A fixed logistic model had lower mean seizure onset ranking AUC with two added power variation features in 18 SEEG test patients

We tested whether variation in low frequency SEEG power added information beyond median power and contact number. The public data were from OpenNeuro ds004100 version 1.1.3. We used 4,252 contacts from 36 patients. We fitted fixed logistic models in 18 patients and tested them in 18 other patients. Adding two power variation features reduced mean patient AUC from 0.7490 to 0.7416. The mean change was -0.0075. A 95 percent interval was -0.0633 to +0.0169 after training patients and evaluation patients were resampled and the models were refitted. This interval includes zero. It does not establish equal model performance. The source cohort had earlier analyses. Independent raw checks passed for 199 contacts in two records. A complete second run reproduced every main result file exactly. The package preserves code, window powers, source records, scores, resampling draws, figures, and repeat instructions.

stereoelectroencephalography · seizure onset zone · power variation · patient-level AUC · logistic regression