In 8 SEEG subjects, key presses did not reduce 70–150 Hz variability in the 10 most responsive channels

Question: does a cued key press decrease the trial-to-trial variability of SEEG band power in single channels (variability quenching)? This is also a test of the equal-spread assumption of LDA BCI decoders. Data: archived features of sci 04683a2a (OpenNeuro ds004473, 8 subjects, adjacent-bipolar log power, 70-150 Hz and 13-30 Hz, rest and key-press windows of 0.5 s). Method: the plan was hashed before analysis. We selected the 10 channels with the largest 70-150 Hz response on trials 1-100. On trials 101+ we calculated Q = median ln(SD move / SD rest) and did an exact sign-flip test over subjects. Result: Q had mean +0.066 and median -0.009, with Q < 0 in 4 of 8 subjects (p = 0.54). Thus the data do not show quenching. Beta power in the same channels, all channels, control channels and detrended data also gave no significant decrease. Secondary (not corrected): channels with a larger 70-150 Hz response had a larger move/rest SD ratio (median Spearman rho +0.27, 7 of 8 subjects, p = 0.016). Thus responsive channels become slightly more variable, not less, during the key press. Limits: offline, one task, 8 subjects. Estimation noise of 0.5 s windows can hide a small quench. The rest and move windows are aligned to different events. Code is pure Python and is in the package, with the features and their hashes.

SEEG · variability quenching · high gamma · beta power · brain-computer interface · linear discriminant analysis · trial-to-trial variability · sign-flip test