Local 2 s beta baselines gave no clear gain over transferred baselines for post-stimulation correlation forecasts in eight SEEG patients

We tested whether a local 2 s beta baseline gave a better post-stimulation correlation forecast than a baseline from another record. We used all 22 records from the eight patients with repeated records in public OpenNeuro ds006519 version 1.0.0. Each of 48 ordered record pairs used matched contacts and edges between different shafts. The beta band was 13 to 30 Hz. Patients had equal weight. Mean squared error was 0.03546 for the local baseline and 0.03534 for the transferred baseline. The transferred-minus-local difference was -0.000114. Its descriptive 95 percent patient bootstrap interval was -0.000771 to 0.000441. The exact patient sign-flip p value was 0.765625. The test did not establish a local gain or equal performance. A fixed zero-correlation forecast had 43.4 percent less error than the local baseline. Both baseline errors were higher than zero error in each patient. The target was a noisy sample estimate. The sign-flip test assumes independent patients and symmetric null differences. Contact masks assume that the supplied bipolar pair labels describe the montage. Record order was not verified. No live interface or stimulation policy was tested. The package preserves input arrays, all forecast cases, source hashes, code, three figures, and raw retrieval instructions. Independent checks passed, and raw extraction reproduced all 22 input arrays exactly.

stereoelectroencephalography · SEEG · beta band · brain-computer interfaces · functional connectivity · correlation networks · forecast error · baseline transfer