A 10-second score offset gave lower mean balanced accuracy for a four-pair seizure reader in 28 SEEG subjects

We tested whether an early score offset improved a seizure reader in a separate SEEG record from the same subject. We used OpenNeuro ds004100 version 1.1.3 and 28 subjects with two records each. The reader used four bipolar pairs. New data supplied the first 30 seconds of all 56 records. Ten seconds was the fixed primary calibration duration. It gave mean balanced accuracy of 76.25%, compared with 79.02% for the frozen reader. The paired mean change was -2.77 percentage points. Its descriptive 95% subject interval was -6.74 to +0.58 points. Sensitivity fell by 6.43 points. Per-record AUC was unchanged. A fixed bipolar voltage gain cancels in the calibrated linear score under positive-feature and no-clipping conditions. This proof did not establish an accuracy gain in the sample. The study used selected clinical seizure records and a reused cohort. It gives no measured command or stimulation effect. The 22.3 MB package preserves new calibration features, audit voltage, fitted readers, all decisions, four figures, source identities, code, and repeat steps.

stereoelectroencephalography · seizure detection · calibration · balanced accuracy · linear reader · brain-computer interface · epilepsy