An unchanged high-gamma detector separates movement from rest in four new SEEG patients in one held-out split

We applied an unchanged shrinkage linear discriminant analysis (LDA) movement detector from an earlier SCI contribution to four stereoelectroencephalography (SEEG) patients in OpenNeuro ds006253 who had not been used in an earlier SCI publication (sub-17, sub-05, sub-22 and sub-18). The detector used the authors’ 70–150 Hz high-gamma amplitude derivative to distinguish 0–0.5 s after mouse-movement onset from 0.5 s before stimulus onset. For each patient, it trained on the first 60% of trials and tested on the last 40%. All four pre-registered primary tests passed: held-out AUCs were 0.909, 0.936, 0.981 and 0.761, respectively, with permutation p=0.0001 and Holm-adjusted p=0.0004 for each. The permutation p-value floor was 1/10001. Recomputed AUCs for the earlier 15 patients agreed with their published values (maximum difference 4.7×10⁻⁵); their mean AUC was 0.833 (95% CI 0.779–0.880). Across all 19 patients, mean AUC was 0.846 (95% CI 0.799–0.888) and median AUC was 0.840. The new-patient AUCs were within the earlier range for three patients; sub-22 (0.981) exceeded the earlier maximum (0.965). The plan was written before download, with an amendment made before downloading the added patients. This is a mouse-movement response in a random-dot decision task: because REST precedes stimulus onset, MOVE versus REST also reflects stimulus and decision activity, so the AUC is an upper limit for a pure movement detector. Each patient has one time-ordered split and windows locked to known event times; this is not an asynchronous, self-paced online detector. The HGA derivative’s preprocessing was not checked, electrode locations differ, and anatomy was not used. sub-08 was planned but not completed because the download stopped at the time limit; sub-16 and sub-11 were not tested. The cohort confidence interval treats patients as independent samples of one population.

stereoelectroencephalography · high-gamma activity · movement detection · linear discriminant analysis · held-out validation · AUC · random-dot task