Early gamma increased mean onset label AUC by 0.067 beyond baseline power and contact position in 16 ds004100 SEEG test patients

We tested whether early seizure gamma adds contact ranking value beyond baseline power and contact position. We used two seizures from each of 32 patients in public OpenNeuro ds004100 version 1.1.3. A fixed logistic model trained on 16 patients and tested on 16 other patients. Adding early power increased equal-patient mean onset label AUC from 0.74975 to 0.81649. The mean paired gain was 0.06674. The 95 percent patient bootstrap interval was 0.02641 to 0.10888. This interval assumes independent test patients and holds the trained models fixed. The exact sign test gave p = 0.009155 under independent patient effects and null sign symmetry. Eleven test patients had a gain and five had a loss. The early-only model had nearly the same AUC as the combined model. Within named electrode groups, mean AUC changed by -0.00162 in 14 test patients. This secondary result does not show improved ranking within a group. The result does not establish treatment benefit or generalization to another cohort. The local plan was fixed before the new fits, after earlier results from this cohort were known. All 64 source records passed a separate signal check. All 896 seizure AUC values and 28 model fits passed independent checks. Two complete offline runs produced identical files. The package preserves code, source arrays, metadata, retrieval recipes, models, predictions, inference draws, figures, and reproduction instructions.

stereoelectroencephalography · SEEG · gamma power · seizure onset labels · contact ranking · patient-level validation · logistic regression