Investigating action topography in visual cortex and deep artificial neural networks

Area of Science:

• Neuroscience
• Cognitive Science
• Artificial Intelligence

Background:

• High-level visual cortex exhibits topographic organization for categories like animacy and size.
• The role of action properties in shaping visual cortex topography remains underexplored.

Purpose of the Study:

• To propose and investigate action as a fundamental organizing principle in visual cortex topography.
• To evaluate the capacity of topographic deep artificial neural networks (DANNs) in capturing action-based visual organization.

Main Methods:

• Functional magnetic resonance imaging (fMRI) was used to analyze brain responses to images.
• Stimuli included body parts and objects varying in action properties.
• Multivariate analyses were employed to identify organizational principles.

Main Results:

• A topographically organized action gradient was identified in the left lateral occipitotemporal cortex.
• This gradient showed overlapping activations for bodies, hands, tools, and manipulable objects.
• Deep artificial neural networks (DANNs) demonstrated organization based on shape and animacy, but not action.

Conclusions:

• Action serves as a significant organizing principle within the human visual cortex, distinct from animacy and size.
• The identified action dimension advances our understanding of visual cortex organization.
• Current DANN models do not fully capture the action-based organization observed in the human visual system.