Separate 30 second corrections reduced mean error from a 5% high gamma crossing rate in three SEEG recording pairs

We tested a short correction for a power threshold in public SEEG data. We used six records from three people in OpenNeuro ds003848 version 1.0.3. The test used 156 adjacent contact pairs and later data that did not set the correction. The power band was 70 to 110 Hz. Each starting threshold had a 5% calibration reference. The equal-person mean absolute rate error was 10.28 percentage points with the fixed threshold. It was 10.57 points with one shared median correction. It was 5.52 points with a separate median correction for each pair. The separate correction reduced error in 2 of three people. The records cover rest to sleep, sleep to a sleep-wake transition, and sleep to sleep. These data have no intended-command labels. The result measures numerical threshold stability. It does not show better task decoding or a safe stimulation rule. The package has code, features, raw examples, source versions, byte hashes, tables, and five figures.

SEEG · intracranial EEG · brain-computer interface · threshold calibration · high gamma · median correction · signal processing