A 25 ms shaft rule gave 69.59% mean HFO list agreement versus 69.62% with equal-count random removal in 135 sleep SEEG records from seven patients

We tested a timing removal rule for high frequency oscillation (HFO) lists in public SEEG sleep data. The source was OpenNeuro ds003498 version 1.1.1. Seven patients supplied 135 records of 300 seconds each. The rule removed an overlap-HFO mark if another shaft had an onset within 25 ms. It removed 24.00 percent of pooled marks. Lists used ceil(0.20 times the pair count), with fractional membership at ties. Each patient had equal weight. The comparator chose uniform random subsets with the same retained count in each record. It used 50000 joint draws. Mean agreement between distinct records was 69.59 percent for the shaft rule and 69.62 percent for random removal. The difference was -0.036 percentage points. Four patient differences were positive and three were negative. The exploratory 95 percent patient interval was -3.769 to +3.899 points. The sign-flip p value was 1.0 under independent patients and null sign symmetry. This near-zero mean does not prove equal population performance. All six fixed checks are reported. The package provides source tables, hashes, code, a count control, every filter decision, results, and reproduction instructions. Independent calculations passed. All 18 numerical files matched an offline repeat. The study tests released marks, not clinical accuracy or artifact identity.

high frequency oscillations · stereoelectroencephalography · epilepsy · event detection · list agreement · repeatability · random removal · shaft timing