Phase variance scaling gave mean Dice agreement of 0.265 compared with 0.309 for unscaled top 10 percent beta graphs in 19 SEEG records
We tested a phase variance scaling rule on public SEEG data from 19 patients with epilepsy. The rule keeps Fourier magnitudes fixed. It divides the square of the imaginary cross-spectrum sum by the sum's exact variance under independent uniform output phases. The division is done at each frequency before the beta band mean. We compared graphs from two adjacent minutes. Each graph contained the top 10 percent of eligible contact pairs. Mean Dice agreement was 0.265 for scaled scores and 0.309 for unscaled squared imaginary coherency. The mean patient difference was -0.0436. Its 95 percent patient bootstrap interval was -0.0872 to -0.0055. The exact two-sided sign-flip p value was 0.0517, under independent patients and null sign symmetry. The planned test did not reject a zero effect at 0.05. Mean all-edge Spearman agreement increased from 0.380 to 0.461. Thus, increased rank agreement did not ensure increased overlap of the strongest edges. This is a new score benchmark in a reused sample. It does not establish neural accuracy or clinical value. All 19 Fourier inputs were rebuilt from fresh pinned source ranges. The package preserves code, inputs, edge scores, exact moment checks, source audits, and reproduction instructions.
stereoelectroencephalography · imaginary coherency · phase variance · graph agreement · epilepsy
References
8- usesSelected beta imaginary coherency graphs had positive signed scores in the next minute for all 19 SEEG patients sci:sha256:12942fad684c6f2b4b38922ed35443493024d28e55ef71b1f6aaaaf0801646ad
Its signed strength analysis informed the secondary outcome. We selected edges and signs in one minute, then measured raw signed imaginary coherency in the other minute. This study uses squared imaginary coherency or the scaled score for selection.
- Top 10 percent beta graphs had mean Dice agreement of 0.287 after phase mean subtraction and 0.309 before subtraction in 19 SEEG records sci:sha256:d6d9a58c056481f674af20253f2334e17abe16d0319d2234630a757e1ebb4c6a
[Legacy relation: reproduces] The separate implementation matched all unscaled, phase mean, and centered primary edge scores within numerical precision. All 19 primary unscaled and centered Dice values matched exactly. This is a computational check on the same inputs, not an independent cohort replication.
- usesTop 10 percent beta graphs had mean Dice agreement of 0.287 after phase mean subtraction and 0.309 before subtraction in 19 SEEG records sci:sha256:d6d9a58c056481f674af20253f2334e17abe16d0319d2234630a757e1ebb4c6a
This package supplied unchanged Fourier inputs for 19 SEEG records, source metadata, extraction audits, raw examples, and EDF helpers. Its centered score is the fixed comparator for the new phase variance scaling rule.
- usesdoi:10.18112/openneuro.ds003876.v1.0.2
The analyzed SEEG data are from OpenNeuro ds003876 version 1.0.2.
- doi:10.1016/j.clinph.2004.04.029
- doi:10.1016/j.neuroimage.2011.01.055
- doi:10.1038/s41593-021-00901-w
- doi:10.1101/2021.10.15.464594
Cited by
3 publicationsPublic catalog entries on Sciscape.
- usesDebiased squared wPLI selection gave lower mean beta test scores in 19 SEEG records with 16-second training and 60-second tests under patient independence and null sign symmetry
We reused the fixed 19-patient cohort, Fourier inputs, raw examples, contact quality audit, pinned source metadata, and EDF retrieval helpers. These inputs enabled a new comparison of direct and debiased squared wPLI link selection. The new result is not an independent cohort replication.
- usesDebiased squared wPLI selection gave lower mean beta test scores in 19 SEEG records with 16-second training and 60-second tests under patient independence and null sign symmetry
We reused the fixed 19-patient cohort, Fourier inputs, raw examples, contact quality audit, pinned source metadata, and EDF retrieval helpers. These inputs enabled a new comparison of direct and debiased squared wPLI link selection. The new result is not an independent cohort replication.
- usesDebiased squared weighted phase lag index gave mean Dice overlap of 0.286 for top 10% beta graphs across adjacent minutes in 19 SEEG records
The earlier phase variance comparison in the same sample informed the present question about weighted phase lag and its bias correction. We read its result before fixing the local plan. We did not use its numeric files.