ECoG contacts outperformed SEEG contacts for visual onset detection in 11 of 13 patients

We compared single-trial visual stimulus onset detection from SEEG depth contacts and ECoG surface contacts in the same visual tasks for 13 patients in OpenNeuro ds004194. Using log broadband power 0.05–0.30 s after onset versus −0.25 to −0.05 s before onset, we selected the best contact on odd trials and tested it on even trials, then reversed the halves. After matching contact counts, mean held-out AUC was 0.64 for SEEG and 0.83 for ECoG; ECoG was higher in 11 of 13 patients (exact sign-flip p = 0.0029; mean difference SEEG − ECoG = −0.19, bootstrap 95% CI −0.27 to −0.10). Count-matched multi-contact LDA gave AUC 0.64 versus 0.85 (p = 0.0024). The fraction of responsive contacts did not differ significantly (0.18 versus 0.24, p = 0.51). Performance varied with patient: the best SEEG contact reached AUC 0.97–0.99 in two patients, while it was near chance in ten. The analysis measured the yield of clinical placement, used no anatomical labels, and tested one broadband feature and stimulus onset only; it does not establish the physical performance limit of either modality.

SEEG · ECoG · visual cortex · single-trial detection · broadband power · brain-computer interface · electrocorticography · AUC