CNeuroMod-THINGS, a densely-sampled fMRI dataset for visual neuroscience

Area of Science:

• Neuroscience
• Artificial Intelligence
• Cognitive Science

Background:

• Neuro-AI modeling demands extensive neuroimaging datasets.
• Existing initiatives like THINGS and CNeuroMod provide valuable resources.

Purpose of the Study:

• To create a large-scale, densely-sampled fMRI dataset (CNeuroMod-THINGS) for neuro-AI modeling.
• To capture neural representations of diverse semantic concepts using well-characterized images.
• To leverage synergies between the THINGS and CNeuroMod projects.

Main Methods:

• Four participants underwent 33-36 fMRI sessions each.
• A continuous recognition paradigm was employed using 4320 images from the THINGS stimulus set (720 categories).
• Behavioral and neuroimaging metrics were collected to ensure data quality.

Main Results:

• The CNeuroMod-THINGS dataset provides high-quality fMRI data.
• The dataset enables the study of neural representations for a broad range of semantic concepts.
• It supports research in both controlled and naturalistic settings.

Conclusions:

• CNeuroMod-THINGS expands the capacity for human vision modeling.
• The dataset facilitates advancements in neuro-AI by providing rich neural data.
• It demonstrates the value of integrating existing large-scale research initiatives.