In one ds007445 feature set, SOZ-to-thalamus lag depended most on the pair summary rule across 144 detector settings

Library results for OpenNeuro ds007445 report seizure onset zone (SOZ)-to-ipsilateral-thalamus recruitment lags from 11 s to 30 s. We varied five detector choices in a pre-registered multiverse of 144 settings, using archived 0.25 s block line length (1–80 Hz) for 54 seizures in 19 patients; the patient was the unit. The mean SOZ-to-thalamus lag ranged from -0.6 to 32.1 s. Pair summary rule had the largest main-effect range (11.5 s): the mean lag was 16.1 s for the median over all pairs, including censored non-recruited pairs; 8.1 s for the first recruited pair; and 4.6 s for the median over recruited pairs. Thus, the long lags mainly reflect fewer thalamic pairs reaching threshold, rather than a later start among recruited pairs. In a joint sign-flip test for thalamus minus ipsilateral white matter (17 patients), the median across settings was +2.1 s (p = 0.45; Holm-adjusted p = 0.45); no setting found thalamus significantly earlier. SOZ was significantly earlier than thalamus in 86 of 144 settings, not all as predicted. The six prior lag values were not recomputed with their own code. Limits include one archived feature set, line length only, data mostly sampled at 200 Hz, and selection bias when the median includes recruited pairs only.

seizure recruitment · stereo-EEG · thalamus · line length · specification curve · detector sensitivity