Adding 30–100 Hz spectrum curvature raised mean SOZ AUROC from 0.598 to 0.703 in patient holdout tests of 31 HUP SEEG patients
We tested whether spectrum curvature improves seizure onset zone (SOZ) contact ranking beyond slope. The public SEEG inputs were two 30-second interictal file sections per patient. The fixed logistic models used 2,879 contacts from 31 HUP patients. Each test patient was excluded from training. Mean patient AUROC rose from 0.598 to 0.703. The mean paired gain was 0.105. Its 95 percent patient resampling interval was 0.020 to 0.198. Each draw refit both models and excluded all copies of the test patient. The gain was positive in 18 patients and negative in 13. All eight planned checks retained a positive mean gain. The result applies to contact ranking in this sample. Clinical benefit and the cause of the curve were not tested. The package preserves input spectra, source recipes, code, scores, draw results, figures, and exact reproduction checks.
stereoelectroencephalography · seizure onset zone · spectrum curvature · spectrum slope · patient holdout · contact ranking · epilepsy
References
6- usesIn 30-second HUP SEEG samples, reference changes reduced mean 30–100 Hz slope rank agreement more than run changes sci:sha256:b9c7b4646dba303292f492835f91dbb8026d4f35fcce938519d1d3fa6bcbff63
The study uses 339 unchanged files from this archive. These files provide the exact SEEG spectra, extraction code, clinical metadata, and pinned public retrieval recipes. All parent file hashes were checked.
- usesA quadratic model reduced mean SEEG spectrum shape error between adjacent windows. Tests used 30 to 100 Hz and a common average reference in 34 patients. sci:sha256:de5f6464d3393d8a43ad2a95cacb1ffe0f3e8632193e766b8e42bce17535cb91
The adjacent-window quadratic shape result motivated the fixed curvature feature. This study tests that feature against clinical SOZ labels with patient holdout. It does not repeat the earlier spectrum prediction test.
- usesRepeated contacts gave higher mean AUROC than unseen patients for a fixed five-neighbor rule on 30-second SEEG spectra from 31 patients sci:sha256:db3385cbd16f5d51ca770d91d015d9d8cdf78e23094788f9c0bcde053342059d
The repeated-contact classifier result motivated exclusion of every test patient from training. This study applies that split rule to fixed slope and curvature models.
- usesdoi:10.18112/openneuro.ds004100.v1.1.3
The analysis uses the SEEG subset of OpenNeuro ds004100 version 1.1.3; its exact release identity and CC0 license are recorded in the preserved source metadata.
- doi:10.1038/s41593-020-00744-x
- doi:10.1523/JNEUROSCI.0146-25.2025
Cited by
0 publicationsPublic catalog entries on Sciscape.
No incoming citations on Sciscape.