In 20 SEEG patients at one center, first-seizure channels were not detectably different from SOZ channels and were earlier than random pairs

A responsive neurostimulation device can monitor only a few bipolar channels. We compared a four-pair montage selected from the onset map of the first recorded seizure (S1) with clinical seizure-onset-zone (SOZ) pairs, random pairs, a montage based on all prior seizures (PRIOR), and the full implant (ALL). We analyzed archived 1-s high-frequency (20–100 Hz) and line-length features from OpenNeuro ds004100 v1.1.3 in 20 SEEG patients with at least two seizures (57 later seizures); run order was assumed to be seizure order. For the primary high-frequency detector (k = 4, z > 3 for 3 consecutive seconds), median patient latency was 1.0 s for S1, −0.4 s for SOZ, and 5.95 s for RAND. S1 was not detectably different from SOZ (n = 19, median difference −0.37 s, p = 0.78; this does not establish equivalence) and was earlier than RAND (n = 20, median difference −3.81 s, p = 0.045). SOZ-pair detectors were earlier than random-pair detectors (p = 0.043). The S1 montage contained a SOZ pair in 58% of test seizures. ALL appeared much earlier at this threshold but had a sham detection fraction of 0.80. In post hoc comparisons at matched sham rates, four-pair montages had similar operating curves, while the full implant with a higher threshold was equal to or better than them. This single-center analysis used clinical onset marks, a 15-s baseline, a 13-s sham window, 1-s resolution, and data ending at +20 s; its findings do not establish equivalence or clinical effectiveness.

seizure detection · SEEG · responsive neurostimulation · electrode montage · detection latency · high-frequency power · seizure-onset zone