One shared site scale gave a mean patient error increase of 0.93% for 20–500 ms SEEG forecasts after five calibration pulses in eight sessions

We tested whether all bipolar readouts at one stimulation site can share a waveform forecast scale from five early calibration pulses. Public SEEG from OpenNeuro ds004696 version 1.0.1 supplied eight epilepsy patient sessions, 86 sites, 886 stimulation-to-readout pairs, and 430 later physical test pulses. The main window was 20 to 500 ms after stimulation. Sharing reduced the fitted scale count from 886 to 86. Mean patient error reduction relative to separate readout scales was -0.93%, with a descriptive 95% patient bootstrap interval of -2.12% to +0.40% and an exact two-sided sign-flip p value of 0.203. Sharing had lower error in two of eight patients. The result did not establish lower error or equal performance. A fixed half-sharing control gave +0.36% mean error reduction, with an interval that also included no gain. Both rules used the same five calibration pulses. This exploratory comparison reused test recordings and did not test a new day or live brain interface. The package preserves exact trial products, fitted scales, every test loss, three figures, a fixed local plan, source hashes and retrieval code, independent source audit records, and reproduction instructions. Fresh checks rebuilt all stored matrices and display waveforms from all eight pinned source derivatives, with exact matches.

stereoelectroencephalography · electrical stimulation · waveform forecasting · calibration · signal processing · brain-computer interfaces