A separate implementation reproduced the ten second calibration accuracy change across two seizures in 32 SEEG patients

A separate implementation reproduced ten second background calibration of a gamma power threshold in public stereoelectroencephalography (SEEG) data. The test used two seizures from each of 32 patients in OpenNeuro ds004100 version 1.1.3. Mean balanced accuracy was 69.53 percent with a fixed threshold and 69.38 percent after calibration. The change was -0.16 percentage points. The 95 percent patient bootstrap interval was -3.28 to +2.81 percentage points. This interval assumes independent patients with a common sampling distribution. It does not prove equal performance. A concurrent publication reported the same primary test. All 64 input hashes matched. All 1,280 primary decisions were identical, and all 32 patient results agreed within numerical tolerance. This is a computational reproduction on the same cohort. Five additional fixed settings also had intervals that included zero. The package adds a calibration utility and its proof for a pure additive change. It preserves source code, unchanged compact power data, full results, source range locks, raw examples, and repeat instructions. Fresh public retrieval checked all 757,770 source power values. An offline repeat reproduced all ten numerical output files exactly. The short excerpts do not measure continuous seizure detection or clinical benefit.

SEEG · epilepsy · seizure detection · gamma power · threshold calibration · balanced accuracy · reproducibility