The interval for added HUP spectral exponent loss gain over contact order included zero in 19 external first-minute SEEG records. Patients were assumed independent and exchangeable.

We trained two fixed models on first-minute SEEG data from 31 HUP patients. We tested them on 19 eligible patients from NIH and Johns Hopkins. The baseline used contact order. The added model used the negative log power slope from 30 to 80 Hz, with 55 to 65 Hz removed. Both models ranked marked and unmarked adjacent bipolar pairs on the same electrode shaft. Mean patient loss gain was -0.019709 nats. Its joint patient bootstrap 95% interval was [-0.094060, 0.023970]. Mean within-shaft AUC change was -0.030190. Its joint 95% interval was [-0.131172, 0.000000]. The joint interval includes zero. The data do not give a clear mean loss gain under the stated bootstrap assumptions. The interval does not prove that the gain is zero. The package contains code, source versions and hashes, feature tables, spectra, predictions, all bootstrap draws, short signal fixtures, checks and reproduction instructions. The target is a clinical source marking. It is not a surgical outcome. The two test sites and the selected records limit the scope.

stereoelectroencephalography · epilepsy · seizure onset zone · spectral exponent · contact order · bipolar pairs · external validation · bootstrap