A detector trained on one movement side had a mean cross-side AUC difference of −0.006 in 10 offline SEEG patients

A brain-computer interface movement detector should work across movement directions. We tested whether a detector trained on movements to one target side could detect movements to the other side in a two-target mouse task. We used archived 70–150 Hz features from 10 SEEG patients in OpenNeuro ds006253 and trained shrinkage linear discriminant analysis models on the first 60% of trials from one side. We tested on the last 40%. Mean AUC was 0.792 for movements to the trained side and 0.786 for the other side (mean difference −0.006; median −0.003; 6 of 10 patients had a negative difference). The pre-registered non-inferiority test with a 0.05 AUC margin passed (one-sided exact sign-flip p = 0.015); the two-sided test of a zero difference was not significant (p = 0.69). One patient had a 0.133 AUC loss. The result applies to offline detection in one task, one session per patient, and two horizontal target sides using the same hand. MOVE and REST were different task phases, so the detector could use visual and decision activity as well as motor activity. Calibration should include all directions when possible, with a patient-level check.

SEEG · brain-computer interface · movement detection · cross-side transfer · linear discriminant analysis · AUC · non-inferiority · mouse movement