In HUP SEEG, a seizure-1 map predicts later onset pairs; a two-lead placement captures 31% of later seizures

In archived HUP stereoelectroencephalography (SEEG) data from OpenNeuro ds004100, we tested whether one recorded seizure predicts later seizure onset pairs and whether a two-lead placement chosen from that seizure contains later onsets. In a replication using unchanged analysis code, the seizure-1 high-frequency (20–100 Hz power) onset map had median AUC 0.737 for later top-10% onset pairs in 13 held-out patients (37 seizures); 12/13 patient values exceeded 0.5 (p=0.0005), and the map outperformed clinical seizure-onset-zone (SOZ) labels (median AUC 0.533; 12/13 positive differences; p=0.0005). Line length gave the same direction (median AUC 0.793 versus SOZ 0.543; map–SOZ p=0.0034). In a separate placement analysis of 35 patients and 90 later seizures, a seizure-1 high-frequency map selected two four-contact leads whose six bipolar pairs contained the later top onset pair in 0.314 of seizures, compared with 0.062 for random placement (p=0.0026) and 0.169 for SOZ placement (p=0.030). Nineteen of 35 patients had at least one hit; 16 had none. For line length, the placement hit rate was 0.302 versus 0.217 for SOZ (p=0.206). A mean map of prior seizures was not better than seizure 1 alone. The study is from one center, uses run order as seizure order, and evaluates a feature-based onset pair rather than a clinical or surgical outcome. The lead geometry and mapping from SEEG to responsive neurostimulation are not modelled.

stereoelectroencephalography · seizure onset · epilepsy · high-frequency power · line length · seizure-onset zone · responsive neurostimulation · held-out replication