A five-pulse geometric median had 0.31% higher mean error than a scaled mean for 20–500 ms bipolar SEEG in eight sessions

We tested a geometric median forecast for centered bipolar SEEG voltage from 20 to 500 ms after a pulse. Both the median and arithmetic mean used five early pulses. Both used the same early leave-one-out rule to fit voltage size. Five separate later pulses supplied the test at 86 sites in eight patients. The analysis used 886 stimulation-to-readout pairs from OpenNeuro ds004696, version 1.0.1. The mean patient error reduction from the matched mean was -0.31%. The descriptive 95% patient interval was -0.86% to 0.19%. The exact sign-flip value was p = 0.359. Three patients favored the median. This test did not show a mean benefit. It does not establish equal performance. All numerical checks passed. Two fresh public source files reproduced the compact inputs exactly. The package contains code, source hashes, exact trial products, every test-pulse score, and three figures. The result applies to offline good-pulse data from one session per patient. It does not establish online BCI performance.

stereoelectroencephalography · SEEG · geometric median · voltage forecast · electrical stimulation · brain-computer interface · robust estimation · forecast benchmark