Adding early gamma power increased mean clinical onset AUC above baseline alone in fixed models for 16 test SEEG patients with common average reference

A fixed gamma model with baseline and early seizure power had higher mean clinical onset AUC than baseline power alone in 16 test patients. Sixteen other patients trained the models. Each patient supplied two seizures from public OpenNeuro ds004100 version 1.1.3. The primary common average SEEG comparison used baseline power from -30 to -10 seconds and early power from 0 to 10 seconds. Mean test AUC increased from 0.723 to 0.797. The paired patient gain was 0.073; its 95 percent patient bootstrap interval was [0.017, 0.130]. The exact two-sided sign-flip probability was 0.02747. These estimates condition on fixed training models and assume independent test patients; the test also assumes symmetric null differences. Early power alone had mean AUC 0.795. Thus, the combination showed little mean gain over early power alone. Prior cohort summaries informed the question. This is not an independent cohort replication or a test of treatment benefit. The package preserves all-contact power, predictions, code, source retrieval records, figures, and passed checks of all 64 source files.

stereoelectroencephalography · seizure · gamma power · clinical onset · area under the curve · patient-level validation · OpenNeuro ds004100