Clinical SOZ pairs show a small power rise 1–5 seconds before the SEEG onset mark

We tested if seizure activity in the onset region starts before the clinical onset mark in SEEG. We used archived 1-s features of OpenNeuro ds004100 (143 seizures, 37 patients, one center). The features are 20-100 Hz log power and line length of bipolar pairs. We z-scored each pair against [-40,-25) s. Then we compared clinical SOZ pairs with other pairs in each window. In the pre-specified primary window [-5,-1) s, SOZ pairs had higher 20-100 Hz power (median +0.18 z, 22 of 36 patients positive, p = 0.007). In the sham window [-25,-15) s, the difference was +0.01 (p = 0.74). Line length gave the same result (+0.28 z, p = 0.003). A sustained departure (target mean z >= 2 for 3 s) started in [-10,-1) s in 17% of seizures per patient. In a sham interval of the same length, the rate was 5% (p = 0.0015). A simulation showed that the FFT filter moves post-onset power into the last second before the mark. Thus, we did not use [-1,0) s for claims. The leakage bound for [-5,-1) s was 0.004 log units, and the observed rise was 0.093 log units (p = 0.003 after we subtracted the bound). Data-selected top-10% onset pairs had a selection bias in the sham window (+0.36 z). Their pre-minus-sham rise was +0.40 z (p < 0.001). For closed-loop BCIs, this result shows that SOZ contacts sometimes give a few seconds of lead time before the clinical mark. But the early signal is weak in most seizures. Code, provenance, tables and figures are in the package.

SEEG · seizure onset · clinical onset mark · seizure-onset zone · intracranial EEG · 20–100 Hz power · responsive neurostimulation