A prior voltage model did not reduce mean forecast error for 20–500 ms pulse size at later sites in eight SEEG sessions

A prior voltage model did not reduce mean forecast error for pulse size at later stimulation sites in eight SEEG sessions. We used public OpenNeuro ds004696, version 1.0.1. The target was log voltage size at 20–500 ms after a pulse. The input was log voltage size at -980 to -500 ms. First-five pulses at earlier sites fitted one slope per patient. First-five pulses at each later site fitted local means. Final-five pulses at later sites tested the forecasts. The constant comparison used the same local calibration. The test included 44 later sites, 469 readout pairs, and 220 unique test pulses. Mean patient error reduction was -0.73%. The descriptive 95% patient bootstrap interval was -2.78% to +1.25%. The two-sided patient sign-flip p value was 0.508. Five patients had higher error and three had lower error. The evidence does not show a benefit of this model and does not exclude benefit from other state features. The target is recorded voltage size, not isolated evoked neural amplitude. The package preserves exact products, the local analysis plan, all predictions, source hashes, code, three figures, and reproduction instructions. It provides a fixed chronological benchmark for later BCI forecast studies.

stereo-EEG · brain stimulation · voltage size · forecasting · brain-computer interfaces · OpenNeuro