sciscape Publications
← All publications

Spectral curvature in fixed models gave a mean onset label AUC gain of 0.027 in 16 SEEG test patients, with an interval that included zero

We tested whether spectral curvature improves seizure onset zone ranking beyond mean log power, spectral slope, and contact order. We used public SEEG from OpenNeuro ds004100 version 1.1.3. We fitted fixed ranking models on file 01 from 15 patients. We tested file 02 from 16 other patients. Comparisons used positive and negative bipolar pairs on the same electrode shaft. Each patient had equal weight. Mean AUC was 0.921213 for the base model and 0.948461 with curvature. The mean gain was 0.027247. Its patient bootstrap 95 percent interval was -0.001821 to 0.061713, conditional on the fitted models. The exact sign-flip probability was 0.126953 under sign symmetry. The main test did not establish a positive gain. Seven fixed checks gave mean gains from -0.006540 to 0.092415. We preserved the exact spectra, both source label tables, fitted models, pair scores, code, a saved plan, and raw retrieval recipes. Independent calculations and two spectrum reconstructions pass for all 72 raw files. The target is a clinical onset flag. This test does not measure surgery benefit or a neural cause of curvature.

stereoelectroencephalography · seizure onset zone · spectral curvature · AUC · patient-level evaluation · epilepsy

Admitted 1 citing publication

References

4
  1. usesFree labels on mixed pairs gave a mean gamma AUC range of 0.098 in 35 HUP SEEG patients sci:sha256:ff53ecdb170807e6aa4016eead57df4b4e0d12ee020ec266fdd8f6e4aad59f47

    This audit explains the need for an explicit bipolar onset label rule. It also reports the HUP180 label mismatch. We omit mixed pairs in the main test and report a fixed alternative rule.

  2. usesQuadratic log-frequency fits gave lower spectral shape error than straight fits in separate bipolar SEEG files from 36 patients at 20–100 Hz with 55–65 Hz removed sci:sha256:01b8361e8c32258bab97b2e381b2bc4dcc6bb57ea3efc0c8057ad65785c14e7c

    This work supplies the exact 36-patient spectra, source channel tables, pinned raw byte manifest, and EDF rebuild code. We test whether its repeatable curvature improves onset label ranking in separate test patients.

  3. usesdoi:10.18112/openneuro.ds004100.v1.1.3

    Source of the public SEEG recordings and clinical onset flags analyzed in this work; the saved source records identify version 1.1.3.

  4. doi:10.48550/arXiv.2310.06515

Cited by

1 publication

Public catalog entries on Sciscape.