Action-related organization of object representations in human visual cortex and topographic models

Decades of research have shown that object representations in occipitotemporal cortex (OTC) are organized by multiple spatial and functional principles, including category-selective areas and large-scale topographic maps structured by properties such as animacy and real-world size. In parallel, recent work has proposed a functional dissociation within OTC, with a ventral pathway supporting object recognition and a lateral pathway supporting action-related processing. However, how these functional pathways relate to the spatial organization of object representations remains unclear. The central aim of my work is to bridge these frameworks by identifying action-related object properties as an additional organizing dimension of OTC. Across neuroimaging and computational modelling studies, I show that action-related information is a key organizing principle of lateral OTC, operating alongside - but independently from - classic object dimensions. Using a stimulus set that dissociates animacy from action properties, I demonstrate that representations in lateral OTC form a smooth topographic gradient reflecting object action-related information, whereas ventral OTC is primarily organized by animacy. Furthermore, in subsequent studies using a combination of image-level functional selectivity analyses and encoding models, I show that body-, hand-, and tool- and even food-responsive areas exhibit robust category selectivity, and multiple areas selective for the same category differ systematically in their feature sensitivity depending on their position within a pathway (ventral vs lateral) or hemisphere. Finally, comparisons with (topographic) artificial neural networks reveal that current models capture aspects of ventral visual organization, such as animacy and surface features, but fail to capture the action-based organization of lateral OTC. Together, these findings identify action-related object properties as a core organizing principle of human visual cortex and motivate new biologically grounded computational models.