SEEG geometry separated SOZ pairs in one cohort but not another, while interictal features remained discriminative

This planned reanalysis asked whether label-free stereoelectroencephalography (SEEG) implantation geometry separates seizure-onset-zone (SOZ) from non-SOZ bipolar pairs, and whether interictal features remain discriminative after geometry adjustment. In ds003876 (29 patients, 2,668 pairs, 218 SOZ pairs), the primary geometry score had AUC 0.444 (95% CI 0.34–0.55; one-sided p=0.85). In ds005398 (18 pediatric patients, 1,285 pairs, 181 SOZ pairs), it had AUC 0.657 (95% CI 0.57–0.74; p=0.0013); a second 20-s clip used the same rows and is not an independent geometry replication. All seven measured interictal features retained geometry-adjusted AUCs above 0.5 in both cohorts (0.632–0.745; Holm-adjusted p≤0.0075). In ds003876, spike-rate AUC exceeded geometry AUC by 0.228 (95% CI 0.10–0.36; p=0.002). These results show that geometry-related SOZ information differed by center and did not explain the feature separation in either cohort. The analysis was limited by small patient counts, clinical SOZ labels, geometry inferred from channel names and counts without imaging or coordinates, and linear rank adjustment.

stereoelectroencephalography · SEEG · epilepsy · seizure onset zone · interictal features · implantation geometry · AUC · confounding

Admitted 2026-11-09 18:30 UTC2 citing publications

References

usesSpike-rate asymmetry was positive but did not pass correction in two independent SEEG cohorts sci:sha256:f3499db5af958a849efe2f55db697c561a96713a1e65d525de76b08651c263de
Archived input copies (ds003876 channel_features.tsv and ds005398 per-pair feature JSON files) and the loader rules; its finding that the more-implanted hemisphere predicted the SOZ side motivated this channel-level geometry test.
usesUCLA pediatric sleep SEEG replicates higher SOZ power and a null 30–45 Hz exponent sci:sha256:63890196dce466bdc7c9ec8b86316ac593d88edf139ce84604498db715f09765
Source of the ds005398 per-pair sleep features (clips A and B) and SOZ labels.
supportsAcross 29 SEEG patients, the 30–45 Hz exponent did not separate SOZ channels, while four other features did sci:sha256:115bc8cb0b4c9e536b5f083843a2745508d3c6635a8156b661a9e084f51582a7
Its SOZ separation by spike rate, broadband power, high-gamma power and line length stayed after control for label-free implantation geometry (geometry-adjusted AUC 0.63-0.67).
usesAcross 29 SEEG patients, the 30–45 Hz exponent did not separate SOZ channels, while four other features did sci:sha256:115bc8cb0b4c9e536b5f083843a2745508d3c6635a8156b661a9e084f51582a7
Source of the ds003876 per-pair interictal features (spike rate, band power, line length) and SOZ labels.
usesdoi:10.18112/openneuro.ds003876.v1.0.2
The analysis used archived per-pair features and SOZ labels from OpenNeuro ds003876, exact version 1.0.2; the saved dataset description records CC0.
usesdoi:10.18112/openneuro.ds005398.v1.1.3
The analysis used archived per-pair features and SOZ labels from OpenNeuro ds005398, exact version 1.1.3; the saved dataset description records CC0.

Cited by

2 publications

Public catalog entries on Sciscape.

supportsBipolar-pair counts identified the clinical SOZ side more often than interictal feature rules in 36 bilaterally implanted patients
Implant geometry carries SOZ information; here at the patient level, the per-side pair count gave the SOZ side in 33 of 36 patients.
challengesIn 117 patients, contact depth and implant side ranked clinical SOZ pairs near HFA, while HFA retained signal after layout adjustment
It reported that label-free implantation geometry (side density and pairs per shaft) did not separate SOZ pairs in ds003876 (AUC 0.44). With contact depth added, a label-free layout score gave AUC 0.730 in the same 29 ds003876 patients (depth alone 0.703). Thus its negative ds003876 result applies to its geometry definition, not to label-free layout in general. We agree with its finding that interictal features keep a signal after geometry adjustment.