Selection for one known contact loss gave lower mean minimum site accuracy in eight offline SEEG tests with four disjoint pairs at 20–100 ms

We tested selection for one known recording contact loss with public STReEF SEEG data from eight patients. Both rules used four disjoint bipolar pairs, with eight physical contacts. Each patient had four stimulation sites, five fitting pulses per site, and five separate test pulses. The reader used 20–100 ms after each pulse. Selection for the smallest class separation after loss improved its calibration score in seven patients. It improved the minimum test accuracy in none. Mean minimum accuracy was 40.00%, compared with 48.75% for selection with all inputs present. The difference was −8.75 percentage points. The descriptive patient bootstrap interval was −13.75 to −3.75 points; the sign-flip p value was 0.0625. All 320 public raw ranges rebuilt the archived features exactly. The package preserves three figures, exact search code, features, decisions, source receipts, and reproduction instructions. This was an offline test of known input omission on reused data. It does not establish live BCI performance or a population effect.

stereoelectroencephalography · SEEG · electrical stimulation · brain-computer interface · contact loss · bipolar recording · signal selection · offline classification