Fixed ridge models with scalp theta increased mean test correlation for local SEEG theta across two memory sessions in 14 patients

We tested whether scalp EEG can predict local SEEG theta power in another verbal memory session of the same patient. We used public OpenNeuro ds004752 version 1.0.1. The test included 14 epilepsy patients, 28 sessions, and 1275 trials without source artifact flags. The target was 4–8 Hz bipolar power at contact pairs with a hippocampus-labelled anchor. We trained fixed ridge models on one session and tested them on the other session in both directions. Adding scalp theta to load and trial-position predictors increased mean patient test correlation from 0.049 to 0.254. The mean gain was 0.205, with a patient bootstrap 95% interval of 0.084 to 0.344. Twelve of 14 patients had positive gains. The exact patient sign-flip test gave p = 0.00390625. The test assumes independent patients and sign symmetry under the null. The interval assumes independent, exchangeable patients and fixes the selected sessions and contacts. Planned reference and penalty checks retained positive mean gains. Checks added after the result retained positive gains for pairs with two hippocampal labels. Removing trials with digital-limit samples also retained a positive gain. The package preserves code, matched features, models, predictions, source hashes, raw audit data, and retrieval instructions. Independent signal checks and an exact repeat analysis passed. The result supports prediction in these session pairs. It does not identify a hippocampal generator, a separate theta rhythm, or clinical performance.

scalp EEG · SEEG · verbal working memory · theta power · ridge regression · cross-session prediction · epilepsy