In 8 SEEG subjects, slower cued presses had higher detector output and fewer misses

We tested whether a shrinkage linear discriminant trained on trials 1–100 could produce weaker single-trial output for later cued key presses in archived features from 8 subjects in OpenNeuro ds004473 v1.0.2. Using 70–150 Hz high-gamma and 13–30 Hz beta log power from bipolar pairs, we compared the move-window margin with fixation-to-press latency on trials 101 and later. Contrary to the preregistered prediction, mean Spearman rho was +0.120 (positive in 7/8 subjects; exact two-sided sign-flip p = 0.016). The pre-fixation rest-window control had mean rho +0.009 (p = 0.75). Miss rate was 0.33 in the fastest third and 0.22 in the slowest third (lower for slow presses in 7/8 subjects; p = 0.016); AUC was 0.821 versus 0.847 (higher for slow presses in 8/8; p = 0.0078). High-gamma alone showed the same direction (mean rho +0.159; 7/8; p = 0.016). Exploratory partial rank correlations controlling trial order remained positive in 7/8 subjects. These offline findings concern one dataset, 8 subjects, one time-ordered split, overt finger presses, two fixed windows, and latency measured from fixation rather than the cue; causes were not tested.

stereo-EEG · brain-computer interface · key-press detection · high-gamma · beta power · linear discriminant analysis · movement latency · single-trial decoding