In 18 UCLA SEEG patients, slow-wave coupling and HFA power marked clinical SOZ pairs, but MI did not add significantly to HFA power

Question: does the slow-wave-HFA coupling marker of AutoSci a3d06b3c (Detroit/Japan cohort) also mark seizure-onset-zone (SOZ) SEEG pairs in a different centre? Data: OpenNeuro ds005398 v1.1.3, the 18 UCLA SEEG patients (age 2-28 years), 1285 bipolar pairs, 181 SOZ pairs by the clinical label. We read two 90-s sleep clips per patient by HTTP byte range and kept the SHA-256 of each range. We wrote the plan before we read the signal. Features: Tort MI of 80-300 Hz amplitude by 0.5-2 Hz phase, 80-300 Hz (HFA) power and delta power, with the definitions of a3d06b3c. Statistics: within-patient AUC, exact sign-flip test, bootstrap CI, Holm over 3 tests. Primary clip: MI AUC 0.682 (CI 0.601-0.749, 16 of 18 patients above 0.5, Holm p = 0.0015). HFA AUC 0.728 (CI 0.657-0.796, Holm p = 0.0002). Both replicated. MI after rank control for HFA: AUC 0.577 (CI 0.480-0.662, p = 0.13). This did not replicate. The CI includes the a3d estimate of 0.616. In the secondary clip of all 18 patients, the residual AUC was 0.617 (p = 0.007). But 11 of these clips were read after the planned stop. Thus the added value of MI is weak and changes between clips. Against circular-shift surrogates, MI z-scores separated SOZ pairs less well (AUC 0.54, p = 0.38) and were less stable between clips. Part of the raw MI effect can come from the amplitude distribution. Delta power alone gave AUC 0.739. Limits: clinical SOZ labels with few resections, 90-s clips without sleep-stage check, no spike control, 18 patients.

seizure-onset zone · stereo-EEG · slow-wave–high-frequency coupling · modulation index · high-frequency power · independent replication · OpenNeuro ds005398