near all GM pairs (median difference −0.025; 2 of 6 decodable subjects positive; p = 1.0). Adding all WM pairs to all GM pairs increased AUC in all 6 decodable subjects (median +0.026, p = 0.031), but also added more pairs and therefore does not isolate a WM effect. Within WM, single-pair effect size did not decrease with distance to cortex up to about 8 mm (median within-subject Spearman rho +0.03, p = 0.84). Deep WM and 13–30 Hz beta analyses gave the same result as the primary test. WM pairs were not over-represented among the 10 best single pairs (median fraction difference −0.09, p = 0.25). Secondary tests were not corrected for multiple comparisons. These data do not support removing WM pairs by default from SEEG BCI channel sets. The study is limited by 8 subjects, offline 60-trial data, no pre-specified equivalence margin, uncertain tissue assignment from one midpoint voxel, possible volume conduction, and a press window that includes visual cue and motor activity. The exact OpenNeuro release analyzed is not established; the source report identifies the live dataset and S3 URLs. See the full methods, subject results, assumptions, and limitations in README.md.

stereo-EEG · SEEG · brain-computer interface · white matter · gray matter · bipolar pairs · key-press decoding · AUC · shrinkage LDA · motor task