Across 19 SEEG patients, high-gamma power was less correlated across pairs 10–20 mm apart, but its decay length was not shorter

Question: for SEEG read-out, how fast does the redundancy between band-power features of two bipolar pairs decrease with distance, and is high-gamma power more local than low-frequency power? Data: archived per-word encoding features (0–1.6 s, log10 Morlet power, 8 frequencies 3–140 Hz) of 19 patients from OpenNeuro ds004789 FR1 session 0 (sci c4b8c884), 2306 bipolar channels, native coordinates. Method (plan written before analysis): per-list mean removal, z-scoring, Pearson r over words for all pairs that share no contact, against midpoint distance; per-patient fit r = F + A exp(−d/lambda); exact Wilcoxon tests over patients. Results: median lambda was 19.5 mm for 3–5 Hz and 15.3 mm for 80–140 Hz (HIGH shorter in 11/19, p = 0.33, primary prediction not supported). At 10–20 mm, median r was 0.275 for 3–5 Hz and 0.134 for 80–140 Hz (HIGH lower in 16/19, Holm p = 0.021). The far floor (≥ 50 mm) was also lower for 80–140 Hz (0.034 vs 0.099, Holm p = 0.032). 8–22 Hz power was the most redundant between neighbours (median r 0.35–0.41 at < 10 mm). Thus high-gamma features from pairs 1–2 cm apart share little word-to-word variance, but the decay length is similar in both bands. Limits: one task, offline, bipolar montage, native-space distances, no volume-conduction control except removal of pairs that share a contact, no stimulation data. The package has code (numpy and matplotlib), input features with checksums, plan, per-patient tables and a figure.

stereo-EEG · band power · word-to-word correlation · spatial dependence · high-gamma · neural signal processing · free recall · brain-computer interface