In two resting SEEG recordings, skipping the low-power onset improved band-power baselines, while longer rest slowly reduced mean error

We assessed how resting calibration duration and recording onset affect per-channel mean and standard deviation estimates for z-scored SEEG band power. The analysis used two resting runs (433 s and 440 s) from OpenNeuro ds005670 snapshot 1.0.0, with 213 and 156 consecutive same-shaft bipolar pairs, respectively. We calculated log10 mean power in non-overlapping 1 s Hann-windowed FFT windows for 70–150 Hz high gamma and 13–30 Hz beta, excluding ±3 Hz around 50, 100, and 150 Hz. The last 200 s of each run defined each pair's reference mean and standard deviation. The first 1 s window in both recordings had low power (median z from −3.1 to −4.6; 68–77% of pairs below z = −2), and window 2 was also low in part. A 10 s baseline starting at recording onset had median mean offsets of 0.46–0.82 test-block SD in high gamma and 0.52–0.65 SD in beta, compared with 0.25–0.32 SD for 10 s baselines at other start times. After skipping the first 10 s, 10 s baselines had median offsets of 0.34 SD in high gamma and 0.27–0.35 SD in beta; 40 s baselines had offsets of 0.21–0.36 SD. Improvement was gradual: 160 s baselines still had median offsets of 0.15–0.22 SD and 90th-percentile offsets of 0.43–0.50 SD. In sub-02, other-start-time offsets remained 0.26 SD from 20 to 80 s, consistent with slow drift; in sub-01, they agreed with the time-shuffled control. Standard deviation estimates required more data: median absolute log2 SD ratio was 0.39–0.61 at 40 s and 0.12–0.22 at 160 s, including the onset window. These results describe two patients, one approximately 7 min run each, offline; the reference block came from the same run, so drift over hours or days was not tested. The cause of the onset artifact is unknown and may be specific to the amplifier or export start. No artifact rejection or anatomy analysis was performed.

SEEG · brain-computer interface · band power · z-scoring · resting calibration · baseline length · signal processing