MNI rank subtraction fitted in nine SEEG patients changed mean SOZ AUC within shafts by -0.004 in eight test patients

We tested whether MNI event count ranks improve seizure onset zone (SOZ) ranking beyond STE ranks. The source was public OpenNeuro ds005398 version 1.1.3. We fitted weights from the first 30 minutes in nine SEEG patients. We tested minutes 60–70 in nine other patients. Eight had SOZ versus other-channel pairs on the same shaft. Mean AUC changed from 0.652821 to 0.649306. The mean change was -0.003516. The descriptive 95% patient bootstrap interval was -0.186979 to +0.126042, with training weights fixed. Five patients improved, two worsened, and one had no change. The fitted MNI weight was negative. In UCLA47, both detectors alone had AUC 1.00. The fitted score reversed seven of 12 pairs and gave AUC 5/12. Exact rank differences explain this failure. A planned exclusion of one training patient removed these reversals and gave mean gain +0.066970. The small test and reused cohort leave the general gain uncertain. The package preserves source bytes, new counts, code, all planned checks, scores, and retrieval instructions. Independent source and numerical checks passed. A clean repeat matched all 19 generated files byte for byte.

stereoelectroencephalography · high-frequency oscillations · seizure onset zone · event counts · patient-level validation · rank model