The nearest same-shaft SEEG stimulation pair adds a small held-out N1 prediction gain in 15 patients, mainly for adjacent pairs

In an offline analysis of archived site-by-contact N1 tables from three OpenNeuro CCEP datasets, we tested whether the measured N1 map of the nearest stimulated site on the same SEEG shaft improves prediction of a held-out site's map beyond log distance and contact propensity. The primary analysis included 15 patients and targets with a neighbor no more than two contact steps away. Adding the neighbor map increased mean held-out AUC from 0.750 to 0.776 (difference +0.026, 95% CI 0.003 to 0.048; 12/15 patients improved; exact sign-flip p = 0.045). Almost all primary targets were adjacent bipolar pairs that share one contact, so the result mainly applies to adjacent pairs. Copying the neighbor map gave AUC 0.785, but mean sensitivity was 0.60 and mean positive predictive value was 0.58. N1 map similarity decreased with separation. The AUC gain differed by dataset: +0.043 for ds005448 (8 patients), +0.026 for ds004696 (4), and -0.021 for ds004080 (3). One dataset contributed most patients and the largest gain; N1 criteria and stimulation currents differed across datasets. Contact separation is measured in contact steps, not millimeters. The analysis used archived binary tables and did not assess amplitude, latency, or closed-loop use.

SEEG · CCEP · N1 response · electrical stimulation · brain connectivity · held-out prediction · epilepsy