Three-second averaging raises AUC for seizure windows 2 to 20 seconds after onset in 28 patients with SEEG recordings

This exploratory study compares a fixed causal three-second mean with a one-second spatial SEEG score in 28 patients from OpenNeuro ds004100 v1.1.3. The main test uses each patient's second recording and local pre-seizure calibration. Both rules test 18 seizure windows from 2 to 20 seconds after onset and 50 pre-seizure windows. Mean patient AUC rises from 0.760 to 0.798. The paired gain is 0.0384, with a descriptive 95% patient bootstrap interval of 0.0219 to 0.0555. AUC rises in 21 patients, falls in 6, and is unchanged in 1. Mean pre-seizure alert windows also rise from 8.86% to 14.57%. Thus better discrimination does not give fewer background alerts at the tested thresholds. Fixed first-record and transferred-calibration checks also raise mean AUC. The package preserves matched band powers, source hashes, code, figures, all scores, results, the original plan, and a change made before temporal results. A missing feature gap prevents a test of the first two seconds. This result does not establish continuous false alarm rates or safe brain stimulation.

stereo-EEG · seizure detection · AUC · causal averaging · spectral power · retrospective study