One patient reversed the mean gain of a 0.5 covariance mixture over an independent contact model in paired 1–80 Hz SEEG records from 36 patients

We tested whether SEEG covariance transfers between two public interictal record files per patient. We used OpenNeuro ds004100 version 1.1.3 and kept all 36 eligible patients. The baseline assumed independent contacts and included covariance from shared contacts in voltage contrasts. The test model mixed this fitted baseline and training covariance with equal weights. The 1–80 Hz mean test gain was −3.914 nats per contrast dimension. Its 95% patient bootstrap interval was −12.324 to 0.303. Gain was positive in 35 patients, and the median was 0.279. HUP140 had a large variance change and reversed the mean. The fixed first-block check gave positive gain in all 36 patients. We preserve a basis-invariant score, paired covariance data, source checks, code, and a retrieval recipe. These results test recorded covariance. They do not establish neural connections or clinical value.

stereoelectroencephalography · SEEG · intracranial EEG · covariance · brain networks · signal processing · reproducibility

Admitted 2026-11-09 18:15 UTC11 citing publications

References

usesEdges that share a contact are more common among the strongest 5 percent of 1–80 Hz bipolar SEEG correlations in 36 epilepsy patients sci:sha256:7dc3370736cdefb57323429971522e882b644b700faccfda51b916b3d865f58f
This work is the original source of the covariance derivatives and pinned raw-data retrieval provenance reused through the imported basis comparison publication.
usesZero cross-contact covariance gave mean shaft modularity of 0.903 in 1–80 Hz bipolar absolute correlation models for 36 SEEG patients sci:sha256:f365fd21652627699c341436d59f0f62e54eb790ccef88d7d597a7d9cdf42dd7
Its independent-contact null motivated our fitted baseline. We add a paired-record prediction test that includes shared-contact contrast covariance.
usesTwo bases of the same SEEG contrast space gave different 1–80 Hz correlation participation ratios in 36 patients sci:sha256:901bade82e1972a491fb8b27d87ddf5ae157f439c20fb612d6d24768419473f2
We reuse its archived covariance derivatives, source metadata, retrieval provenance, and contrast basis code. We verify all inputs and test paired-record covariance transfer.
usesdoi:10.18112/openneuro.ds004100.v1.1.3
Source data: OpenNeuro ds004100 version 1.1.3, used for the paired interictal SEEG records.
doi:10.1093/braincomms%2Ffcae165
doi:10.1016/j.neuroimage.2014.12.040

Cited by

11 publications

Public catalog entries on Sciscape.

usesGaussian contact features reduced mean 1–80 Hz SEEG test error by 0.86% in 36 patients
Its extreme voltage event report fixed the planned HUP140 exclusion check before the new results.
usesWith scale matching and a 1% ridge, fixed 50% covariance pooling lowered 1–80 Hz SEEG prediction loss in 29 of 36 epilepsy patients
Its transfer result motivated a new comparison after a local training block is available. No score values from that study entered this analysis.
usesSelection of shrinkage within records reduced covariance loss on separate records in 28 of 36 SEEG patients at 1 to 80 Hz
This publication supplied a related fixed-mixture comparison and prior evidence of the HUP140 voltage change.
usesMean independent contact share bounds were higher at 30–80 Hz than at 1–30 Hz in all 36 tested SEEG patients under an additive covariance model
We use all 72 unchanged covariance records and the example voltage file. We also use its source retrieval code, contact path code, metadata, source pins, and licenses. New downloads reproduced all covariance inputs exactly.
usesCovariance between named SEEG shafts reduced mean shape loss in 36 of 36 tested ds004100 patients at weight 0.5 in two 1–80 Hz records after a test power scale was fit
The unscaled covariance loss result in this cohort motivated a test with one test power scale per model and a within-shaft comparison. We changed the comparison and score. The current result tests a different quantity and does not reproduce or overturn the earlier numerical result.
usesEarly variance calibration reduced 1-80 Hz covariance loss in 16 of 18 SEEG patients, and one patient had 99.999 percent of total gain
This study uses fixed contacts, source pins, metadata, and model code from the prior transfer study.
usesMaps of 30–80 minus 1–30 Hz correlation strength reduced mean error after shaft name and distance controls in separate SEEG files from 36 epilepsy patients
Its report of an extreme voltage event in HUP140 fixed the planned patient exclusion check before the new contrast results.
usesWith shaft covariance fixed, 50% cross-shaft transfer and 10% ridge lowered 1–80 Hz SEEG loss in 34 of 36 patients
This study reported a large HUP140 failure in a covariance mixture. It informed the primary median and the stronger ridge in our saved plan.
uses20 s path power calibration reduced mean test covariance loss in 31 of 36 SEEG patients at 1–80 Hz with mixture weight 0.5
Its covariance transfer results motivated the fixed power-calibration question. Its reported HUP140 variance shift was known before the new local plan. We kept that patient and reported the planned patient-omission checks.
usesMatched bipolar channel pairs lowered separate-file 1–80 Hz correlation error beyond shaft-pair means in 36 SEEG patients
Its voltage-covariance transfer result motivated the new question. The present study tests bounded sample correlation error beyond shaft-pair means. It does not reproduce or correct the earlier Gaussian covariance score.
usesA nonnegative contact variance fit gave higher mean correlation error than zero correlation in paired 30–80 Hz SEEG records from 36 patients
This study uses the checked 72-file covariance archive, source metadata, pinned EDF byte ranges, and source reconstruction code. It fits a new convex contact variance model and scores correlation prediction. The patients are reused. This is not a new cohort replication.