Interictal SOZ-likeness falls weakly with distance and does not reliably locate an unsampled SOZ

In 34 patients with stereo-electroencephalography (SEEG), 3,541 bipolar contact pairs, and 378 clinical seizure-onset-zone (SOZ) pairs, we examined whether interictal SOZ-likeness extends into nearby non-SOZ tissue and can indicate an SOZ missed by an implant. A leave-one-patient-out, class-balanced L2 logistic model using 18 features had a median SOZ AUC of 0.81. Mean within-patient score was +0.98 SD at SOZ pairs, +0.17 SD at non-SOZ pairs 0–10 mm away, and −0.28 SD at 30–50 mm. Among non-SOZ pairs, score decreased weakly with distance (mean Spearman rho −0.088; 23/34 patients negative; one-sided p = 0.015); this remained after normative-map correction (p = 0.025), but was not significant on shafts without SOZ contacts (p = 0.15). One-to-four-hertz power alone showed no gradient (p = 0.24). In a simulated SOZ miss, the highest-scoring remaining pair was at the 40th–44th distance percentile (chance 50th; p = 0.02–0.09). These results indicate that the measured interictal signal is mostly focal and can rank sampled contacts, but does not reliably show the direction of an unsampled SOZ. Limitations include fsaverage template coordinates, clinical labels, one center and one five-minute clip, simulated rather than real misses, partial controls for shaft and region effects, and no multiple-test correction for H2 and secondary H1 tests.

stereo-electroencephalography · SEEG · seizure-onset zone · interictal · epilepsy · electrode localization · OpenNeuro · spatial gradient