Unrestricted positive SOZ label additions can reverse the mean unfiltered STE count AUC lead over MNI in 18 first-hour SEEG records

This study gives exact limits for a detector comparison when positive seizure onset zone (SOZ) labels can be missing. It uses released first-hour HFO marks from 18 SEEG sleep records in OpenNeuro ds005398 version 1.1.3. There are 1,285 good bipolar channels. All source positive labels remain positive. Counts stay fixed. Both detectors use the same changed labels. The source mean STE minus MNI count AUC is +0.2011. With at most floor(5% of source negatives) additions per patient, its exact range is -0.0925 to +0.3264. Under a shared total budget, 21 additions first permit a nonpositive mean. When additions are limited to shafts with source positive labels, the mean stays positive over every permitted set. The method uses fixed score ranks, exact fractions, and dynamic programming. Full enumeration on small examples and independent score comparisons check the results. All nine output files match in an offline repeat. The package preserves code, source bytes, source pins, an exact proof, attaining label sets, and retrieval instructions. These finite-cohort limits test label assumptions. They do not measure actual missing labels or predict surgical outcome.

stereoelectroencephalography · high-frequency oscillations · seizure onset zone · area under the curve · sensitivity analysis · reference labels · exact bounds · OpenNeuro