In 15 SEEG patients, HGA time-course decoding was not significantly better than a 1 s mean

In an offline analysis of archived 70–150 Hz high-gamma activity (HGA) features from 15 patients in OpenNeuro ds006253, we compared a shrinkage LDA using five joined 0.2 s bins from 0 to 1 s after mouse-movement onset (ST) with the mean of those same bins (M1) and the mean from 0 to 0.5 s (M05). Trials without a valid answer or with missing values were removed; error trials were retained, and the chosen side was derived from stim_dir and correctness. For each patient, trials were ordered by stimulus onset and split 60/40 for training and testing. ST did not outperform M1 (mean difference +0.007 AUC, 95% CI −0.013 to +0.030; exact two-sided sign-flip p = 0.55) or M05 (+0.021, 95% CI −0.011 to +0.054; Holm-adjusted p = 0.48). Mean AUCs were 0.565 (ST), 0.558 (M1), and 0.544 (M05); ST passed a patient-level label-permutation test in 4/15 patients. Two additional pre-movement bins did not change mean AUC in the exploratory analysis. These results do not establish equivalence. The analysis is offline, aligned to known movement onset, and uses 0.2 s bins; side codes were not mapped to left/right. The exact source dataset release is not recorded.

high-gamma activity · stereoelectroencephalography · brain-computer interface · chosen-side decoding · shrinkage LDA · OpenNeuro ds006253