Adding time concentration to a fixed 80 to 150 Hz power model raised mean test AUC by 0.021 in 17 SEEG patients

We tested whether time concentration adds seizure onset label information beyond mean 80 to 150 Hz SEEG power. We used two 60 second clips from each of 34 patients in OpenNeuro ds004100 version 1.1.3. A fixed split gave 17 training patients and 17 separate test patients. Training used run 01. The primary test used run 02. Mean patient AUC rose from 0.7362 to 0.7576 after concentration was added. The mean gain was 0.0214. Its 95 percent patient bootstrap interval was 0.0019 to 0.0422, with training models held fixed. Eleven patients improved, three declined, and three had equal AUC. The planned two-sided sign test gave p = 0.0574. It did not detect an effect at the 0.05 level. The gain was 0.0085 in the other test recording. Comparisons within shafts gave no mean gain. The result does not establish a stable gain across recordings or clinical populations. This is a new analysis of a prior cohort, not an independent cohort replication. Pair selection used clinical labels and excluded mixed pairs. The package contains the fixed plan, all test predictions, code, source records, figures, and repeat instructions. Fresh retrieval of all 72 source prefixes rebuilt all 68 feature files with zero numerical difference. The package also contains independent numerical checks.

stereoelectroencephalography · seizure onset zone · time concentration · high-frequency power · patient holdout · area under the curve · intracranial EEG