Across two patients, category decoding showed no measurable duration trend; SEEG image detection was strong in sub-p14

We asked whether shorter image presentations (17–533 ms) reduced single-trial image detection and six-category decoding from intracranial broadband power. We analyzed six runs from two patients in the sixcatloctemporal task of OpenNeuro ds004194: four runs from sub-p13 and two from sub-p14. Using adjacent bipolar contacts, 70–180 Hz broadband power, and leave-one-run-out six-class LDA, pooled one-pulse category accuracy was 0.27–0.34 across SEEG and ECoG sets, above the within-run permutation 95th percentiles (0.20–0.22) but far below perfect decoding. Accuracy varied from 0.12 to 0.48 across duration cells, with no monotonic trend; correctness had |Spearman rho| < 0.02 in all four sets. In sub-p14, which had 38 SEEG pairs and many visually located contacts according to the archived derivative list, detection AUC was 0.89–0.96 across durations, including 0.92 at 17 ms; the response peaked at about 100 ms. In sub-p13, most SEEG shafts were outside visual cortex and AUC was 0.58–0.72. A sensitivity analysis using one mean feature per channel reduced decoding to near chance in sub-p14 SEEG (0.222, equal to the null 95th percentile), indicating that temporal profile features carried much of the category information. The evidence is limited to two patients, 24–48 trials per duration, 40 permutations in the main analysis, and unmatched SEEG and ECoG channel sets. Confidence intervals were wide (SE about 0.06–0.09), and effects smaller than about 0.15 in accuracy could not be detected. No multiple-comparison correction was applied across durations. Thus, within these data, 17-ms images were not measurably worse than 533-ms images for category read-out; this does not establish equivalence or generalize beyond the small sample.

SEEG · ECoG · broadband power · visual cortex · image detection · category decoding · stimulus duration · intracranial EEG · LDA · OpenNeuro ds004194