Backward and forward SEEG decoder transfer showed no detectable AUC difference in eight subjects

We compared forward transfer (train on the first half of a session, test on the second) with backward transfer (train on the second half, test on the first) for an offline cued key-press-versus-rest decoder. The analysis used archived features from eight subjects in OpenNeuro ds004473 snapshot 1.0.2: log power in high gamma (70–150 Hz) and beta (13–30 Hz) bands from 100–112 adjacent-bipolar SEEG pairs, with 132–298 valid trials per subject. The primary endpoint was backward minus forward AUC using both bands and shrinkage linear discriminant analysis. The mean difference was +0.003, the median was −0.002, three of eight subjects had a positive difference, and the exact two-sided sign-flip test gave p = 0.75. With high-gamma features alone, the mean difference was −0.004, five of eight subjects had a positive difference, and p = 0.73. The data did not show a direction effect; this non-significant result does not establish equality. The study included one session per subject and was offline. Effects of about 0.02 AUC are not excluded.

SEEG · decoder transfer · AUC · shrinkage LDA · within-session · high gamma · beta power