Signed 10 ms covariance gave a mean patient error reduction of 3.56% in 19 selected SEEG records filtered at 1 to 40 Hz
We tested whether signed lag covariance in one interval improves covariance prediction in a later interval. We used 19 selected public SEEG records from OpenNeuro ds003876 version 1.0.2, with 2,067 contacts. Each native one-minute interval was processed separately at 200 Hz and filtered from 1 to 40 Hz. The two retained 56-second intervals had a four-second gap. At a fixed 10 ms lag, adding the first signed covariance term to a symmetric model gave a mean patient error reduction of 3.56%. Sixteen patients had positive gains. The median was 0.44%. The patient bootstrap 95% interval was 0.80% to 7.13%, under patient sampling assumptions. All nine fixed checks had positive means. Direct sample checks, synthetic controls, and an offline repeat passed. All 19 first source retrievals passed pinned-source checks. A second rebuild matched exactly for 11 patients; eight did not finish within the run limit. The package has code, source pins, covariance statistics, one raw prefix, patient results, pair maps, and reproduction instructions. The result shows repeatable second order temporal structure under this test. It does not establish neural causation, whole-process time irreversibility, or clinical value.
SEEG · stereoelectroencephalography · lag covariance · time reversibility · epilepsy · covariance prediction
References
6- usesRaw sink scores from a 10 ms ridge model had mean reversal rank correlation 0.841 in 19 filtered SEEG records sci:sha256:3d57be5f8d1e099afb6c038349049004f966440a839215cfd0c4885735d54d25
Uses the fixed cohort, source lock, EDF contact maps, and extraction code. The new extractor processes each native interval separately. The new test predicts signed lag covariance rather than sink ranks.
- usesLower contact numbers gave mean AUC 0.790 for public onset marks within inferred shafts in 19 external SEEG patients sci:sha256:2a48b5b2805c661fdcb824661f42d0fdcf0ade0dadbd4a88453e239cfc987f74
Uses the selection history of the 19-patient NIH and Johns Hopkins cohort. This history fixes the duration scope and documents the exclusion of short Miami records.
- usesdoi:10.18112/openneuro.ds003876.v1.0.2
Source dataset, version 1.0.2; its pinned files and metadata supply the 19 selected SEEG records analyzed here.
- doi:10.1093/brain%2Fawac300
- doi:10.1101/2021.10.15.464594
- doi:10.1038/s41593-021-00901-w
Cited by
4 publicationsPublic catalog entries on Sciscape.
- usesLag relations repeated beyond site scores in 19 SEEG patients and 50 resting fMRI adults; a ten-person film test failed
We reuse its fixed SEEG cohort, source covariance statistics, retrieval pins and voltage extractor. The new analysis uses bipolar squared correlations and residual-flow repetition.
- Human SEEG and BOLD lag scores did not support a shared N3 decrease in 56 patients and 16 records
Prior evidence for repeatable antisymmetric lag structure in other human SEEG data; no data or code reused.
- usesA sleep decrease in repeated lag asymmetry was not confirmed in 39 SEEG patients and 16 fMRI records
Its repeated signed lag covariance result motivated this sleep-state and second-modality test.
- usesUnder fixed prediction, signed 10 ms covariance reduced mean error by 1.89 percent for bipolar SEEG between shafts in 19 selected patients at 1–40 Hz
We used the 19 contact covariance files, raw EDF prefix, source pins, metadata, and extraction code. We added bipolar transforms and a mask for links between shafts.