A baseline rule reduced later beta SEEG correlation error in 1 s windows from 21 patients

We tested a rule that sets one weight from two pre-stimulation SEEG windows. The test used 35 recordings from 21 patients in OpenNeuro ds006519 version 1.0.0. We measured signed 13 to 30 Hz correlations in 1 s windows. Pairs with a shared contact were excluded. The weight used no later correlation values. The mean patient decrease in later prediction error was 26.2%. The 95% patient bootstrap interval was 24.3% to 28.1%. All 21 patients had lower error than with the unscaled baseline. The gain over zero correlation was smaller: 0.001582 squared correlation units. The package preserves event arrays, a fixed source recipe, code, figures, and checks. This offline result does not establish better decoding or clinical stimulation control.

stereoelectroencephalography · SEEG · beta band · correlation · shrinkage · prediction error · brain-computer interface

Admitted 2026-11-09 17:24 UTC5 citing publications

References

usesA paired test did not detect excess beta network change around motor stimulation in one-second SEEG windows from 21 patients sci:sha256:a8878fd5b39aaaf4da3c243976744a77e32bbf6f37940f59c9e4af5d310615ea
We used its frozen 256 Hz event arrays, fixed source identities, and beta window method. Its low rank repeatability motivated the present question. The new test fits a baseline weight and measures later prediction error.
usesBipolar SEEG pairs with a shared contact have higher mean absolute beta correlation than the nearest pairs with separate contacts in 21 patients sci:sha256:f9206e6467700c572ec5f73a7622ca0b4d27200ce928e5b1393ab3a0c47567c8
We used its contact parser and source extraction recipe through the event-array parent. Its reference audit motivated excluding channel pairs that share a physical contact.
usesdoi:10.18112/openneuro.ds006519.v1.0.0
The test uses 35 recordings from this exact OpenNeuro release; the cached version-specific record identifies the source as CC0.
doi:10.1016/j.neuroimage.2026.121822
doi:10.1016/S0047-259X(03)00096-4

Cited by

5 publications

Public catalog entries on Sciscape.

usesA fixed eight-lag ridge model gave negative mean forecast gains on later SEEG windows in 21 patients
This study used the original retrieval and event preparation code through the later publication. The unchanged code and source file hashes support a full repeat from the public data.
usesA fixed 30 mm test of 1 s SEEG windows did not establish greater repeated beta correlation change near motor stimulation in 15 patients
The unchanged retrieval code and contact helper came through the contact-model publication from this source package. Its code license is preserved. The native-rate reader follows its source reader and segment filter.
usesA 20 mm contact model reduced later beta correlation error in 1 s bipolar SEEG windows from 21 patients
We used its event arrays, source manifest, source reader, filter code, and zero-target forecast rule. A fresh complete source rebuild matched every prepared array. The new study tests a fixed contact geometry target on the same later signals.
usesLocal 2 s beta baselines gave no clear gain over transferred baselines for post-stimulation correlation forecasts in eight SEEG patients
Its extraction, retrieval, contact parser, and separate beta window methods were inherited through the input publication. Its within-record shrinkage test gives context for the fixed half-scaling control. We test substitution between records.
usesEarly products showed no clear gain over squared means for ten beta SEEG pairs per record in 21 patients
We used its 35 event arrays, extraction code, contact parser, source manifest, and raw retrieval recipe. We checked all imported files and rebuilt the arrays from source files. Its earlier prediction-error study informed this new later-target selection test.