In eight SEEG subjects, one-key calibration detected presses of the other key with a mean AUC cost of 0.003

We tested if a SEEG press-vs-rest detector must see each movement in calibration. In OpenNeuro ds004473 (8 subjects), we used archived bipolar 70-150 Hz and 13-30 Hz log power (sci 04683a2a) and the left/right key of each trial from the OpenNeuro behavior files. For each subject, we trained three shrinkage-LDA models with the same number of training trials (42-50, from trials 1-100): left-key only, right-key only, and mixed. We tested them on presses of each key in trials 101+. The pre-specified primary result failed: the same-key model was not better than the other-key model (mean AUC 0.808 vs 0.805, D = +0.003, 95% CI -0.008 to +0.014, 4 of 8 positive, one-sided sign-flip p = 0.33). The cost was 0.028 AUC or less in each subject. HG alone gave the same result (D = -0.000). Move windows separated the two keys only weakly (mean key AUC 0.556); the 2 subjects with the largest cost had key AUCs of 0.77 and 0.64. Thus calibration with one key was sufficient for two similar finger movements that the contacts do not separate. The result does not show transfer between limbs or effectors. Limits: one dataset, 8 subjects, one time-ordered split, offline.

SEEG · brain-computer interface · movement decoding · shrinkage LDA · cross-key transfer