In eight SEEG subjects, beta-band gain was not smaller with 20 than 100 calibration trials

We tested whether adding 13–30 Hz beta log power to 70–150 Hz high-gamma (HG) log power improves a shrinkage-LDA detector when calibration data are limited. Using archived bipolar band-power features from eight subjects in the cued left/right key-press task in OpenNeuro ds004473 version 1.0.2, we compared HG with HG+B (twice as many features) for calibration sizes N = 10, 20, 40, 60, and 100. All models were tested on trials 101 to the end. The preregistered primary endpoint was G = D(100) − D(20), where D(N) = AUC(HG+B) − AUC(HG); an exact one-sided sign-flip test over eight subjects gave mean G = −0.003, median +0.003, G > 0 in 5/8 subjects, and p = 0.57. Thus, the hypothesis that beta gain is smaller with 20 than 100 calibration trials was not supported. For first-N calibration, mean D was +0.047 at N = 10 (7/8 positive, two-sided p = 0.016) and +0.043 at N = 20 (6/8 positive, p = 0.047). Random draws of N trials from trials 1–100 yielded mean gains from +0.035 to +0.040 across N. At N = 100, mean D = +0.040, reproducing the prior result. Secondary tests were not corrected for multiple comparisons. The primary test included only eight subjects, so a small interaction (|G| < about 0.03) cannot be excluded. This was an offline analysis of one session per subject, one cued task in epilepsy patients, and one dataset. The rest window was pre-fixation, so the detector may use cue or attention signals as well as movement signals. First-N calibration also combines trial count with trial recency; the random-draw analysis isolates trial count.

SEEG · brain-computer interface · calibration · beta band · high-gamma power · shrinkage LDA · AUC · sign-flip test · OpenNeuro ds004473