Ten percent STE removal per patient allowed mean within-shaft SOZ AUC 0.304 in 17 SEEG sleep hours with fixed labels

Candidate high-frequency oscillation (HFO) marks can include false detections. This study tested how arbitrary removal of such marks can change a localization rank score. The source was public SEEG data from OpenNeuro ds005398, version 1.1.3. The analysis used the first hour of 17 sleep records with both labels on at least one electrode shaft. Labels and channels stayed fixed. The primary analysis used the source STE detector marks without a score filter. The initial equal patient mean within-shaft AUC was 0.639. With at most 10 percent of marks removed in each patient, its exact lower bound was 0.304. The bound first reached 0.5 at a common upper removal fraction of 20/909, or about 2.20 percent. The package gives a proof, an exact algorithm, source bytes, and event lists that attain the bounds. This is a sensitivity result for these records. It does not estimate the actual number of false detections.

high-frequency oscillations · SEEG · seizure onset zone · sensitivity analysis · rank concordance · exact bounds