Label-informed removal of at most 10 percent of STE marks reduced SOZ count AUC to 0.5 or less in eight of 16 first-hour SEEG records with initial AUC above 0.5

We tested exact sensitivity to mark removal in 18 first-hour public SEEG sleep records from OpenNeuro ds005398 v1.1.3. Labels, channels, and observation times stayed fixed. Sixteen records had initial STE count AUC above 0.5. A removal set that used labels reduced AUC to at most 0.5 within 10 percent of all marks in eight of them. Two other records already had AUC below 0.5. The median minimum was 8.97 percent of all marks, or 30.05 percent of SOZ marks, across the 16 records. A new exact integer solver and a separate integer program agreed on all 18 primary thresholds. The package includes a proof, source bytes and pins, code, count tables, exact cost curves, mark row witnesses, figures, and repeat instructions. These are possible changes under unrestricted removal. They do not estimate actual artifact rates or patient benefit. This study reuses an existing cohort and does not supply independent replication.

high-frequency oscillations · stereoelectroencephalography · seizure onset zone · area under the curve · sensitivity analysis · integer optimization