In OpenNeuro ds004194, clinical SEEG gave weaker single-trial visual read-out than equal-count ECoG in 7 of 8 patients, but stronger in 1

In eight patients with simultaneous ECoG and SEEG during the OpenNeuro ds004194 spatialpattern task, we compared single-trial broadband power (70–170 Hz) using leave-one-run-out shrinkage LDA and equal contact counts (50 random ECoG subsets per patient). SEEG stimulus-presence AUC was lower than equal-count ECoG in seven of eight patients (median difference −0.25; two-sided exact sign test p = 0.070); bipolar SEEG showed the same direction (−0.23; p = 0.070). The exception, p05, had 20 responsive SEEG contacts of 25 and SEEG AUC 0.97, above all 50 ECoG subsets. SEEG identity accuracy was near chance (0.07–0.15) in all patients. In p11, 31 of 32 SEEG contacts were outside the Benson14/Wang15 visual maps. These results suggest that clinical depth-contact location can limit visual read-out; they do not establish a general depth-versus-surface difference. The sample was eight patients, and the sign tests did not reach p < 0.05. Only 48 of 58 runs were processed; p01 used 60-Hz band limits despite 50-Hz line noise; contact tests used a normal approximation; and identity decoding used one 0.5 s power window. The analyzed files came from an unversioned public S3 URL; source-file hashes are preserved in the run record.

SEEG · ECoG · visual cortex · broadband power · brain-computer interface · single-trial decoding · intracranial EEG · OpenNeuro