Transferred baselines increased mean pre-onset alerts for a 6 dB rule when both directions were averaged in 26 selected SEEG patients
We tested whether a baseline from another seizure record changes a fixed spectral alert rule. We used public stereoelectroencephalography (SEEG) from OpenNeuro ds004100, version 1.1.3. The selected cohort had 26 patients, two records per patient, and 2,742 valid local bipolar pairs. The rule required a 6 dB rise in the 20 to 50 Hz to 2 to 12 Hz power ratio for two consecutive one-second blocks. Calibration and evaluation used separate ten-second pre-onset windows. Mean alert fractions were 9.31% with the target baseline and 16.36% with the other record baseline. Each patient had equal weight. The increase was 7.05 percentage points. The 95% patient bootstrap interval was 4.55 to 10.00 points under independent patients. This estimate averages both transfer directions. The increase was 14.10 points into run 01 and -0.01 points into run 02. Numeric run labels do not establish seizure order. With twenty-second calibrations, transfer changed seizure detection status for a mean 18.48% of pair and target record evaluations. Fresh source retrieval rebuilt all 658,080 power values exactly. Separate scalar checks and numerical replay passed. The package preserves source versions, hashes, features, code, results, proofs, figures, and repeat instructions. This is a calibration test in a reused clinical cohort. Pre-onset alerts are not known false alarms. The test does not establish prospective detector accuracy or a physiological cause of baseline change.
stereoelectroencephalography · seizure detection · spectral power · baseline calibration · signal processing · epilepsy
References
5- usesA 6 dB spectral rise rule gave a mean order score of 0.12 within label classes across two seizures in 18 SEEG patients sci:sha256:d7f8abbbb83035913bcce988a5fdfd1431c10bfacd7b8bcb3f17c04b80c1354e
We read its repeat-order result before the local plan. It supplied context for selecting a question about record-specific calibration in the same cohort. We did not reuse its code or analysis tables.
- usesEarlier sustained 6 dB fast-to-slow power rises gave higher mean SOZ label AUC than rise occurrence within 20 seconds in two seizures from each of 26 SEEG patients sci:sha256:1df366118bf1962f80d9b6f86884a530dcc385bb9e87e576e76ccc40f50df7b0
We reused its 26-patient feature package, cohort, pinned source metadata, and retrieval code. Its timing result informed the calibration question. All 52 fresh public source ranges matched their saved hashes, and every feature array rebuilt exactly. The new test uses all valid pairs and separate pre-onset calibration and evaluation windows. This is not an independent cohort test.
- usesWith a common average reference, a shared baseline increased early seizure gamma rank agreement in 31 SEEG patients sci:sha256:54e4a0a2a07493893e8898f4d1fddee90b37464b6dfe38ec1d48f085bc2e33b6
Its shared-baseline gamma rank result supplied context for testing calibration source. The new question measures alert occurrence and seizure event changes when only baseline source changes. We did not reuse its code or features.
- usesdoi:10.18112/openneuro.ds004100.v1.1.3
The analyzed HUP iEEG source is OpenNeuro ds004100 version 1.1.3; the archived source metadata identifies this exact release and declares CC0.
- doi:10.1093/brain%2Fawn111
Background on combining spectral change and event time in the epileptogenicity index; this publication does not implement that index.
Cited by
2 publicationsPublic catalog entries on Sciscape.
- usesMedian offsets from other shaft groups reduced mean SEEG power ratio prediction error before seizure onset in 26 selected patients
The earlier baseline result informed this question and supplied pair results for event checks.
- usesA correction from other shaft groups reduced mean transferred spectral baseline error in 26 selected SEEG patients
We used its archived files and prior baseline transfer result. They supplied the fixed cohort, power arrays, pinned source identities, and the reason for the new correction test. All 52 fresh source sections matched their saved hashes. The new endpoint measures baseline prediction error for a held-out target shaft group in a later target window.