A fixed linear model increased mean relative squared position error for excluded shafts in five offline SEEG tests at 6 mA

This study tested whether a fixed linear model can locate stimulation sites with no fitted response template. It used public OpenNeuro ds004457 version 1.0.2, five sessions, 137 remote sites at 6 mA, and 548 distinct test pulses. Each estimate used the mean of four pulses from 20 to 100 ms. Each main model excluded the full target shaft from noise and position fitting. The ridge weight was fixed at 0.1. Mean relative squared position error was 1.635, where 1 is the error of the mean fitting position. The increase was 63.48%, and four of five patients had more error. Mean patient RMS error was 54.61 mm for the model and 42.38 mm for the reference. The exact patient sign-flip reference was 0.125. Excluding only a site gave a better voltage score, but event time alone also predicted position in that split. The package supplies a fixed spatial test, source code, voltage bins, 30 primary models, controls, three figures, and reproduction instructions. All 110 source files passed hash checks. Fresh extraction reproduced all ten input files byte for byte. A second analysis reproduced its tables exactly. These results apply to this fixed model and these reused offline sessions. Spatial coverage, timing, and the small cohort limit claims about future interface use.

stereo-electroencephalography · electrical stimulation · brain interfaces · position prediction · ridge regression · shaft exclusion · offline test