Pre-movement high-gamma did not confirm left/right choice decoding in four new ds006253 SEEG patients (AUC 0.530; p = 0.31)

This preregistered replication tested whether all-channel 70–150 Hz high-gamma activity (HGA) in the 0.5 s before mouse-movement onset decodes left/right choice in new patients from OpenNeuro ds006253. An earlier secondary, uncorrected result in seven patients reported mean AUC 0.533. We applied the same shrinkage LDA pipeline with a time-ordered 60/40 split to four new patients available at the stop time. Their choice AUCs were 0.604, 0.550, 0.571, and 0.393; mean AUC was 0.530 and the exact one-sided sign-flip p-value was 0.31. The result did not confirm the earlier finding. With four patients, the smallest possible exact p-value was 0.0625, so this test could not reach 0.05; the result is not confirmed, rather than shown to be absent. Pooling the four new patients with the seven source patients gave exploratory mean AUC 0.534 (10/11 above 0.5, p = 0.038), which is not independent evidence. On error trials, the choice-trained score did not reliably follow choice or stimulus direction (mean AUC 0.517 new, 0.527 all; p > 0.1). A movement-detection control worked (AUC 0.877 new, 0.815 all 11). The effect is small and does not support single-trial BCI control from these all-channel pre-movement features. The package preserves the plan, code, derived features, results, figure, source hashes, and provenance. The analysis assumes the author HGA derivatives and event times are correct; one time-ordered split may be affected by drift, error-trial tests had low power, and H2/H3 tests were not corrected for multiple comparisons.

SEEG · high-gamma activity · choice decoding · movement intention · brain-computer interface · replication · OpenNeuro ds006253