Two selected SEEG pairs had higher mean response detection error than eight pairs after condition changes in eight people

We tested how the number of selected SEEG pairs affects an offline brain reader after a task condition change. We used OpenNeuro ds004473 version 1.0.2 and eight people with epilepsy. Each model used 64 early calibration trials. The later test set contained 979 trials. Two selected pairs had mean response detection Brier error 0.2133. Eight pairs had error 0.1996. The mean increase was 0.0137, with a descriptive 95 percent person bootstrap interval of 0.0025 to 0.0264. The exact sign flip probability was 0.0703. Mean balanced accuracy was 66.66 percent with two pairs and 69.69 percent with eight pairs. Left versus right key accuracy remained between 51.22 and 52.96 percent across six budgets. Mean key Brier error did not improve on the constant 0.5 output at any budget. The result separates response detection from key reading when testing recording input requirements. These test trials were used in earlier sci work. The response windows use future event times. The results are exploratory and do not establish performance for a live interface. The package preserves code, input arrays, source retrieval recipes, models, scores, figures, and independent checks.

stereoelectroencephalography · SEEG · brain-computer interfaces · pair selection · Brier error · condition transfer · response detection · key reading