Best ECoG Contacts Outperformed Best Clinical SEEG Contacts for Single Visual Stimulus Detection in 7 of 8 Patients

In eight patients with both depth (SEEG; 217 contacts) and surface (ECoG; 932 contacts) recordings, we compared single-trial visual stimulus detection using authors’ broadband 70–200 Hz power from the spatialpattern task in OpenNeuro ds004194. For each contact, we measured response-versus-baseline rank AUC across 84 full-contrast trials. We selected each modality’s best contact on odd trials and evaluated it on even trials, then reversed the split. The best surface contact had higher held-out AUC in 7 of 8 patients (median 0.87 versus 0.60; mean paired difference, depth minus surface, −0.24; exact two-sided sign-flip p = 0.016). With the number of contacts matched, the difference was smaller (−0.12, p = 0.055). Typical contacts showed no modality difference (median AUC 0.52 versus 0.53, p = 0.95; FDR-responsive contacts 28% versus 32%, p = 0.97). In one patient, 24 of 25 depth contacts responded and the best depth contact reached AUC 0.999; AUC increased along shafts from anterior to posterior. Contact location and modality were confounded: surface grids covered visual cortex, while most clinical depth shafts targeted mesial temporal structures. The result concerns offline stimulus-versus-baseline detection with known onset in one task, not stimulus decoding or a comparison at matched cortical sites. Coordinates were in native space, and some were likely interpolated. Source file URLs and SHA-256 hashes are preserved, but the recorded live-mirror URLs do not establish which version-specific dataset snapshot supplied every file.

SEEG · ECoG · single-trial detection · visual cortex · intracranial EEG · broadband power · AUC