Interictal SEEG features rank clinical seizure onset zone channels above chance, and this agreement is not lower in failed surgery (33 patients, ds004100, 90 s clips)

We tested whether interictal SEEG features identify the clinically labelled seizure onset zone (SOZ), and whether agreement is lower when surgery fails. We used 33 SEEG patients from OpenNeuro ds004100 snapshot 1.1.3 (19 Engel I, 14 Engel II-IV), with 90 s from one interictal clip per patient and a bipolar montage (3,319 channels, 347 SOZ). SOZ channels had higher total power, delta power, line length, and spike-like transient rate; relative alpha, beta, and high-gamma power were lower, and the 30-100 Hz spectral slope was steeper (Holm-corrected permutation p < 0.01 for each). The steeper slope did not remain after removing the effect of total power. A leave-one-patient-out logistic model gave a mean within-patient AUC of 0.786 (32 of 33 patients above 0.5; label-permutation null mean 0.500). AUC was 0.778 in Engel I and 0.797 in Engel II-IV patients (p = 0.72; difference -0.019, bootstrap 95% CI [-0.114, 0.084]). Agreement between interictal features and the SOZ label did not identify failed surgery in this cohort. Limits include short clips, a simple spike detector, no HFO analysis, no anatomical control, and low power for small outcome effects.

SEEG · epilepsy · seizure onset zone · interictal · spectral features · surgical outcome · OpenNeuro ds004100