Causal HFA and line length ranked clinical SOZ pairs above other pairs within 3 seconds of marked onset in 35 SEEG patients

We tested how fast simple causal markers find the clinical seizure-onset zone (SOZ) after a seizure starts. This is the target choice that a closed-loop stimulation BCI must make. Data: OpenNeuro ds004100 v1.1.3, SEEG ictal runs 01-02, 35 patients with SOZ labels, 64 seizures. We read 80 s per seizure (onset-60 s to onset+20 s) from the EDF files with HTTP range reads. Method: bipolar pairs of adjacent good contacts on the same shaft; 1-s windows; log 65-115 Hz power (HFA), log line length (LL), log 13-115/3-12 Hz ratio (ER) and log 3-12 Hz power (THETA); robust z-score against [-60,-15] s; score = mean z from onset to T. We computed the AUC for SOZ vs non-SOZ pairs per seizure and then the mean per patient. Tests are over patients (sign-flip, bootstrap CI). Results: HFA AUC was 0.68 [0.62-0.74] at T=1 s, 0.72 at 3-5 s and 0.74 [0.67-0.81] at 20 s. LL gave 0.68 at 1 s and 0.73-0.75 at 3-20 s. ER was slower (0.62 at 1 s, 0.73-0.74 at 10-15 s). THETA was near chance (0.51-0.58). The same score in a sham window that starts 12 s before onset gave AUC 0.55-0.60. The paired real-minus-sham gain was +0.12 at 1 s and +0.17 at 3-5 s (p<=0.002). For resected pairs in 20 good-outcome patients, HFA AUC was 0.66 at 1 s and 0.73 at 5 s. In 29 patients with two seizures, the per-seizure AUCs correlated (r=0.68). Assumptions and limits: one center; clinicians marked the onset time and the SOZ from these seizures, thus the agreement is partly circular; the AUC is lower than the 0.79-0.82 AUC that earlier sci results report for multi-feature interictal models. Code, tables, figures and file provenance are in the package.

seizure localization · seizure-onset zone · SEEG · high-frequency activity · line length · causal markers · closed-loop stimulation · AUC