A reader with early post-stimulation SEEG power had lower AUC than a descriptor reader in a fixed pair test of 17 people

This study tested moving and stationary visual effect labels in 86 public SEEG records from 17 people. Each positive and negative test record used one fitted model. The fit excluded all records from both test people. The test used fixed features, equal-person record weights, and a logistic penalty of 0.1. The reader with early post-stimulation power had AUC 0.544. The six-descriptor reader had AUC 0.659. The early reader remained lower at three fixed penalties and after each full re-analysis that removed one person. A reader that used only training label frequency had pair AUC 0.500, versus pooled leave-one-person-out AUC 0.086. A proof and score-transform checks explain this difference. The method applies established leave-pair-out AUC to whole people. The cohort was reused from an earlier study. The results are descriptive and do not establish a population effect or live BCI function. The package contains source spectra, pinned raw retrieval code, features, all primary fits, pair scores, three figures, and exact replay checks. Four fresh raw record checks and the numerical replay passed.

SEEG · leave-pair-out AUC · grouped cross-validation · visual effects · brain-computer interface · intracranial EEG