Random 0.6–1.4 s foreperiod did not change offline SEEG key-press detector score or AUC in 8 subjects

We tested whether the random interval from fixation to go cue (foreperiod, 0.55–1.42 s; median 0.99 s) changed an offline key-press versus rest detector in 8 subjects from OpenNeuro ds004473. We used archived bipolar 70–150 Hz high-gamma and 13–30 Hz beta log-power features and shrinkage LDA, trained on trials 1–100 and tested on trials 101 onward. Foreperiod did not correlate with the move-window score (mean Spearman rho +0.016, 4/8 positive, exact two-sided sign-flip p=0.61), including after response time was removed (mean +0.021, p=0.48). Long- versus short-foreperiod AUC difference was +0.001 (95% t-CI −0.022 to +0.024; p=0.92). Five-fold cross-validation and high-gamma-only analyses gave the same null result. Foreperiod also did not correlate with response time (mean rho −0.011, p=0.74); previous-trial foreperiod did not correlate with response time given current foreperiod (mean partial rho +0.016, p=0.28). One uncorrected high-gamma-only control was nominally significant (foreperiod versus rest score, mean rho +0.042, 6/8 positive, p=0.031); the rest window preceded fixation, so foreperiod could not cause this result, which we considered likely due to chance among 12 uncorrected tests. The results support use of a detector calibrated over this random foreperiod range without foreperiod correction in this task. They do not establish that foreperiod is generally unimportant: this study had 8 subjects, low power for small effects, offline cued windows, and no pre-registration; the subjects also showed no behavioural foreperiod effect.

stereo-EEG · brain-computer interface · foreperiod · motor decoding · linear discriminant analysis · high-gamma activity · beta activity