A contact number rule ranked adjacent seizure onset contacts better than a fixed power score in two short SEEG recordings from each of 34 patients

We tested whether a fixed delta to slow power score could rank adjacent clinical seizure onset labels. We used the first 120 seconds of two public SEEG recordings from each patient in OpenNeuro ds004100, version 1.1.3. The local test had 126 opposite-label pairs on 90 named shafts from 34 patients. Each pair had consecutive contact numbers. Each patient had equal weight. A fixed rule predicted seizure onset for the lower contact number. It used no signal and no fitted parameter. The rule gave 78.0 percent correct ranks. The fixed power score gave 50.8 percent. The paired difference was 27.3 percentage points, with a 95 percent patient interval of 14.3 to 40.2 points. The corrected primary p value for this difference was 0.00068 under the stated sign-flip assumptions. For the same patients, full contact AUC was 0.648. Thus, whole implant association and local label ordering gave different estimates. The result applies to this fixed score, these short recordings, and the source clinical labels. Contact numbers are not verified physical distances. The score interval permits positive effects and does not establish equivalence to chance. The package preserves source features, a reusable pair table, code, provenance, figures, a pinned raw check, and reproduction commands.

stereoelectroencephalography · seizure onset zone · intracranial EEG · contact ranking · spectral power · adjacent pairs · patient-level inference