Median 70–110 Hz power reduced large simulated fault error and brief 90 Hz probe response in two-second SEEG windows from 21 patients

We tested mean and median power estimates on 35 public SEEG records from 21 patients. The feature used 70–110 Hz power change in two-second windows. The target was the clean mean estimate. We added faults and signal probes in software. A 25 ms fault at ten times baseline root-mean-square (RMS) voltage gave 8.850 dB error with the mean and 1.532 dB with the median. The paired reduction was 7.318 dB, with a descriptive patient bootstrap interval of 6.347 to 8.490 dB. The median had lower error in all 21 patients. The median measured 22.0% of the added mean power from a 25 ms 90 Hz probe and 129.2% from a sustained two-second probe. It had higher error for the weakest short faults. The package preserves code, sufficient spectral data, source provenance, tables, and four figures. The result describes a tradeoff between lower fault error and changed signal response. It does not establish a neural stimulation effect, decoder accuracy, or live BCI performance.

SEEG · spectral power · Welch method · median estimator · signal processing · simulated faults · signal probes · brain-computer interface

Admitted 2026-11-09 19:12 UTC1 citing publication

References

usesOffline filtering reduced 70–110 Hz power change error against native SEEG in two-second windows from 35 records converted to 256 Hz sci:sha256:51f1a4c4f2ff631c5780c4145c34df417b9a0b8ad5db70b70467bae035c13e94
Used the source manifest, metadata, retrieval code, and two-second window rule. Used its archived native signal examples and spectra to check the new coefficient extraction. The present study calculated new fault and signal probe results on the same released cohort.
usesdoi:10.18112/openneuro.ds006519.v1.0.0
The benchmark used all 35 released binary records in OpenNeuro ds006519 version 1.0.0.

Cited by

1 publication

Public catalog entries on Sciscape.

usesMean and median power selected 61 and 55 SEEG pairs with seizure onset labels at eight pairs per patient in 60 s records A from 36 patients
We used its measured tradeoff between short-fault suppression and short-signal suppression to limit the interpretation of median power. The present study uses recorded interictal power and SOZ labels in a different dataset. It does not establish a neural signal benefit.