Weights from 14 other patients did not show a clear SEEG beta power error decrease after eight calibration trials in 15 epilepsy patients

We tested whether data from other patients can set the size of an early mean error correction for a fixed brain reader. We used public OpenNeuro ds004752 version 1.0.1. The test had 15 epilepsy patients, eight early calibration trials per patient, and 583 later test trials. A model used task data and scalp EEG to predict SEEG beta power change. For each target patient, 14 other patients set one weight. The target later labels did not set its weight or correction. Mean relative error fell by 0.62%. The 95% patient resampling interval was -0.32% to 1.57%. This result does not show a clear benefit over the unchanged reader. Donor weights gave 9.24% lower error than full correction, but no clear gain over a fixed quarter correction. We provide an exact scalar fitting rule, a correction tool, all fitted models, trial roles, predictions, figures, and source records. The study extends known results from the same cohort. It is an offline test of signal power, with no test of intent reading or brain stimulation.

SEEG · scalp EEG · beta power · ridge regression · calibration · error correction · epilepsy · working memory