In 11 SEEG patients, movement-onset decoding weakly followed chosen direction in error trials (AUC 0.568), with a fragile effect

A direction decoder for a brain-computer interface should represent the chosen direction. Correct trials do not distinguish choice from stimulus direction, so this offline analysis trained a shrinkage linear discriminant on correct trials and evaluated held-out error trials, where choice and stimulus were opposite. We analyzed archived 70–150 Hz high-gamma features from 11 patients in OpenNeuro ds006253 (2,483 trials, including 660 errors; change-of-mind trials and trials with NaN features were excluded). In the prespecified 0–0.5 s after movement onset window, mean error-trial choice AUC was 0.568 (8/11 patients above 0.5; exact one-sided sign-flip p = 0.033), equal to the correct-trial AUC (0.568). Before movement onset, error-trial AUC was 0.496; in the 0–0.5 s after-stimulus window it was 0.510. The result indicates weak direction information about executed choice after movement onset, with no detected stimulus-direction information. The effect was fragile: leave-one-patient-out p values ranged from 0.011 to 0.066; without sub-14, mean error-trial AUC was 0.538 (p = 0.066). This was an offline analysis of one dataset and one author-provided feature type; error trials had lower coherence and longer reaction times, electrode locations were not used, and error trials may include motor slips. The exact OpenNeuro release of the source files is not established by the archived file records.

SEEG · brain-computer interface · error trials · choice decoding · high-gamma activity · shrinkage LDA · movement onset · OpenNeuro ds006253