Regression fitted to voltage before pulses reduced mean SEEG shape error 100 to 500 ms after later pulses in five patients with simultaneous receiver data

We tested simultaneous waveform reconstruction in public SEEG data from OpenNeuro ds004457 version 1.0.2. The test used five patients, 19 anatomy-selected contacts, and 120 stimulation blocks. A scalar regression slope used only voltage before five training pulses. A target's after training mean and simultaneous donor test voltage reconstructed its shape on five later pulses. The primary window was 100 to 500 ms after stimulation. The equal patient mean squared error reduction was 80.95 percent relative to the training-mean control. All five patients had positive reductions. The patient bootstrap 95 percent interval was 59.31 to 93.01 percent. The two-sided patient sign test gave p = 0.0625. Correct pulse matches gave lower error than four wrong matches in every patient. A disjoint bipolar check gave 14.70 percent mean reduction in three patients. A synthetic common reference produced a large gain without an evoked neural response. The result supports recorded waveform reconstruction. It does not establish neural coupling or removal of noise. All 9,414 input epochs rebuilt exactly from pinned raw files. Code, inputs, predictions, results, checks, and repeat instructions are preserved.

stereoelectroencephalography · SEEG · electrical stimulation · waveform reconstruction · simultaneous prediction · signal processing · five-patient analysis · shared voltage