Early gamma change increased mean onset mark AUC under all allowed pair labels in fixed tests of 32 SEEG patients

Clinical contact marks do not supply one binary label for every bipolar SEEG channel. We tested whether pair label choices can remove the added onset mark agreement from early gamma change. We used 64 seizures from 32 patients in public OpenNeuro ds004100, release 1.1.3. There were 3,212 adjacent pairs, including 111 pairs with different endpoint marks. Both models excluded the complete test patient and used equal endpoint pairs for fitting. We held equal endpoint labels fixed and permitted either binary label for each mixed pair. Labels were shared across both records and both models. Exact mean AUC gain limits were 0.030055 and 0.085369. A 1,000-draw patient bootstrap fitted both models again. The percentile 95 percent interval for the minimum allowed gain was 0.004587 to 0.059806. It assumes independent patients, fixed source selection, and the stated label constraints. Six patients had negative gain under every assignment. Planned reference and ridge checks retained positive minimum group gains. The package supplies an exact bound tool, its proof, attained labels, code, source powers, raw retrieval recipes, and full results. Direct label enumeration checked the bounds. Fresh raw checks passed for all 64 records and 757,770 power values. A complete analysis repeat gave identical bytes for all eight numerical result files. This is a secondary analysis of the same source patients. It concerns clinical mark agreement and does not establish treatment benefit.

stereoelectroencephalography · seizure onset · gamma power · area under the curve · label uncertainty · paired model comparison · patient bootstrap · exact bounds