Abstract
Human brain networks involved in visual perception have been extensively investigated with fMRI studies such as the Natural Scenes Dataset (NSD), in which 8 healthy participants viewed thousands of naturalistic images. However, spectro-temporal characteristics are limited in fMRI. In contrast, intracranial EEG (iEEG) boasts sub-millisecond resolution and meaningful spectral power changes, but is limited by sparse spatial sampling. By collecting a large, high-quality iEEG dataset with matching stimuli to an fMRI dataset, we enable integration of the two modalities to leverage benefits of both. We recorded iEEG data in 12 human participants implanted for epilepsy monitoring, while they viewed the same 1000 stimuli presented to the NSD-fMRI participants. At each electrode, broadband power was calculated as a measure of local neuronal activity. Electrodes with significant signal-to-noise ratio across 6 repetitions of a 100-stimuli subset were considered visually responsive. Of 1650 total electrodes, 92 were visually responsive: 16 early visual, 10 lateral occipital, 5 dorsal occipital, 31 ventral temporal, 6 lateral temporal, 2 temporal pole, 18 frontal, 2 cingulate, and 2 amygdala. To integrate the iEEG and fMRI datasets, broadband power at each electrode was correlated with fMRI beta weights at all vertices across stimuli, yielding brain-wide correlation maps. 73 electrodes exhibited significantly positive correlations with fMRI vertices within a 3 mm radius, indicating robust local correspondence. Globally, most electrodes showed one of five spatial correlation patterns, which each consisted of positive and negative correlations across early visual areas and ventral, lateral, and dorsal streams; and which showed preference for contrast, places, faces, food, and bodies. This novel dataset integrates iEEG and fMRI on the same set of naturalistic stimuli. The recorded iEEG activity correlates well with fMRI and reveals global patterns of stimulus preference. This dataset opens the door to time-frequency analyses that can elucidate high-resolution dynamics within these networks.