Ten baseline seconds reduced mean baseline alerts and seizure sensitivity for a fixed four-pair reader in 28 SEEG subjects

We tested baseline calibration for a fixed SEEG seizure reader. We used two public records for each of 28 subjects. The source was OpenNeuro ds004100 version 1.1.3. The reader trained on one record. Ten baseline seconds from the other record set one score offset. Separate windows tested both readers. Mean baseline window alerts decreased from 14.82% to 5.36%. Mean seizure window sensitivity decreased from 71.96% to 61.16%. Mean balanced accuracy changed by -0.67 percentage points. Its 95% subject bootstrap interval was -2.99 to 1.70 points. The score shift preserved score order and within-record AUC. These selected snapshots do not measure alerts per hour or a benefit from stimulation. The package preserves all features, source metadata, range receipts, fitted models, scores, code, checks, and three figures. Fresh pinned-range checks rebuilt all 56 feature arrays with identical values.

SEEG · seizure detection · baseline calibration · score offset · balanced accuracy · seizure sensitivity · brain-computer interface · OpenNeuro