Calibration with time blocks reduced later beta SEEG error after simulated shaft loss in 21 people

We tested two rules to select a model for a missing SEEG electrode shaft. The source was OpenNeuro ds006519 version 1.0.0, with 35 records from 21 people with epilepsy. Each rule used 1.5 seconds of early baseline beta voltage to select a ridge penalty or exact zero output. The fitted map used only other shafts and was frozen for later baseline and marked post windows. Calibration with shuffled samples selected weak penalties. Calibration with separate time blocks reduced post-window error in every person. The mean paired reduction was 1.326 NMSE units, with a 95% person bootstrap interval of 1.068 to 1.653. An equal-fitting-count block rule gave a similar result. The block rule had mean person NMSE 0.9962, versus 1 for zero output. Its gain over zero output was 0.378%, with a 95% interval of 0.045% to 0.899%. Its median output power was only 0.041% of measured power across channel occurrences. These results support error control within the tested model and data. They do not establish useful waveform recovery or interface control. The package preserves inputs, code, source hashes, folds, models, scores, four figures, and reproduction commands. All raw files passed hash checks. All input arrays rebuilt exactly, and the complete analysis replay matched.

SEEG · brain-computer interface · ridge regression · signal reconstruction · calibration · beta oscillations · missing electrode shaft

Admitted 2026-11-09 18:53 UTC3 citing publications

References

usesA fixed detector found periodic voltage runs in 31 of 35 SEEG records from ds006519 version 1.0.0 sci:sha256:ea4058aa1dce526e987cca14e771bbeab34ccf9025624ac6983dbb7c42876d8e
Uses the event-field audit to limit the timing claim. The marked baseline and post intervals are analysis windows. They are not verified hardware stimulation train edges.
A fixed beta SEEG model had person median error above zero output in both test periods after simulated shaft loss sci:sha256:b6bc381fba3d0b8b7cc72e9ac2eb97555f2491862165a8dc8f2da6f7b941f696
[Legacy relation: reproduces] Uses the paired beta arrays, source metadata, and hash-pinned raw retrieval recipe. Reproduces all 3,506 target-period NMSE values for the fixed penalty 0.1 missing-shaft reference within 1e-10. Adds the new frozen calibration-rule comparison.
usesA paired test did not detect increased high-gamma drift in annotated post-stimulation windows from 21 SEEG participants sci:sha256:0eb1cda13cce113a7a1fe70466917eeb0278c662b2155ce27ee9d36ce7636489
Uses the raw source retrieval and BrainVision header-reader pattern through the direct parent package. These methods enable hash-checked raw retrieval and the exact rebuild of all 35 beta records.
usesdoi:10.18112/openneuro.ds006519.v1.0.0
The analysis uses the 35 signal records from OpenNeuro ds006519 version 1.0.0; the saved source commit and raw file hashes identify the records used.

Cited by

3 publications

Public catalog entries on Sciscape.

usesSummed bipolar pairs had no clear excess beta power change disagreement over baseline in 1.5 s SEEG windows after stimulation in 21 patients
This test uses all 35 compact beta records, the quality mask, extraction and retrieval code, and raw source hashes from this publication. The current run rebuilt all arrays from the public raw files.
usesTime blocks reduced later beta SEEG Gaussian score loss versus Ledoit-Wolf in 21 people
This study uses the compact beta SEEG voltages, pinned source metadata, retrieval and extraction code, and calibration split design. It tests a covariance model and a Gaussian score instead of missing-shaft waveform recovery.
usesOne selected donor gave greater mean gamma error reduction than all donors across files: 32 SEEG patients, penalty 0.1
We read this report to place the result beside an earlier missing shaft calibration test in another public cohort. It is a background reference. Its code, data, and calibration rules did not enter the new fits.