In 43.2-50 Hz SEEG trains, responses mostly scale down, but gain-only models leave small shape changes

Question: during a high-frequency SEEG stimulation train, the pulse-locked response decreases to approximately 0.6-0.7 of its initial size. Is this a pure gain change, or does the waveform shape also change? Method: we used the per-run features archived in sci 533cdd82 (ds005169, visual-effect sites) and sci bb7fae83 (ds006519, negative-motor-response sites). For 267 responsive bipolar channels (31 runs, 15 + 7 subjects), we compared the mean half-sum waveform 5-18 ms after the pulse in the first 0.5 s and in the last 1 s of the train. We fit a gain-only model and divided its residual by the pre-stimulus noise (Q). We compared Q with a within-channel control (the last two bins of the train). Results: the noise-corrected waveform correlation was 0.96-0.97 and the peak time did not move (median shift 0 ms). But Q was larger than the control in 18 of 22 subject units (p = 0.002; ds006519 7 of 7, ds005169 11 of 15, p = 0.07; with first-bin SNR > 5, 11 of 14, p = 0.02). In ds006519 the 10-18 ms part decreased less than the 5-10 ms part (7 of 7, p = 0.016); in ds005169 it did not (7 of 15). The shape change did not increase with the artifact-to-response ratio, so a constant residual artifact does not explain it. Meaning for BCIs: one gain factor per channel compensates most of the train decrease, and a fixed template stays well matched, but a small shape change remains. Limits: secondary analysis of a subset of runs, 5-18 ms window only, and the noise model assumes the pre-stimulus noise.

SEEG · intracranial EEG · cortical stimulation · evoked response · waveform shape · gain change · brain-computer interface