Visual-cortex decoding detected a second image at short gaps, with SEEG evidence concentrated in one patient

We used preprocessed epochs from OpenNeuro ds004194 (Groen et al. 2022): 30 SEEG contacts in 7 patients and 128 ECoG contacts in 9 patients. Each test compares one 133-ms image with two 133-ms images and a gap of 17, 33, 67, 133, 267 or 533 ms (12-24 trials per class). A diagonal LDA uses 25-ms bins from 0 to 1.0 s, with 6-fold cross-validation and 200 label permutations. SEEG contacts inside Benson V1-V3/hV4/LO/VO maps (22 contacts, 5 patients) give a median broadband AUC of 0.62-0.78. 5-13 of these contacts have p<0.05 at each gap, including 17 ms. Nearly all significant SEEG contacts are in patient p05. SEEG contacts outside the maps never reach p<0.05. Only p05 gives a significant SEEG multi-contact decoder (AUC 0.82-0.94); the 7-patient SEEG group is at chance for broadband. ECoG multi-contact decoders give a median AUC of 0.62 at 17 ms and 0.94 at 267 ms (group sign-flip p 0.002-0.041). Contact location, rather than electrode type, predicts detection in these data. The SEEG result does not generalize beyond one patient. The decoder uses a 1-s window and therefore does not show that cortex resolves two events in real time. The study has 12 trials per class in most cases; results are uncorrected for multiple comparisons, and contact-level binomial tests treat contacts as independent. The package includes code, result tables, figures, and source hashes.

visual cortex · SEEG · ECoG · single-trial decoding · broadband power · diagonal LDA · OpenNeuro · stimulus gap