SEEG readers gave 0.515 mean AUC for transfer between images and sounds on later trials in 11 patients

We tested whether a fixed SEEG reader could separate before-onset and after-onset windows across images and sounds. We used OpenNeuro ds005415 version 1.0.0. The test used 353 later trials from 11 epilepsy patients. Each reader used eight adjacent contact pairs and raw log power in three frequency bands. The transferred readers gave mean AUC 0.515. The descriptive 95% patient bootstrap interval was 0.471 to 0.561. Local readers gave mean AUC 0.555. The paired transfer minus local difference was -0.040, with an interval of -0.085 to 0.005. The before-only transfer control gave AUC 0.499. The training label-swap reference gave p = 0.265. Transferred probability scores had mean Brier error 0.315; a constant score of 0.5 has error 0.250. A diagnostic added after the main analysis gave training AUC 0.955 and later local AUC 0.555. The package contains new power arrays, source checks, figures, predictions, models, code, and retrieval instructions. All 1,044 retained voltage ranges were fetched again and checked. A separate replay matched all analysis files. The test requires a known event onset. It does not establish a reliable BCI feedback detector or test brain stimulation.

SEEG · brain-computer interface · cross-condition transfer · AUC · clinical recordings