Interictal SOZ agreement was not higher in seizure-free patients in two SEEG cohorts

We tested whether agreement between interictal SEEG features and the clinical seizure-onset zone (SOZ) was higher in patients who became seizure-free after surgery than in patients whose surgery failed. The pre-specified primary analysis used 20 surgery patients from OpenNeuro ds003876 at NIH, Johns Hopkins, and Miami, excluding RNS patients. The score was each patient's SOZ AUC from a frozen HUP-trained model. Median AUC was 0.63 in seizure-free patients (n=10) and 0.75 in patients who were not seizure-free (n=10); the one-sided permutation Mann–Whitney test for seizure-free greater than failed gave P(S>F)=0.34 and p=0.89. The median difference was −0.12 (95% bootstrap CI −0.30 to 0.07). Across 21 primary and descriptive tests, none showed higher scores in seizure-free patients at p<0.05. Theta power in ds003876 gave medians 0.65 versus 0.68 (p=0.80). In 35 HUP SEEG patients from ds004100, theta gave 0.77 versus 0.71 (p=0.39) and spike rate 0.80 versus 0.82 (p=0.21). In 55 pooled patients, theta gave 0.70 versus 0.71 (p=0.64). These results do not support using agreement with clinical SOZ labels as evidence that an interictal map is surgically correct. The primary groups were small (10 per outcome; only large effects were detectable). Analyses used the supplied clinical SOZ and outcome labels, short interictal clips without sleep-state or artifact control, and included ablation patients as surgical cases in ds004100. Follow-up time was not controlled.

SEEG · epilepsy · seizure-onset zone · interictal features · surgical outcome · AUC · OpenNeuro