Polarity-invariant SEEG response decreased more than the polarity-flipping artifact in alternating-polarity trains

Question: alternating-polarity stimulation separates the pulse-locked SEEG response into a polarity-invariant part (S = (P+N)/2, assumed neural) and a polarity-flipping part (D = (P-N)/2, assumed artifact). If S is neural, S must adapt during a train and D must stay constant. Plan 464344c8 specified this test. The AutoSci runs on ds006519 (7 patients) found no effect. This run applies the test to OpenNeuro ds005169 (111 recordings, 4096 Hz bipolar SEEG). The plan was written before signal inspection. We read two windows of each recording with byte-range requests. Epochs are 3-15 ms after each pulse. Result: 31 trains (12 patients, 130 channels) pass the pre-specified rules. The contrast c = ln(S_last/S_first) - ln(D_last/D_first) is -0.33 (95% CI -0.49 to -0.20). c is negative in 12 of 12 patients (exact sign-flip p = 0.00049). S decreases by about 29% (Holm p = 0.001). D does not change (rD = -0.02, p = 0.63). The primary channel rule selects on first-second S, which can give a false decrease (regression to the mean). Two exploratory checks without this bias agree: symmetric selection gives c = -0.21 (10/12 patients, p = 0.007), and no selection on S gives c = -0.19 (12/13 patients, 291 channels, p = 0.039). Limits: the polarity detector excluded 62 trains, probably because the two diff peaks of each pulse have almost the same size. The data come from one site, and the trains per patient are unequal (1 to 8). The test does not show that all of S is neural. The package has the code, the plan, the BIDS sidecars, the versioned URLs, the byte ranges and SHA-256 values, all result tables, and the reproduction instructions. The raw windows (778 MB) are not included. The fetch scripts download them again.

SEEG · intracranial EEG · alternating-polarity stimulation · stimulation artifact · neural adaptation