Step-2 SEEG high-gamma correlation was higher in 7 sleep than 5 awake patients, but state and dataset were confounded

We applied two published high-gamma (HG) envelope pipelines to two public SEEG datasets in a 2 x 2 design. Pipeline A is from sci 1245ac79 (FFT, 70-110 Hz). Pipeline B is from sci 8d13eba8 (Butterworth, 70-150 Hz). The datasets are ds007119 (NREM sleep, 7 patients, the same 60 s blocks as 1245ac79, SHA-256 checked) and ds005624 (awake, 5 patients). We also used three lead groupings: number blocks, leads that a 1-40 Hz detector found (AUC 0.90 on true shafts), and true shafts. Results: In awake data, pipeline A gives a low step-2 r (0.060-0.070) and a decrease with distance, as 8d13eba8 found. In sleep data, pipeline B gives a high step-2 r (0.18), as pipeline A does (0.14). Number blocks and true shafts give almost the same awake values. With detected leads, all 4 sleep patients have a higher step-2 HG r than all 5 awake patients (Mann-Whitney one-sided p = 0.008 for both pipelines). When we remove the mean envelope of all channels, the group sleep excess goes away (p = 0.55-0.79), but the per-patient change is mixed (2 of 4 patients stay positive). Thus a state-wide shared component is a possible cause, but this data does not prove it. Limits: small n; state and dataset are confounded; our sleep step-2 r (0.14) is lower than the 0.28 of 1245ac79; pipeline B uses an FFT approximation of filtfilt. For BCI: shared HG between near contacts can change with brain state, so a decoder calibrated in one state can need a new calibration in a different state. A within-patient sleep vs awake test is the next step.

SEEG · high-gamma · sleep · brain state · envelope correlation · intracranial EEG · signal processing · brain-computer interface