Five short pre-pulse windows did not show a clear forecast gain from local proxy coupling over geometry in eight SEEG sessions

We tested whether short voltage recordings near a stimulation site improve forecasts of remote SEEG responses. OpenNeuro ds004696 version 1.0.1 provided eight sessions, 45 sites, and 481 readout pairs. The first five pulses supplied coupling estimates from separate proxy contacts within 15 mm. The final five supplied response power at 20 to 500 ms. Each person was excluded from model fitting. Mean squared error changed from 64.319 to 65.112 dB squared when coupling was added to geometry. The mean error reduction was -0.792 dB squared (-1.23%), with a whole-person 95% bootstrap interval of [-3.623, 1.215]. An added exploratory test supplied early response power to the model. Coupling then gave an error reduction of -0.063 dB squared, with interval [-0.618, 0.414]. The test used nearby proxy contacts and processed data from people with epilepsy. It did not test pulse savings, a clinical benefit, or control of a brain interface. The package preserves source checks, code, compact inputs, forecasts, figures, and rebuild instructions.

SEEG · electrical stimulation · brain networks · response forecasting · signal coupling · ridge regression