In 10 SEEG subjects, all bands shifted more across days, while high-gamma was not significantly more stable than theta-alpha

We tested whether high-gamma SEEG band power is more stable between days than theta-alpha power, for BCI decoders without daily recalibration. Data: archived per-trial log band power (6 bands, 4-6 s maintenance, bipolar pairs) of OpenNeuro ds004752 v1.0.1, 10 subjects with same-day and cross-day sessions. For each session pair and band we calculated the feature shift (median |mean difference| / pooled SD) and the transfer AUC of a one-band shrinkage LDA (set size 8 vs 4). We wrote the plan before the results. Primary test H1: extra cross-day shift of 55-145 Hz minus 4-13 Hz, exact sign-flip test. Result: in each band the shift between days was 2.9-3.4 times the shift within one day (9 of 10 subjects per band). The extra shift was smaller for 55-145 Hz (median 0.30 SD) than for 4-13 Hz (0.52 SD) in 7 of 10 subjects, but H1 was not significant (p = 0.072). The AUC loss between days was small in all bands and did not differ between high and low bands (p = 0.42). Cross-day AUC was higher with 4-13 Hz features (median 0.64) than with 55-145 Hz features (0.54) in 8 of 10 subjects (p = 0.035, Holm p = 0.070). Limits: 10 subjects, day is confounded with time gap and clinical state, one window, one task and one decoder, offline only. The package has the code, features, checksums, per-pair and per-subject tables and one figure.

SEEG · intracranial EEG · band power · cross-day stability · working memory · brain-computer interface · LDA · OpenNeuro ds004752