Early HFO rank repeatability selected detectors with lower mean later SOZ AUC than STE in 18 SEEG sleep records

We tested whether early count-rank repeatability selects the HFO detector with better later seizure onset zone (SOZ) channel ranking in 18 public SEEG sleep records from OpenNeuro ds005398 version 1.1.3. Between minutes 0–15 and 15–30, the rule selected MNI in 14 patients. At minutes 60–70, mean patient AUC was 0.5341 for the selected detector and 0.7160 for fixed STE; the mean paired difference was -0.1819 (95% patient bootstrap interval -0.2619 to -0.1069). The exact two-sided sign-flip value was 0.000366 under patient independence and null sign symmetry. All fixed secondary checks had negative mean differences; early event-count matching gave -0.1249. This is a retrospective table benchmark using supplied marks and labels. Prior first-hour results informed the question and comparator, so the cohort is reused and the new time interval does not provide independent patient evidence. The bootstrap assumes independent, exchangeable patients. No new detector fitting, event adjudication, prospective performance, or clinical benefit was tested.

high-frequency oscillations · stereoelectroencephalography · seizure onset zone · detector selection · rank repeatability · AUC · retrospective benchmark