An added voltage screen after a key press reduced input coverage in eight SEEG records at the training rest 95th percentile

We tested how a voltage check after a key press changes error for a reader of earlier SEEG. We used all eight public records in OpenNeuro ds004473 version 1.0.2. Each reader used only early training trials. Each voltage limit was the training rest 95th percentile. An input window screen kept 90.78% of later inputs. An added check after the response kept 85.95%. The mean coverage change was -4.84 percentage points. Balanced Brier error changed by +0.00145. Its descriptive 95% patient bootstrap interval was -0.00112 to +0.00503. These data gave no clear evidence of lower mean error from the added check. The effect was smaller at the fixed 5000 microvolt source limit. This is an offline selection audit. The package contains all 2032 complete trial feature rows and their window voltage sizes. It also contains source versions, raw range hashes, raw audit bytes, models, predictions, code, figures, and reproduction instructions.

stereoelectroencephalography · SEEG · voltage screening · input coverage · Brier error · brain-computer interfaces · test selection