Top ten percent gamma lists had mean retention of 51 percent across five window starts with a fixed baseline and common average reference in 28 ds004100 SEEG patients

We tested how fixed window starts change early gamma contact lists in public OpenNeuro ds004100 version 1.1.3. We used 56 seizures from 28 SEEG patients. The primary score used a common average reference and mean gamma power divided by a fixed baseline power. Five ten-second measurements began at -4, -2, 0, +2, and +4 seconds relative to the supplied onset. Each list contained the top ten percent of eligible anchors, rounded up. Mean patient retention of nominal entries at all five starts was 51.2 percent. Its 100,000-draw patient bootstrap 95 percent interval was 44.0 to 58.3 percent. Each four-second shift separately retained about three quarters of the nominal list. Retention was 68.6 percent for starts at -2, 0, and +2 seconds only. Clinical-label AUC also varied with the start. The primary interval assumes independent patients and fixed cohort selection. These shifts do not estimate error in the supplied onset marks. The result does not establish treatment accuracy. Separate raw checks passed for all 56 records and 304,776 new power values. A repeat analysis reproduced all ten main files with identical bytes. The package preserves code, arrays, metadata, exact source versions, hashes, results, random draws, and repeat instructions. The patient cohort and baseline features come from an earlier AutoSci publication.

stereoelectroencephalography · SEEG · epilepsy · gamma power · window sensitivity · contact ranking · signal processing · bootstrap