Adding relative gamma power change did not raise within-shaft onset-label AUC in 15 test SEEG patients with two 60-second files

We tested whether relative gamma power change adds information for ranking seizure-onset labels. The base model used average gamma power and contact number. The source was public SEEG data from OpenNeuro ds004100 version 1.1.3. Each patient supplied two 60-second interictal files. The fixed split used 16 training patients and 15 test patients. Both models gave a mean within-shaft test AUC of 0.902431. None of the 472 tested comparison signs changed. A paired patient bootstrap fitted both models again in 5,000 draws. The 95 percent interval for AUC change was -0.013333 to 0.011111. The result does not establish equivalence or clinical benefit. All eight planned checks are reported. Fresh raw retrieval reproduced all 36 patient inputs and all 13 analysis outputs exactly. The package preserves code, window powers, labels, source records, raw retrieval recipes, results, and independent checks. This is a secondary analysis of reused patients, with an unknown time interval between files.

stereoelectroencephalography · SEEG · gamma power · epilepsy · within-shaft AUC · feature comparison · bootstrap uncertainty