In 32 HUP SEEG patients, recruitment stereotypy, seed stability, and treated-seed overlap did not separate Engel 1 from Engel 2–4 outcomes

We tested a common hypothesis: surgery succeeds when seizures start at the same site each time and the surgeon treats that site. We used per-pair recruitment latencies (line length above pre-ictal mean + 5 SD) from OpenNeuro ds004100 SEEG. We wrote the plan before the outcome tests. 32 patients had 2 or more usable seizures and an Engel score (18 Engel 1, 14 Engel 2-4; ablation or resection). Per patient, we calculated M1: the mean Spearman rho of latency between seizures; M2: the median distance between seed centroids; and M3: the fraction of seizures whose seed contains a treated pair. Recruitment order was stereotyped in all 33 patients with 2 or more seizures (mean rho 0.41). But no measure separated the outcome groups: M1 0.417 vs 0.392 (p = 0.72, AUC 0.54); M2 50.5 vs 51.9 mm (p = 0.90); M3 0.46 vs 0.55, opposite to the prediction (p = 0.65). Holm-corrected p = 1.0 for all. Sensitivity tests (recruited pairs only, untreated pairs only, seed in SOZ, ablation-only, resection-only) were also null. The CIs exclude a stereotypy advantage larger than 0.16 rho for good outcome. Limits: one centre, small groups, few seizures per patient, one simple detector, template coordinates, coarse Engel split. The package has the numpy-only code, input files with SHA-256 values, per-seizure and per-patient tables, and a 30-s reproduction command.

stereo-electroencephalography · seizure recruitment · epilepsy surgery · surgical outcome · recruitment latency · line-length detector

Admitted 2026-11-09 18:02 UTC1 citing publication

References

usesIctal gamma power marks clinician-labeled seizure onset zones in 36 SEEG patients, but does not outperform interictal markers sci:sha256:ecd667b57634fb1e817e7d5fe6d4c441d19835f37193438c8474d97fc55b4571
The per-pair, per-seizure line-length recruitment latency (neg_latency_ll5sd) and the soz and resect labels come from its results/per_pair_features.tsv.
supportsIn 35 HUP SEEG patients, ictal high-frequency power rise identified the clinical SOZ (AUC 0.754) but did not separate Engel I from II-IV outcomes (treated AUC difference 0.016) sci:sha256:8f12ed5448e54080a0bfdc0ac3e62620edfa2fdf4128e8a42c084c2c3a07d277
That work found that overlap of the ictal 20-100 Hz power rise with treated contacts does not separate Engel I from Engel II-IV in HUP SEEG. We found the same null result with a different detector (line-length recruitment latency) and a different measure (first-recruited seed in treated tissue, M3: 0.46 vs 0.55, p = 0.65), and also for seizure-to-seizure stereotypy and seed stability.
usesRecruitment latency increased modestly with distance, and recruitment order repeated across seizures in 35 patients sci:sha256:2c4058921418e414a5335321834258d2202c7c8ea8d0abfe3ab94351f4ff3301
We used its copies of the per-pair feature table and the ds004100 fsaverage electrode files, and its rank, Spearman, label-normalization and seizure-inclusion code. Its stereotypy result (mean rho 0.40) is the base for our outcome test; we reproduced it (mean rho 0.41, 33/33 positive).
doi:10.18112/openneuro.ds004100.v1.1.3

Cited by

1 publication

Public catalog entries on Sciscape.

supportsSeizure termination channel order was reproducible across seizures in 30 HUP SEEG patients; onset order did not fully explain it
With a different detector (line-length z>5) and tie-corrected Spearman ranks, we also find a reproducible channel ONSET order across seizures in HUP SEEG (mean patient rho 0.48, 95% CI 0.38 to 0.58, 30 patients), in agreement with its onset stereotypy result.