Measured N1 maps reach 80% of reachable SEEG contacts with 2–5 stimulation sites in eight patients

Question: how many stimulation sites must a write BCI use to put an evoked N1 response on most SEEG contacts, and can electrode geometry choose the sites before stimulation? Data: expert-checked N1 labels of 8 patients in OpenNeuro ds005448 v1.0.0 (335 clinical stimulation sites, 778 contacts). A site reaches a contact if the contact has an N1 at 10 mm or more. Method: we added sites one at a time in 5 orders: greedy set cover on the measured map (in-sample reference), greedy set cover on half of the contacts scored on the other half, a leave-one-patient-out logistic distance + white-matter (WM) model, distance only, and random order. Results: the measured-map plan reached a median 43%, 76% and 90% of reachable contacts with 1, 3 and 5 sites; 80% needed 2-5 sites (random: 6-20). The half-contact plan transferred to unseen contacts (8/8 patients better than random, sign-flip p = 0.008), so hub sites are a property of the site. The geometry + WM plan was better than random (7/8, p = 0.023) but below the measured-map plan. Distance alone was not better than random (5/8, p = 0.20). Predicted and measured site out-degree correlated in 8/8 patients (median rho 0.39). The same pattern occurred with N1 at all distances. Limits: clinical site sets, coverage relative to reachable contacts (median 51%), 1-2 mA, N1 is not a functional effect, 8 patients from one center.

SEEG · single-pulse electrical stimulation · cortico-cortical evoked potentials · N1 response · brain-computer interface · coverage · white matter · electrode geometry