In 8 SEEG patients, geometry-based selection reached 80% of N1-reachable contacts with a median of 5 sites; measured-response oracle needed 3

We asked how many single-pulse stimulation sites are needed to reach most contacts that respond to at least one clinically tested site, and whether contact geometry can guide site selection before stimulation. We used the archived expert-checked N1 table for 8 SEEG patients (335 tested pairs) from OpenNeuro ds005448 snapshot 1.0.0. A greedy order selected from measured responses reached 80% of reachable contacts with 1 to 4 sites (median 3); this in-sample oracle is an optimistic upper reference. A leave-one-patient-out log-distance model using contact positions needed 2 to 12 sites (median 5), and random orders needed 5 to 14 (median 10). Mean coverage over sites 1 to 10 was higher for geometry than random in all 8 patients (mean difference +0.155; exact sign and Wilcoxon p = 0.0078), and lower than the oracle in all 8 (mean difference -0.166; p = 0.0078). Adding stimulated-pair tissue labels did not consistently improve the geometry order (AUC10 higher in 3 patients and lower in 4; Wilcoxon p = 0.81). These results suggest geometry can guide selection before stimulation, while a short measured survey adds value. The study included 8 epilepsy patients from one center and clinically tested pairs only; N1 is an early evoked response, not evidence of behavioral utility or controllability. The tests were not corrected for 12 comparisons.

SEEG · single-pulse electrical stimulation · N1 · brain connectivity · brain-computer interface · site selection · coverage · OpenNeuro