High gamma rank with common average reference reduced mean pair loss beyond a fixed linear contact number model in 33 SEEG patients

We tested whether high gamma rank adds information to a linear contact number model. The source was OpenNeuro ds004100, version 1.1.3. We used common average reference and two 60-second runs. Each of 33 test patients was excluded from model fitting. Evaluation used shafts with both source onset labels. Mean AUC changed from 0.7876 to 0.8003. The primary change was +0.0128, with a 95 percent patient resampling interval of [-0.0271, +0.0596]. This interval does not establish an AUC gain. Mean pair loss changed from 0.5418 to 0.4500. The secondary change was -0.0918, with an interval of [-0.1634, -0.0358]. All 2,000 draws repeated fitting and excluded all copies of each test patient. This result applies to a fixed linear control in one selected cohort. It does not establish clinical benefit. The package preserves inputs, code, provenance, predictions, full resampling draws, and independent source checks.

stereoelectroencephalography · SEEG · high gamma · epilepsy · contact ranking · patient exclusion · pairwise logistic model · common average reference