Early timing fits increased mean relative late SEEG forecast error by 1.92 percent compared with amplitude fits in five sessions after four calibration pulses per known site

We tested whether the early response to a stimulation pulse can improve a forecast of its later response. We used public stereoelectroencephalography (SEEG) data from five patient sessions in OpenNeuro ds004457 version 1.0.2. The test had 183 stimulation sites, 15 recording pairs, and 732 test pulses. The first four pulses at each site set a mean template. The final four pulses tested each forecast. The site was known. A fit used only voltage from 20 to 100 ms of each test pulse. It chose one offset within 20 ms and one amplitude value for all recording pairs. The target was centered bipolar voltage from 100 to 480 ms. The control fit the amplitude with the offset fixed at zero. Timing increased mean relative error by 1.92 percent after equal patient weighting. Error increased in four of five sessions. The descriptive 95 percent patient interval was -0.07 to 4.91 percent. The exact two-sided patient sign-flip p value was 0.1875. The fixed sensitivity checks gave no mean timing gain. This result does not establish a general harm or a neural delay. All 110 public source files passed hash checks. A fresh extraction and a repeated analysis matched the saved arrays and scores. The package has voltage bins, source hashes, code, pulse forecasts, and three figures. This is an offline test with source quality labels from the full record. It does not test a live brain-computer interface.

stereoelectroencephalography · SEEG · electrical stimulation · waveform forecasting · signal processing · brain-computer interface