Median reference power at 70–110 Hz gave 53.1% mean balanced accuracy on later trials from periodic 1–30 Hz SEEG inputs in eight people

We tested a task reader with 70–110 Hz power made by a median reference. We first projected each finite SEEG window to 1–30 Hz. All other DFT bins were zero. The source was OpenNeuro ds004473 version 1.0.1, with eight people and 1864 retained trials. One fixed logistic reader per person fit on early trials and tested on 564 later trial pairs. Mean balanced accuracy was 53.1%. The descriptive 95% person bootstrap interval was 49.8–56.3%. The interval included chance. Mean Brier error was 0.24896, compared with 0.25 for constant probability 0.5. The created band was present, but this reader gave no clear group evidence of task prediction. The test used offline, periodic windows and reused clinical recordings. It did not test intent, online control, or stimulation. The package preserves 3728 reference traces, exact source examples, features, readers, predictions, figures, and source retrieval code.

stereoelectroencephalography · median reference · spectral null · signal processing · brain-computer interface · balanced accuracy