Abstract
Humans can search for and interact with objects based on many properties including pose, texture, and shape. Flexible perception is thought to rely on a network of areas engaged by many tasks, termed the Multiple Demand Network, spanning regions in the parietal, frontal, and lateral occipital cortex. However, the degree of functional specificity vs domain generality among regions in the MDN is still unclear. To address this question, we constructed an fMRI experiment in which human participants performed three different tasks while viewing the same set of simple three-part shapes. For each trial, participants either passively fixated or indicated whether a) medial axis structure or b) set of local shapes comprising the object or c) object texture was different from that of the previous object. Performing multiple tasks while presenting the same stimulus allowed us to induce maximum variance along task dimensions of voxels independent of the stimuli. We created a simple encoding model with indicator variables for task and found that it explained unique variance over and above stimulus-based models in regions of the frontal, parietal, and lateral occipital cortex. Among these regions, we found that posterior LOTC ventral to hMT+ and dorsal to OFA responded more in the local shape task. A region just anterior to this responded more in the medial axis task. The texture task engaged more posterior regions. We also found spatially distinct activations in the IPS for different tasks. Responses in IPS did vary with subjective task difficulty, which varied across participants, but some IPS regions were reliably activated by the medial axis task. The involvement of distinct regions of the parietal cortex in computing local shape and medial axis point toward functional specificity of regions within the Multiple Demand Network, with different regions representing relatively more global and local aspects of object shape.