A fixed low frequency score added 0.034 to mean onset-label AUC above a coordinate and number model in 36 SEEG patient holdouts

We tested whether a fixed low frequency SEEG score adds information to a location and contact-number model. The source was OpenNeuro ds004100, version 1.1.3. The test used two 120-second interictal records per patient. We matched 4,209 contacts and 360 clinical onset marks in 36 patients. Each test patient was excluded from model fitting. Mean patient AUC was 0.712 for coordinates plus number and 0.746 after the EEG score was added. The mean gain was 0.034. A 95% interval from 1,000 patient bootstrap samples was 0.003 to 0.077. Each sample fitted the models again and removed all copies of its test patient from training. The lower bound is close to zero. Within 90 mixed-label shafts from 34 patients, the mean AUC change was -0.009. That secondary result has a different contact scope and no planned interval. The signal gain applies to the fixed control and supplied labels. It assumes complete onset marks and comparable source coordinate registration. Source coordinate units and reference space are unresolved. The original signal reference is unknown. This is one clinical cohort, not external validation or a test of surgery benefit. The package preserves all feature inputs, source metadata, immutable raw retrieval references, a checked raw example, code, complete results, and reproduction instructions.

stereoelectroencephalography · seizure onset labels · low frequency power · contact ranking · patient holdout · bootstrap · spatial bias