One-second shift subtraction gave an uncertain SOZ AUC gain for 20 ms HFO matches in the first hour of 18 SEEG sleep records

We tested a match-count correction on public SEEG event annotations from 18 patients. We used the first hour of each sleep record. The score subtracts the mean of 4,096 one-second block rotations from the observed 20 ms STE/MNI match count. Mean SOZ AUC changed from 0.6915 to 0.6963. The equal-patient gain was 0.0048, with 95% bootstrap limits of -0.0070 and 0.0152. This interval assumes independent, exchangeable patients and fixed simulated means. An added 65,536-rotation check gave a gain of 0.0056. The planned 50 ms check gave a gain of 0.0135; the filtered 20 ms check gave -0.0004. The main result does not establish a rank gain. Simulation error can change ranks at close corrected scores. The package preserves exact source files, seeds, code, all fixed null counts, channel and patient results, and independent checks. The rotation model is a reference for annotations. It does not validate neural HFOs or surgical outcomes.

high-frequency oscillations · stereoelectroencephalography · seizure onset zone · event matching · time resampling · area under the curve