Press-locked channel selection did not make SEEG key-press decoder outputs press-locked

We tested whether selecting press-locked bipolar channels makes a high-gamma SEEG decoder output track key-press timing. Using the publication-reported OpenNeuro ds004473 v1.0.2 cued key-press data from 8 subjects (1,977 trials), we applied the alignment-index rule in training folds and evaluated the same shrinkage LDA decoder across 10 five-fold cross-validation splits. Among 7 subjects with all-channel AUC >= 0.6, the PRESS-minus-ALL latency-slope difference was +0.23 median (mean -0.12; positive in 5 of 7; exact sign-flip p = 0.84). The PRESS median slope was 0.17, below the prespecified press-locked threshold of 0.8 (p = 0.016 versus 1); score alignment was unchanged (median 0.3 for both sets). AUC decreased in all 7 subjects (median 0.83 to 0.67; difference -0.13; p = 0.016). Cue-locked control channels gave a similar slope (PRESS-minus-CUE median +0.03; p = 0.94; 6 subjects). Across 10 splits, the ALL slope range exceeded 0.4 in 4 of 7 subjects, showing that single-split timing estimates can vary substantially. Few pairs passed selection, confidence intervals were wide, and the half-rise endpoint was noisy for decoders with few channels. These offline results provide no evidence that this selection rule works for this task and feature set; they do not show that all channel-selection methods fail.

SEEG · brain-computer interface · key-press decoding · high-gamma activity · channel selection · latency · cross-validation · shrinkage LDA