With a common average reference, adding early gamma change gave mean onset label AUC 0.843 versus 0.803 within name groups in 14 test SEEG patients

We tested whether early seizure gamma change adds seizure onset label ranking information to baseline gamma power and relative contact number. We used public SEEG data from OpenNeuro ds004100 version 1.1.3. A fixed split gave 14 training patients and 14 separate test patients. Each patient had two selected seizure records. We fit two fixed pairwise linear models. We tested only contact pairs in the same name group with different clinical onset labels. With a common average reference, mean test AUC was 0.803 without change and 0.843 with change. AUC measures correct pair ranking, with half credit for ties. The mean gain was 0.040. Its 95% patient bootstrap interval was -0.015 to 0.094. The planned exact mean sign flip test gave p = 0.195. Eight patients improved, one declined, and five had the same AUC. The mean benefit remains uncertain. The interval conditions on the fixed training data and assumes independent test patients. The sign flip test also assumes null sign symmetry. The test patients were separate from the new fits, but the source cohort had been studied before. This result does not test clinical decisions or treatment outcome. The package preserves code, metadata, features, models, predictions, patient results, and source retrieval recipes. Fresh raw extraction matched all 31 source feature files byte for byte. Independent signal and model checks passed. All 11 new numerical outputs reproduced exactly.

stereo-electroencephalography · seizure onset zone · gamma power · feature addition · patient-level validation