Referential SEEG montages produced more far detections than bipolar montage, and most matched near responses

We compared as-recorded monopolar (M), common-average (CAR), common-median (CMR), and bipolar (BIP) references in single-pulse stimulation SEEG from 8 patients in OpenNeuro ds005448 (STReEF). Ten stimulation pairs per patient and 10 pulses per pair were analyzed with an exact sign-flip test of responses from 10–50 ms. Detection was similar across references at 0–20 mm. At 40–60 mm, pooled detection probabilities were 0.008 for BIP, 0.041 for M, 0.044 for CMR, and 0.107 for CAR; the test false-positive rate was 0.002. Far detection rates exceeded BIP in 7/8 patients for CAR (p=0.016) and 6/8 for M and CMR (p=0.031). Far referential detections closely matched the strongest near response (median |r|=0.91–0.96), while far BIP detections did not (median |r|=0.40); only 5–9% of far referential detections were also detected in a bipolar channel containing the same contact. These findings indicate that, in this dataset and protocol, most far responses under referential montages reflected volume conduction or reference contamination rather than a local response. For evoked-response brain-computer interfaces, local montages should be used before far responses are treated as evidence of a connection. The result is limited to one dataset, one protocol, 10 pairs per patient, and the 10–50 ms window; no Laplacian montage was tested. The decay-slope analysis was inconsistent across patients, so no claim is made about it.

stereo-EEG · single-pulse stimulation · volume conduction · reference contamination · bipolar montage · evoked responses · epilepsy