In awake bipolar SEEG, same-pair 4–30 Hz power did not track high-gamma baseline drift

We tested whether high-gamma (70–150 Hz) baseline drift in bipolar stereoelectroencephalography (SEEG) pairs is a broadband shift also visible in 4–30 Hz power, and whether low-band power can serve as a label-free reference. We analyzed archived features from 24 patients in OpenNeuro ds005624 session 0: 474 windows and 3,073 adjacent-contact pairs after the previously reported cleaning rules. Each window was 4 s, sampled once per minute. Offsets were measured from the first three windows. Across subjects, the median within-pair correlation between high-gamma and low-band offsets was −0.066 (8/24 positive; p = 0.010); for five-window running means it was −0.115 (8/24 positive; p = 0.068). The pooled slope was 0.32, below 1, despite being positive in 18/24 subjects (p = 0.0002). Using high-gamma minus low-band power as a reference worsened the late mean squared offset: median 20.2 versus 2.65 with fixed calibration, worse in 24/24 subjects (p < 0.0001). Its short-term standard deviation was 2.6 times that of high-gamma alone. A slope fitted on the first half of post-calibration windows also worsened second-half offsets (median log ratio +0.30; worse in 18/24; p = 0.0003). Global re-centering reduced the late offset (median 2.25 versus 2.65; lower in 24/24; p < 0.0001). In this dataset and setup, same-pair low-frequency power did not provide a useful label-free reference for high-gamma drift. Task periods were unknown, so task-related beta decreases and high-gamma increases could contribute to negative correlations. Sessions were 13–55 min, only one 4-s window was sampled each minute, analyses used bipolar pairs, anatomical labels were unavailable, and decoder accuracy was not tested. The low-band measure averaged three log-power bands; it was not a fitted 1/f slope or offset.

stereoelectroencephalography · high-gamma · baseline drift · bipolar recordings · signal processing · neural signals · OpenNeuro