Average references reduced low frequency seizure onset ranking AUC in two short SEEG recordings from each of 36 patients

We tested a fixed 2 to 4 Hz divided by 0.5 to 1 Hz power score in public SEEG data. We used the first 120 seconds of each of two recordings from 36 patients with epilepsy. The same 4,252 contacts entered all main comparisons. The mean patient AUC was 0.647 for stored signals, 0.602 for a common average reference, and 0.578 for a shaft average reference. The paired changes were -0.045 and -0.069. Both passed Holm correction for five planned tests. All three group AUC values exceeded chance. Stored contact ranks had a mean correlation of 0.507 between recordings. Their mean top 10 percent overlap was 0.363. Only 19 patients had an AUC greater than 0.5 in both recordings. We saved a sensitivity check for six discordant source labels. The package contains code, window-level features, source versions and hashes, results, figures, and a signal retrieval recipe. All feature hashes passed. A second EDF reader confirmed one source segment. A repeat analysis gave identical result file hashes. The study tests short recording label association. It does not test treatment benefit.

stereo-electroencephalography · seizure onset zone · low frequency power · reference effects · recording stability · area under the curve