Pretrial Rest Does Not Predict Movement-Window Decoder Noise in SEEG Key-Press Data (8 Subjects)

We tested whether decoder-axis fluctuations in a pretrial rest window continued into the key-press window of the same trial, and whether same-trial baseline subtraction improved separation. Archived features from eight subjects in the cued SEEG task were analyzed with a shrinkage LDA trained on trials 1–100; all reported tests used held-out trials 101 onward. For high-gamma features, the mean same-trial rest-to-movement Spearman correlation was −0.016 (95% CI −0.085 to +0.052; 3/8 positive; exact two-sided sign-flip p = 0.58). The rest score explained at most 1.3% of movement-score variance in any subject. With high-gamma and beta features, mean correlation was −0.082 (1/8 positive; p = 0.016), and paired d′ was smaller than unpaired d′ in all eight subjects. Movement-window variance was not consistently greater than rest-window variance (high-gamma mean log variance ratio +0.040; 95% CI −0.185 to +0.265; p = 0.71), so the analysis found no evidence for a gain change, while a small effect remains possible. In these windows, about two seconds apart, decoder-axis fluctuations did not continue from pre-fixation rest to key press, and a pretrial baseline did not correct this noise. The study is limited to one dataset, eight subjects, a cued key-press task, fixed windows, and offline analysis; subject 11 had 32 held-out trials. The observed excess decoder-axis variance was only 1.3–1.6 times the channel-shuffled value, smaller and less consistent than in the related earlier analysis. These results do not establish the time scale beyond the tested windows, and a small fast shared component remains possible.

stereo-EEG · decoder-axis noise · baseline subtraction · key-press task · shrinkage LDA · high-gamma power · beta power · within-trial correlation