Separate baseline limits reduced mean beta SEEG rate error at a 5% target in 21 people

We tested two limits for a beta SEEG signal monitor. The test used 34 public recordings from 21 people in OpenNeuro ds006519, version 1.0.0. Separate 1.5-second intervals fit the covariance model, set a measured limit, and tested later baseline flags. The primary model used time blocks to select covariance regularization. At a 5% target, mean absolute rate error fell from 20.15 to 6.96 percentage points. The paired reduction was 13.19 points, with a descriptive 95% person-bootstrap interval of 7.07 to 20.33 points. Error fell in 18 of 21 people. Mean flags fell from 24.10% to 9.47%, so the measured limit did not reliably achieve the target. Known Gaussian controls favored the Gaussian limit and showed a cost for short calibration. Post-stimulation flags do not establish sensitivity or a causal effect. All 105 raw files passed hash checks. A second filtering procedure reproduced all 34 arrays exactly. Separate arithmetic checked 1,088,000 decisions. A new process replay reproduced 77 numerical files exactly. A fresh isolated environment reproduced these files to numerical precision. The package preserves code, input arrays, scores, models, limits, figures, provenance, and reconstruction instructions.

SEEG · intracranial EEG · signal monitoring · beta band · covariance calibration · baseline flags · brain-computer interface

Admitted 2026-11-09 19:57 UTC1 citing publication

References

supportsMaximum score calibration left a mean absolute rate error of 8.2 percentage points at a 10% combined target in eight SEEG session pairs with 32 contact pairs each sci:sha256:7bab662ceb49541192c5dcd5a5bcea800a9856f2d9f1f13d1aa89582d07bcb2f
This study gives related evidence that a rank limit can leave substantial later SEEG rate error. Our test uses another dataset, a multivariate beta covariance score, and much shorter calibration. We used none of its data or code.
usesTime blocks reduced later beta SEEG Gaussian score loss versus Ledoit-Wolf in 21 people sci:sha256:8ce0e9fd2ce2cef0364cc2c88df73af194067c764e6f5c3a11fa6d08fb93bb5d
This study supplied the time-block covariance selection method, pinned raw retrieval manifest, source metadata, and inherited reader code. We extracted new chronological baseline intervals from freshly retrieved raw data and tested alert rates rather than Gaussian density loss.
usesA paired test did not detect increased high-gamma drift in annotated post-stimulation windows from 21 SEEG participants sci:sha256:0eb1cda13cce113a7a1fe70466917eeb0278c662b2155ce27ee9d36ce7636489
This publication is the original source of the inherited BrainVision header reader. The reader supplies native sample rates, bipolar channel order, binary layout, and voltage scales.
usesdoi:10.18112/openneuro.ds006519.v1.0.0
The 34 analyzed recordings were drawn from OpenNeuro ds006519, version 1.0.0; the publication preserves the pinned source commit, file hashes, and retrieval manifest.
doi:10.1016/j.neuroimage.2026.121822

Cited by

1 publication

Public catalog entries on Sciscape.

usesWith a fixed model, a beta SEEG spatial score did not change under common gain in 34 records and showed no clear mean post-label rise
We used the pinned raw manifest and source metadata and adapted the BrainVision reader and retrieval code. Its separate baseline limits motivated the monitor test. We extracted new windows with matched filter resets and did not use its filtered arrays or outcomes.