SEEG high-gamma activity did not decode left/right mouse choice before movement in 5 ds006253 patients

We tested when the left/right choice becomes readable from SEEG high-gamma activity (HGA, 70-150 Hz) in the ds006253 random-dot-motion task. We used binned epochs of 5 patients from sci 8d5572e9 (138-269 trials, 50-97 channels per patient). We removed trials with a change of mind. A shrinkage LDA with 5-fold cross-validation used per-channel log HGA from -0.5 to 0 s before mouse-movement onset. Median AUC was 0.518. No patient had permutation p < 0.05 (200 permutations). The group sign-flip p was 0.156 and the Fisher p was 0.284. A decoder trained on correct trials (-0.25 to +0.25 s) and tested on error trials gave median AUC 0.517 (sign-flip p = 0.25). Thus the test did not show whether the signal follows the choice or the stimulus. In the movement-locked time course, no 150 ms window was reliable after a max-statistic correction (p = 0.71). Exploratory result: from 0.25 to 0.75 s after movement onset, error-trial AUC was above 0.5 in 5 of 5 patients (median 0.569, p = 0.0625, the smallest p this test can give). Limits: 5 patients selected by file size, clinical electrode placement, no anatomy check, and few error trials. The package contains the plan, the numpy-only code, the results, an SVG figure and SHA-256 hashes of all inputs.

stereo-electroencephalography · high-gamma activity · choice decoding · brain-computer interface · random dot motion