SEEG band-power drift was large in 4 of 8 patients and small in the other 4

We compared cross-session and within-session changes in SEEG band-power maps for eight patients with two to four sessions each (19 sessions total) in OpenNeuro ds005624 v1.0.0. We analyzed four 15-s segments per session, with the recording segments selected at 10%, 35%, 60% and 85% of recording duration. In four patients, the all-band between/within change ratio was 1.40–2.54; in the other four it was 0.89–1.09. All-band session identification accuracy ranged from 0.58 to 1.00 in the first group and 0.25 to 0.63 in the second. Group median ratios were 1.03–1.18 for individual bands and 1.24 for all bands. Session explained a median 17–29% of feature variance, compared with 8–10% under permutation; session-ID accuracy was 0.53–0.65, against mean chance of 0.45. In high gamma (70–170 Hz), one patient had a ratio of 4.7 and between-session map correlation of 0.24 (within-session 0.91), while four patients had ratios at or below 1.05. Removing global gain did not remove drift: median ratios remained 1.08–1.21, and alpha exceeded 1 in all eight patients (sign test p = 0.004). Results suggest patient-dependent cross-session drift at this 15-s scale. The sample is small, task markers and session dates are absent, and clinical activity, quantization, and unremoved channel artifacts may affect the measurements. The data do not establish how much time separated sessions or distinguish task from rest states.

SEEG · band power · cross-session stability · brain-computer interface · signal processing · electrophysiology · session identification · OpenNeuro