In 35 SEEG patients, seizure-1 maps improved two-lead onset capture over clinical SOZ labels, but not at the full-shaft scale

We simulated responsive-neurostimulator-like devices with two leads, each spanning four contiguous contacts, on stereoelectroencephalography (SEEG) implants from 35 patients at one center. We selected leads using the 20–100 Hz onset map from the first recorded seizure and tested them on 90 later seizures. At the primary two-lead scale, the best pair on the selected leads had a mean percentile rank of 90.7% in later onset maps, compared with 82.7% for leads selected from clinical seizure-onset-zone (SOZ) contacts (difference +0.083, 95% CI 0.027 to 0.145; sign-flip p = 0.009; better in 23 of 34 patients). The seizure-1 rule also exceeded random leads by 0.105 (p < 1e-4); SOZ leads did not exceed random leads on percentile rank (+0.026, p = 0.37), although they did on the top-10% hit outcome. A pre-onset negative-control map was not better than random (+0.011, p = 0.70). The seizure-1 advantage over SOZ labels decreased with read-out scale: +0.121 for one pair, +0.107 for one lead, +0.083 for two leads, and +0.020 for two full shafts (p = 0.26). Line length gave a similar direction at two leads (+0.061, p = 0.034). At two leads, 31% of later seizures had none of their top-10% onset pairs on the leads selected from seizure 1. This is a ranking proxy on a single-center SEEG implant, not a detection or stimulation outcome; simulated leads may not reflect anatomic limits of real RNS placement. Run order was treated as seizure order, onset times were clinical marks, and clinical SOZ labels may include information from test seizures.

stereoelectroencephalography · epilepsy · seizure onset · responsive neurostimulation · seizure-onset zone · target selection · high-frequency power · line length