Adding high gamma gave an uncertain mean AUC gain of 0.023 beyond contact numbers in a fixed test of 35 SEEG patients

We tested whether high gamma power adds onset label information beyond electrode contact numbers. We used 70 one-minute public SEEG records from 35 ds004100 patients, version 1.1.3. A fixed logistic model used scaled contact numbers and number ranges. A second model added 80 to 150 Hz power. Both records from each test patient were excluded from model fitting. Mean patient AUC rose from 0.661 to 0.684. The mean gain was 0.023. Its 95 percent patient bootstrap interval was -0.011 to 0.056. The jackknife interval with all models fitted again was -0.010 to 0.055. Both intervals crossed zero. Within mixed shafts, the mean gain was -0.003 in 33 patients. The bootstrap kept predictions fixed. Both intervals are approximate. A planned low gamma check gave a larger descriptive gain of 0.060. The main result is a small, uncertain high gamma gain in this fixed test. It does not show model equivalence or clinical benefit. The package contains code, pinned inputs, raw retrieval recipes, predictions, and complete checks. Four raw power checks passed. A complete numerical repeat gave the same file hashes.

stereoelectroencephalography · high gamma · contact numbers · seizure onset · AUC · patient-level cross-validation