Signals from other shafts raised mean 120 to 500 ms SEEG forecast error with past calibration at ridge penalty 1 in eight patients

We tested whether early signals from other electrode shafts improve a late SEEG waveform forecast with past calibration. The public source was OpenNeuro ds004696 version 1.0.1. We used 317 conditions and 1,933 test pulses in eight patients. Each condition used five calibration pulses. Shared fitting used only other calibration groups that ended before the query input. At ridge penalty 1, added shaft inputs raised mean 120 to 500 ms forecast error by 4.27%. The 95% whole-patient bootstrap interval was 3.31% to 5.12%. Error rose in all eight patients. The unrestricted calculation matched the prior 1.82% error increase. At fixed penalty 10, the past-data increase was 0.29%, with an interval that included zero. All eight native source reconstructions matched the compact inputs exactly. Time checks, future-data changes, independent error calculations, and offline replay passed. The package preserves data, source hashes, code, forecasts, tables, three figures, and repeat instructions. This is an offline test of selected clinical data. It does not establish a deployed interface benefit or the absence of useful information in other shafts.

stereoelectroencephalography · brain-computer interface · forecasting · ridge regression · temporal validation · electrical stimulation