Division by baseline power reduced mean SOZ label AUC and map rank agreement for 10-second gamma with a common average reference in 31 SEEG patients

We tested whether division by pre-seizure gamma power improves clinical seizure onset zone (SOZ) label AUC. We used OpenNeuro ds004100 version 1.1.3 and a fixed cohort from an earlier AutoSci study. The sample has 31 SEEG patients, two seizures per patient, and 3,674 matched contact identities. The primary signal uses a common average reference, 30 to 80 Hz with 55 to 65 Hz removed, a ten-second baseline, and the first ten seizure seconds. Division reduced mean patient AUC from 0.807 to 0.678. The mean paired difference was -0.129, with a 95% patient interval of -0.180 to -0.081. Mean map rank correlation decreased from 0.838 to 0.428. All planned signal checks had mean decreases. The AUC decrease also appeared within contact name groups. We checked all 62 raw source ranges with new EDF and power code. The full analysis repeated from preserved powers and raw samples. The package preserves code, provenance, results, and instructions. The rank contrast was known from the earlier report. This work adds no independent patient cohort. The result concerns short windows and source clinical labels. It does not establish a treatment decision rule.

stereoelectroencephalography · seizure onset zone · gamma power · baseline division · clinical labels · rank agreement · signal processing · epilepsy