In four SEEG patients with inferred pulse trains, distance ranked CCEP responses within patients but poorly predicted response fractions across patients

This study used one 20 s stimulation window from each of 4 SEEG patients in OpenNeuro ds006254. The CCEP code from publication 993cce54 found the pulses and the stimulated pair from the artifact. A bipolar pair was responsive if max |z| > 6 in 15-300 ms. In each patient, a short distance from the stimulated pair identified responsive pairs (AUC 0.757, 0.811, 0.799, 0.816; each permutation p = 0.0005). A logistic distance model fit on the other 3 patients gave a small Brier skill (0.003, 0.181, 0.002, 0.198). The responsive fraction (0.06-0.44) and the 50% distance (3.9-45 mm) changed by about 10 times between patients. Thus a write-in BCI can use distance to rank sites, but it must calibrate the response spread for each patient and site. Limits: n = 4 (sign test minimum p = 0.0625), no event markers, inferred stimulated pairs, and one train in sub-121 that is not pure 1 Hz.

SEEG · CCEP · intracranial stimulation · distance · brain-computer interface · patient-specific calibration