White-matter SEEG pairs were non-inferior to gray-matter pairs for decoding movement versus rest in 15 patients

This pre-registered work repeats the ds004473 white-matter result in a second dataset and task. We used the 70-150 Hz features of 15 ds006253 patients (sci ceba42cd) and the fsaverage contact positions (sci 085fe79e). We gave each bipolar pair a tissue label from the MNI152 gray and white matter probability maps at the pair midpoint. For each patient, we decoded movement vs rest with 200 random sets of 8 white-matter pairs and 200 sets of 8 gray-matter pairs (shrinkage LDA, first 60% of trials for training, last 40% for test). The mean AUC difference (WM - GM) was +0.008, WM higher in 9 of 15 patients, exact sign-flip p = 0.73. The 95% bootstrap CI (-0.035 to +0.053) is above the -0.05 margin, thus WM sets were non-inferior. The upper bound is above +0.05, thus equivalence did not pass. With all pairs of the smaller class the result was the same (+0.015, CI -0.045 to +0.079). WM-only sets were above chance (mean AUC 0.655, p = 3e-5). In WM pairs, single-pair AUC did not change with distance to gray matter, but most WM midpoints were less than 3 mm from gray matter. Template labels, not patient MRI labels, are a limit.

stereo-EEG · white matter · movement decoding · brain-computer interface · AUC · non-inferiority