Beta power made 50-trial calibration non-inferior to 100-trial high-gamma calibration in eight SEEG subjects

Calibration time is a cost for BCI users. We measured how many calibration trials 13-30 Hz (beta) power saves when it is added to 70-150 Hz (high-gamma) power in an SEEG key-press detector. Data: archived bipolar log band power of sci 04683a2a (OpenNeuro ds004473, 8 subjects, rest and move windows), with hashes checked. Shrinkage LDA was calibrated on the first 10-100 trials and tested on trials 101+. The pre-registered primary test was non-inferiority of high-gamma plus beta at 50 trials to high-gamma at 100 trials (margin 0.02 AUC). The result supports it: mean difference +0.015, median -0.003, 7 of 8 subjects above the margin, exact sign-flip p = 0.020. The result stayed significant when we removed any one subject. With 30 trials, non-inferiority was not shown (mean -0.007, p = 0.28). Beta added AUC at all calibration sizes (mean +0.020 to +0.047), and the gain did not become smaller with 10 trials. The saving was largest in subjects with weak high-gamma signals and zero in 2 subjects with high-gamma AUC near 0.9-0.95. Limits: offline, cued fixed windows that include the press, one session, 8 subjects, first-n calibration order only.

SEEG · brain-computer interface · calibration · beta power · high-gamma power · shrinkage LDA · non-inferiority