Early signals from other shafts raised mean 120 to 500 ms SEEG forecast error at ridge penalty 1 in eight patients

This study tested whether early SEEG signals from other shafts improve a late waveform forecast beyond signals from the target shaft. It used OpenNeuro ds004696 v1.0.1 from eight patients with epilepsy. The test included 317 stimulation conditions and 1,933 distinct later pulses. Each query used five pulses to set means and scales. Ridge weights used the first five pulses of other conditions in the same patient. Input bins covered 20 to 100 ms; target bins covered 120 to 500 ms. At ridge penalty 1, adding other shafts raised mean error by 1.82%. The 95% patient bootstrap interval for the rise was 0.82% to 2.53%. Seven of eight patients had higher error. The error rise before the pulse was 2.11%. The paired difference did not establish a change specific to the after-pulse period. At fixed penalty 10, the error difference was near zero. An unplanned zero-waveform check did not establish a gain for either main model. The result limits this fitted model; it does not show that other shafts lack useful information. This is an offline transfer test. Other-condition calibration data can occur later in source time than query test data. The package preserves code, compact inputs, native examples, forecasts, scores, three figures, source metadata, checked retrieval recipes, and reproduction records. A clean replay matched all 11 generated numerical files in the recorded environment.

SEEG · waveform forecasting · ridge regression · brain stimulation · neural signals