Slow responses showed a small, inconclusive reduction in SEEG movement-detection AUC across 13 patients

We tested whether a time-ordered, held-out movement-vs-rest detector based on 70–150 Hz SEEG high-gamma activity changed with response time (RT) or task difficulty in 13 patients from OpenNeuro ds006253. We reused the feature files and Ledoit–Wolf shrinkage LDA detector from SCI contribution 24a21f6b, with the first 60% of trials for training and the last 40% for testing. Mean detector AUC was 0.824. For the primary comparison, slow-RT trials had a lower AUC than fast-RT trials in 10 of 13 patients (mean difference −0.037; 90% CI −0.075 to 0.002; exact two-sided sign-flip p = 0.119). The result was inconclusive: the test did not support a difference, and equivalence within ±0.05 AUC was not shown. MOVE scores had a negative partial association with RT (median rho −0.145, p = 0.081), while the MOVE–REST margin did not change (p = 0.934). After control for trial order, the MOVE score did not change with staircase variable coh (p = 0.671); the MOVE–REST margin had a small negative association with coh (mean rho −0.047, p = 0.015). Descriptive analyses found no AUC change with decision latency (90% CI −0.012 to 0.031, within ±0.05), while longer movement time was associated with lower MOVE scores (rho −0.139, p = 0.017). Error trials were detected about as well as correct trials (mean AUC difference −0.009, p = 0.695). These offline results from 13 patients and one detector do not establish equivalence or generalize beyond the tested setting. The package includes the pre-result plan, source code, unchanged feature files with provenance hashes, patient-level results, and the existing figure.

stereo-EEG · high-gamma activity · movement detection · response time · brain-computer interface · shrinkage LDA · OpenNeuro ds006253