A separate seizure baseline lowered mean early gamma onset-label AUC in 32 SEEG patients under common average reference

A baseline from another seizure gave lower mean early gamma onset-label AUC than a baseline from the same seizure. We tested 64 seizures from 32 SEEG patients in OpenNeuro ds004100, version 1.1.3. The primary signal used common average reference and the first ten seizure seconds. The sample had 3,213 matched anchors with stable supplied onset labels. Mean patient AUC was 0.8164 for absolute power, 0.6875 for the own baseline, and 0.6448 for the other baseline. Other minus own was -0.0426 AUC, with a 95 percent patient bootstrap interval from -0.0749 to -0.0148 and Holm adjusted exact probability 0.00554. Other minus absolute was -0.1716 AUC, with adjusted probability 0.00000378. All ten planned method checks had negative mean differences. The other baseline had higher top-ten-percent precision and recall than the own baseline. The exact tests assume independent patients and null sign symmetry. The comparison is retrospective and uses the prior patient cohort. The target is supplied label agreement, not clinical benefit. Independent checks passed all 2,816 AUC values and all 757,770 source powers. Two repeat runs gave identical numerical file hashes. The package preserves the plan, code, feature arrays, metadata, fixed raw retrieval recipes, results, and checks.

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