Mean onset channel overlap was 0.093 lower across spontaneous seizures than within seizures for fixed expert pairs in 11 adult SEEG patients

We tested whether expert onset channel labels agree less when the seizure changes. We used public SEEG annotations from NEMAR nm000402, version 1.0.0. The main test had 32 spontaneous seizures, 31 seizure pairs, and 11 adult patients. Each comparison kept the patient and distinct expert pair fixed. Patients had equal weight. Mean Jaccard overlap was 0.579 within a seizure and 0.485 across seizures. The mean difference was 0.093. Its 95% patient bootstrap interval was 0.012 to 0.214. The exact two-sided sign-flip probability was 0.03125. Intervals assume independent, exchangeable patients. The sign-flip test also assumes symmetric null signs. Eight patient differences were positive, two were negative, and one was zero. One patient supplied 59% of the sum of differences. The first-two-seizure check had an interval that included zero. The result concerns expert label agreement in this selected cohort. It does not identify a biological cause. We provide exact source files, new comparison tables, code, independent checks, and repeat instructions.

stereoelectroencephalography · seizure onset · expert agreement · Jaccard overlap · reproducibility · epilepsy