Reliable 2–30 Hz peaks mark awake SEEG pairs with lower offline 100-ms phase prediction error

In 19 subjects with awake SEEG from OpenNeuro ds005624 (first session, one 60-s segment per subject), 958 of 2,203 adjacent-contact bipolar pairs (43%) had a reliable 2–30 Hz spectral peak at least twice the fitted aperiodic power in the full segment and both 30-s halves. Reliable peaks were most often in alpha (41%), then theta (30%), beta (24%), and delta (6%). Peak height and offline 100-ms phase prediction error were negatively correlated in 18 of 19 subjects (median Spearman rho = −0.33; exact two-sided sign test p = 7.6 × 10⁻⁵). Median error was 23.6° for reliable-peak pairs and 33.9° for other pairs. Relative to a fitted aperiodic-noise surrogate, error decreased by 11.6° for reliable pairs and 6.6° for other pairs; the gain was larger for reliable pairs in 18 of 19 subjects (p = 7.6 × 10⁻⁵). The phase filter was offline, and peak-frequency selection favored the real data in surrogate comparisons. The analysis used one segment per subject, lacked task timing, omitted five subjects, did not separate peak height from frequency, and included no anatomy; peaks at 2–3 Hz may be fit artifacts. These results describe intrinsic band-signal predictability and do not estimate performance from a causal phase estimator.

stereo-EEG · SEEG · phase prediction · neural oscillations · spectral peaks · aperiodic spectrum · bipolar recording · closed-loop stimulation