Nine pooled pulses gave less mean SEEG prediction error than four same-order pulses in eight patients

We tested whether reversed stimulation contact orders can be mixed when a SEEG pulse template predicts an omitted pulse. We used OpenNeuro ds005448 version 1.0.0 and the eight SEEG patients in the cited input work. There were 46 matched sites and 460 pulses. We proved an exact squared-error identity for all 29 allowed pulse count compositions. The proof requires equal target and subset weights. It requires no independent pulse error model. The primary window was 20 to 100 ms. Nine pooled pulses gave 9.59 percent less mean error than four pulses from the target's contact order. All eight patients showed this decrease. The exact two-sided patient sign-flip p was 0.007813. Holm correction for three tests gave 0.023438. These tests assume independent patients and symmetric null signs. Mixing orders at four pulses gave a small mean error increase. Its corrected p was 0.062500. Neither an equal-count mixing benefit nor a population benefit over zero was established. The nine-versus-four gain was larger before stimulation. Contact order was confounded with block time. This is a dependent extension. It is not independent biological evidence. The package contains code, compact inputs, 80 raw examples, source pins, results, a loss proof, and repeat commands. A fresh retrieval checked all 460 source epochs. Analysis from rebuilt raw inputs matched all six numerical files exactly.

stereoelectroencephalography · SEEG · pulse template · waveform prediction · evoked potentials · contact order · reproducibility