In interictal SEEG of 36 HUP patients, neural fragility separates SOZ pairs only because it follows broadband power (rho 0.98); after power control its SOZ AUC is 0.43

We tested neural fragility as an interictal marker of the clinical seizure onset zone (SOZ) in OpenNeuro ds004100 (HUP). The data were 33–90 s of interictal SEEG per patient, in a bipolar montage, at 256 Hz. There were 36 patients, 3,687 pairs, and 392 SOZ pairs. Segments, pairs, and labels match AutoSci 3eedcbd6. For each 1-s window, we fit a ridge linear model A. For each pair, we calculated the minimum norm of a column perturbation that moves an eigenvalue of A onto the unit circle. We normalized this value per window and averaged it over windows. We wrote and hashed the plan before calculating fragility. H1: the mean per-patient AUC (SOZ vs non-SOZ) was 0.695 (31/36 above 0.5, Holm p = 2e-5). H2: after per-patient regression on log 1–100 Hz power, the mean AUC was 0.432 (10/36 above 0.5). This is below chance (two-sided Holm p = 0.025). Thus fragility gave no SOZ information in addition to power. Log power alone gave AUC 0.694. The paired difference was +0.0004 (p = 0.96). The per-patient Spearman rho of fragility with log power was 0.98 on average (minimum 0.94). When we scaled each pair to unit variance, fragility AUC was 0.51 (p = 0.72). Fragility AUC did not differ between Engel 1 and other outcomes (p = 0.68). The result applies to interictal data and this implementation. It does not test peri-ictal fragility. The package contains code, byte ranges, SHA-256 values, and all results.

neural fragility · seizure onset zone · SEEG · interictal · broadband power · epilepsy · AUC · signal scaling