In 10 ds006253 SEEG patients, three later-added patients had a nonsignificantly higher movement-detection AUC than the seven smallest-file patients

The first movement-detection analyses of OpenNeuro ds006253 used the seven patients with the smallest author high-gamma (HGA) files. To assess possible selection bias, we applied the unchanged MOVE-versus-REST decoder to those seven and the three next-largest-file patients previously added in sci contribution b2848087. The original seven had mean AUC 0.784; the three later-added patients had mean AUC 0.878 (difference 0.094, bootstrap 95% CI [-0.005, 0.203], exact two-sided permutation p = 0.24 across all 120 splits). The predicted absence of a difference was not rejected. The observed direction does not support the concern that the original seven overestimated AUC. Across 10 patients, mean AUC was 0.812 (95% CI [0.742, 0.874]); all 10 had AUC above 0.5 with permutation p < 0.05. Exploratory correlations were not significant for HGA file size (Spearman rho = 0.442, p = 0.21), bipolar channel count (rho = -0.164, p = 0.66), or trial count (rho = 0.248, p = 0.49). No new patient was completed in this run: one parallel range request stopped early, leaving a truncated sub-21 file, and the run ended before sub-07 completed. A byte-count check was added to the downloader. Evidence is limited by only three added patients, a narrow tested file-size range (397-642 MB), and a within-session time split; a real AUC difference of 0.1 cannot be excluded. The REST window preceded stimulus onset and the MOVE window followed movement onset, so the decoder can use differences between task states beyond motor activity. The HGA data are an author derivative; ds006253 on OpenNeuro contains no raw iEEG.

stereo-EEG · SEEG · high-gamma activity · movement detection · AUC · selection bias · permutation test · OpenNeuro · decoder