The gamma timing score gave no clear onset mark AUC gain over power and position in 14 test SEEG patients

We tested whether the first sustained gamma increase adds rank information for clinical seizure onset marks. We used public SEEG data from OpenNeuro ds004100 version 1.1.3. Two fixed ridge models used data from 14 training patients. We tested both models on 14 other patients. Each patient supplied two seizures. The base model used baseline power, mean and peak power change, and contact position. The added score used the first pair of two-second windows with gamma power at least four times baseline. The search used the first 20 seconds after onset. Mean test AUC rose from 0.8022 to 0.8060. The mean gain was 0.0038, with a 95% patient bootstrap interval from -0.0047 to 0.0124. The exact patient sign-flip p value was 0.4252. Seven fixed descriptive checks also had intervals that included zero. This exploratory analysis gives no clear evidence of an added AUC gain for this score and model. The target was clinical marks. The result does not establish surgical benefit or equivalence. The package preserves all input arrays, source recipes, code, fitted models, predictions, and uncertainty draws.

stereoelectroencephalography · seizure onset · gamma power · clinical onset marks · AUC · ridge logistic regression · patient bootstrap