Separate binary labels for mixed contact pairs gave a mean AUC bound width of 0.098 for a fixed score in 36 SEEG patients

This study tests how contact onset labels affect a fixed bipolar SEEG score. It uses two 120-second interictal records from each of 36 patients in OpenNeuro ds004100 version 1.1.3. Of 3678 retained contact pairs, 126 have different endpoint labels. The target model keeps the common label for equal-label pairs. It permits a separate binary label for each other pair. The positive target count can change. Under this model, the exact AUC bounds have a mean width of 0.098. A patient bootstrap gives a 95% interval of 0.076 to 0.122 for mean width. The bounds include 0.5 in five patients. Four fixed endpoint rules give a mean within-patient AUC range of 0.069 in the 35 patients with both target classes under all rules. Removing mixed pairs removes all pair coverage for 15 of 360 source-positive contacts. The package contains a rank-based bound tool, its proof, all pair powers, scores, bound witnesses, source metadata, fixed source versions, and repeat instructions. Full enumeration checks all 36 patient bounds. An independent reader and direct Fourier calculation check one raw prefix. A fresh calculation matches all eight statistical output files exactly. The bounds describe target-rule sensitivity under stated assumptions. They can permit different rules on overlapping pairs. They do not establish true bipolar onset labels or clinical accuracy.

stereoelectroencephalography · SEEG · bipolar signals · seizure onset zone · area under the curve · AUC bounds · label sensitivity · interictal EEG · benchmarking