An early SEEG reader gave 49.4% site accuracy at 100–500 ms for remote neighboring sites in five offline sessions

We tested whether an early SEEG reader could identify the same neighboring stimulation sites at later response times. The reader used data at 20–100 ms after a pulse. We used OpenNeuro ds004457 version 1.0.2. The main set had 91 site choices in five patients. Each choice had two neighboring sites on a shaft with no retained recording contacts. Each model used the first four eligible pulses per site. The same last four pulses tested all six reading intervals. The early reader gave 65.5% mean early accuracy and 49.4% mean accuracy at 100–500 ms. Chance was 50%. Readers fitted at each later interval gave 52.3%. The planned gain was 2.9 percentage points, with a two-sided patient sign-flip p-value of 0.125. One reader fitted across all six intervals gave 50.2% late accuracy. The package contains data, models, all decisions, three figures, code, and a checked raw-data recipe. The result applies to known site choices in offline tests. Stimulation blocks and five reused patients limit the result. It does not show that late neural signals lack site information.

stereoelectroencephalography · SEEG · brain–computer interface · site identification · response-time transfer · intracranial EEG