Interictal gamma gave an uncertain gain in early seizure rank prediction after baseline adjustment in 29 ds004100 SEEG patients

We tested whether interictal gamma power improved prediction of early seizure power ranks beyond a nearby baseline and contact order. We used public OpenNeuro ds004100 version 1.1.3. The test included 29 patients, 3,062 contacts, and 358 electrode shafts. Models were fitted on the first seizure of other patients and tested on the second seizure of the excluded patient. Adding power from two 60-second interictal records increased mean rank prediction skill by 2.17 percentage points relative to zero prediction. The patient bootstrap 95 percent interval was -0.17 to 4.25 points. It includes zero. Interictal ranks closely tracked baseline ranks, with mean association 0.867. All five fixed secondary tests had positive mean gains. The study does not establish a positive population benefit or prospective seizure warning. Fresh source checks passed for 128 signal excerpts. A separate offline run reproduced 41 files exactly. The package preserves source code, parent features, compact matched inputs, source versions and hashes, predictions, bootstrap results, and reproduction instructions. The claim applies to a fixed ridge model, name-based shafts, and this clinical cohort.

stereoelectroencephalography · epilepsy · gamma power · rank prediction · interictal activity · patient-excluded validation