In one patient, N1 AUC rose with pulse averaging for SEEG-to-ECoG responses, while specificity fell at a 3.4 SD threshold

This analysis asked how averaging 1 to 10 single-pulse electrical stimulation responses affected automatic N1 detection against expert labels for SEEG depth stimulation recorded on ECoG contacts in one patient (OpenNeuro ds004774, sub-UMCU59). The source contained 14 stimulation pairs, 10 pulses per pair at 4 mA, and 658 expert labels, including 77 responses. For each pulse count, 100 random pulse subsets (seed 0) were averaged. The N1 score was the negative minimum from 9 to 100 ms divided by baseline SD from -100 to -10 ms. AUC increased from 0.773 (SD 0.033) at 1 pulse to 0.903 at 4 pulses and 0.954 at 10 pulses. At the fixed 3.4 SD threshold, sensitivity rose from 0.760 to 1.000 while specificity fell from 0.652 to 0.389. Thus the score ranked expert-labelled responses well with 4 or more pulses, and AUC continued to increase through 10 pulses; 10 pulses was not a saturation point in this patient. The short 90 ms baseline made the SD estimate noisy and the fixed threshold too lax. A possible cause, not tested, is that averaging reduces baseline SD more than slow background deflections and residual artifact in the N1 window. This is one patient, one rater, and 14 stimulation pairs; the labels cover ECoG recording contacts only, with no SEEG-to-SEEG labels. Random pulse subsets do not model time order or habituation. Do not use the 3.4 SD threshold without calibration or generalize these findings to SEEG-to-SEEG responses.

single-pulse electrical stimulation · evoked potentials · N1 detection · SEEG · ECoG · pulse averaging · expert labels · AUC · specificity · OpenNeuro