An unchanged task-locked SEEG detector classified movement above chance in 12 patients, with no detected relation to channel count

We continued a pre-registered replication of an unchanged movement-versus-rest detector in OpenNeuro ds006253. The detector used author-provided 70–150 Hz high-gamma activity (HGA), Ledoit–Wolf shrinkage linear discriminant analysis (LDA), and a time-ordered split that trained on the first 60% of trials and tested on the last 40%. In the newly analyzed patient sub-07, held-out AUC was 0.851 (within-trial permutation p = 0.0001; 120 test trials; 89 channels), within the earlier 11-patient range of 0.569–0.966. Across all 12 patients, mean held-out AUC was 0.827 (median 0.838); all AUCs exceeded 0.5 and all patient-level permutation p values were below 0.05. The exact sign-flip test gave p = 0.00024, the smallest possible value for n = 12. AUC showed no detected relation to channel count (Spearman rho = 0.035, permutation p = 0.92; 50–148 channels). This was an offline analysis of one mouse-movement task, with task-locked MOVE and REST windows from different task phases, clinical SEEG contacts, author-provided HGA rather than raw voltage, and one split per patient. Patients were processed in pre-registered order, not selected at random; 10 of 22 patients remain unprocessed.

stereoelectroencephalography · SEEG · high-gamma activity · movement decoding · linear discriminant analysis · AUC · replication · OpenNeuro