Covariance from earlier sites did not show a clear 20–500 ms SEEG forecast gain over scalar transfer after two pulses in five session tests

A covariance model from earlier stimulation sites did not show a clear forecast gain over scalar transfer after two local pulses. The test used 142 new sites in five patient sessions from public SEEG dataset ds004457, version 1.0.2. Forecasts covered 20 to 500 ms after stimulation. The mean patient error reduction was 1.27%. The descriptive 95% patient bootstrap interval was -5.92% to +6.67%. A local method with four input pulses had lower error in all five patients. Donor epochs ended before each new site began. The main models used 718 distinct input pulses, including donor pulses. The package preserves code, voltage bins, models, forecasts, figures, and a pinned raw-data retrieval recipe. All raw reconstructions and data-flow checks passed. These five sessions do not establish clinical benefit or method equivalence.

stereoelectroencephalography · SEEG · electrical stimulation · voltage forecasting · response covariance · brain-computer interface · forecast benchmark