In 8 SEEG subjects, visual cues affected move-vs-rest detectors, while line length was not less movement-specific than high-gamma power

This offline analysis tested whether detectors trained to distinguish cued key presses from rest also respond to a visual-only window, and whether line length (LL) is less movement-specific than 70–150 Hz power (HG). The preregistered analysis used 120 trials each from 8 subjects in OpenNeuro ds004473, with 0.5 s rest, cue, and move windows and a shrinkage LDA trained on the first 60% of trials and tested on the last 40%. All five feature families responded to the cue window. For the all-channel move-vs-rest decoder, cue-vs-rest AUC was 0.628 for HG (7/8 subjects above 0.5, p = 0.023) and 0.614 for LL (7/8, p = 0.016); the cue window provided about half of above-chance move-vs-rest separation for both. The primary endpoint did not support the hypothesis that LL is less movement-specific than HG: move-vs-cue AUC was 0.649 for LL and 0.646 for HG (difference +0.003; 6/8 subjects higher; sign-flip p = 0.80). Cue windows used as additional training negatives lowered cue responses, while a threshold calibrated to 5% false alarms on training rest yielded 24–56% false alarms on test rest for all-channel read-outs. These offline results from 8 epilepsy patients show that absolute move-vs-rest AUCs partly reflect visual-event detection. Move vs cue is not a pure movement contrast because the Go cue is also visual. Secondary tests were not corrected for multiple comparisons.

stereo-EEG · brain-computer interface · line length · high-gamma power · visual cue · movement detection · AUC · shrinkage LDA · OpenNeuro ds004473