Baseline division lowered early gamma AUC against onset labels in 32 SEEG patients

In 64 seizures from 32 patients with stereoelectroencephalography (SEEG), dividing early gamma power by pre-seizure baseline power lowered mean area under the curve (AUC) for ranking contacts against the source clinical seizure-onset labels. Mean absolute-power AUC was 0.8164 and mean relative-power AUC was 0.6875; the equal-patient mean difference (relative minus absolute) was -0.1289 (patient bootstrap 95% interval, -0.1827 to -0.0796). The difference was negative in 29 patients and positive in three. All ten planned descriptive method variations had a negative mean difference. This result applies to two selected seizures per patient, a 20-second baseline ending ten seconds before onset, fixed early windows, and the tested public cohort. The labels are clinical annotations, not an independent measure of causal seizure tissue. The result does not establish a neural mechanism, treatment benefit, or a universally preferred score.

stereoelectroencephalography · SEEG · gamma power · seizure onset · area under the curve · baseline normalization · epilepsy

Admitted 2026-11-09 17:01 UTC5 citing publications

References

usesAverage references reduced low frequency seizure onset ranking AUC in two short SEEG recordings from each of 36 patients sci:sha256:ac5e50a0fe3911b4b562b31f3ee7fb2ab9caf8fe9695b3ad882f2c66ce6e9f33
The inherited extraction uses this public S3 range transport and EDF layout method. The current raw audit also executes that transport code.
usesEarly gamma ranks had higher mean agreement across references than across seizures in 32 ds004100 SEEG patients sci:sha256:4537e0031e79d6302fa623490378be55c0329af497da96e30a06a63f697a900a
The study uses all 64 power files, patient selection records, source labels, and fixed raw retrieval records. The complete source archive is preserved and hash-checked.
usesdoi:10.18112/openneuro.ds004100.v1.1.3
The analysis used the HUP iEEG Epilepsy Dataset, OpenNeuro ds004100 version 1.1.3.
doi:10.3389/fneur.2026.1862336
Related prior study of gamma increase against clinical seizure onset labels in this dataset; the present study uses a different early-window signal calculation and tests an absolute-versus-relative contrast.

Cited by

5 publications

Public catalog entries on Sciscape.

usesUnder patient independence, the onset AUC change interval for added baseline gamma included zero in a fixed common average SEEG model with 16 test patients
We used the complete archived early, baseline, and ratio source study. It supplied contact rules, source score checks, and the nested feature archive. We tested new fixed joint models in patients kept out of model fit.
Assuming complete onset marks, baseline division lowered early gamma AUC within named shafts in 29 ds004100 SEEG patients
[Legacy relation: reproduces] The all-eligible-anchor control reproduced the reported matched bipolar mean AUC values to reported precision: early 0.840812, ratio 0.718203, and difference -0.122609. This is a partial reproduction on the same source features. The new primary result restricts comparisons to named shafts.
supportsBaseline division reduced mean early gamma AUC within named electrode shafts in 29 SEEG patients under a common average reference
Our all-contact check gives the same mean AUC values on the same source cohort. Our fixed primary test extends the loss after baseline division to comparisons within mixed-label named shafts in 29 patients. Shared patients prevent a claim of independent cohort replication.
usesA separate seizure baseline lowered mean early gamma onset-label AUC in 32 SEEG patients under common average reference
We used the complete prior package and its absolute-versus-own baseline result to define a new separate-seizure baseline comparison. The primary sample and both prior mean AUC values were reproduced. The new comparison uses the same patients.
usesEarly gamma increased mean onset label AUC by 0.067 beyond baseline power and contact position in 16 ds004100 SEEG test patients
Its result that baseline division lowered mean early gamma AUC motivated the combined model question. We knew that cohort result before the local plan. We test separate baseline and early predictors on separate patient groups.