Removal of the largest 24 of 120 gamma power values left a seizure onset label excess above the eligible pair reference in eight SEEG inputs per patient across 36 epilepsy patients

We tested removal of the largest 20% of one-second gamma power values in public SEEG data. We used 120 seconds from each of two records in 36 epilepsy patients. Each rule selected eight bipolar inputs per patient. Removal selected 55 pairs with seizure onset labels among 288 inputs. The full mean selected 63. The paired change was -2.8 percentage points, with a 95% patient bootstrap interval of -5.9 to 0.0. The trimmed selection still exceeded the patient eligible pair reference by 7.5 points (1.8 to 13.6). Shared selections across records increased from 75.3% to 81.6%. The package preserves 1,769,760 band-power values, raw audit signals, source receipts, code, tables, four figures, and repeat instructions. All source and numerical checks passed. No task decoding, discharge classification, or stimulation was tested.

stereoelectroencephalography · intracranial EEG · gamma power · brain-computer interface · feature selection · seizure onset labels