Unvalidated IED candidates did not raise detector scores in eight 240-second SEEG blocks

We tested whether interictal epileptiform discharge (IED) candidates raised the output of a causal high-gamma plus beta key-press detector or were linked to excess false detections. We used the same 240-second blocks from eight subjects in OpenNeuro ds004473 v1.0.2, verified byte-for-byte against a prior SCI contribution. An offline IED candidate detector was not expert-validated and detected 24–158 candidates per minute. Among six subjects with enough IED-free frames, the mean standardized difference in detector margin for IED-near versus IED-free no-press frames was −0.13 (two subjects positive; exact sign-flip p = 0.25). Four of nine false detections occurred near a candidate, compared with 3.0 expected by time (pooled circular-shift p = 0.36). Removing channels frequently marked by candidates did not increase evidence of an effect. The IED-triggered margin was flat from −1 to +1 seconds. Evidence is weak: there were only nine false detections, detector mean hit rate was 0.10, and candidate labels were unvalidated, with a high candidate rate that may include physiological transients or noise. These results do not establish absence of an effect of true IEDs.

SEEG · interictal epileptiform discharges · brain-computer interface · causal detection · high-gamma · key-press detector