Interictal SEEG power separated SOZ pairs in 13 HUP patients, but cross-session reliability was underpowered

In 13 patients represented in OpenNeuro ds005624 (HUP SEEG, color change detection task) and ds004100 (clinical seizure-onset-zone labels), we analyzed 20-s segments starting at 300 s and bipolar pairs of adjacent contacts on the same shaft. Clinical labels identified whether each pair was in the SOZ. Mean patient-level AUC was 0.710 for log 1–45 Hz power and 0.718 for log 1–4 Hz power; 12 of 13 patients had AUC above 0.5 for each feature (Holm-adjusted exact sign-flip p=0.0037 and 0.0029). HUP140 was below chance (AUC 0.326 and 0.438). Only four patients had multiple task sessions, so cross-session rank reliability could not be tested with useful power. Mean Spearman rank correlations were high within sessions (0.977 and 0.965) and between sessions (0.967 and 0.961); the differences were not significant (p=0.25 and 0.625; minimum possible two-sided p=0.125). The analysis uses 13 patients from one center, short task-state segments, and labels from a separate recording. Task state was not controlled, and stable channel properties may contribute to the observed rank stability. The assumed identity of patients, implants, and contact names across the two datasets is not independently established.

SEEG · epilepsy · seizure onset zone · interictal · power · bipolar pairs · rank reliability · OpenNeuro