A radial template space reveals organization of feature and task-selective regions in lateral occipitotemporal cortex

Human lateral occipitotemporal cortex (LOTC) represents information about objects and object properties, motion, and bodies, and is also modulated by tasks, including mental imagery, working memory, and attention to different features. However, the organization of this region is still debated, partly because variability in anatomy blurs cross-subject functional maps. We address this by developing a flexible template space for LOTC to average responses across participants. Following work that has suggested organizing principles centered on motion-selective region hMT+, we use hMT+ to define the center of a radial grid. We fix the polar angles of the grid to evenly sample space between anatomically and functionally defined regions including the superior temporal sulcus, the Occipital Place Area, and the Fusiform Face Area. We define the radial dimension by distance along the cortical surface from the hMT+ centroid. We then use data from multiple experiments to investigate the organization of feature and task representations in LOTC. In previous experiments, we developed fMRI encoding models based on motion energy, object boundary contours, and body parts and their locations. We project the prediction accuracy of these models into the radial grid space and average across subjects. The models each predict unique variance (relative to each other) in fMRI responses to rendered and naturalistic stimuli in different locations around hMT+: the boundary contour model predicts responses posterior to hMT+, the body part model superior and anterior, and the motion energy model in the center of hMT+. Separate analyses of experiments in which participants performed tasks while watching stimuli reveal that object-focused tasks (searching for objects and a one-back task comparing object parts) activate regions ventral and anterior to hMT+, while a self-motion focused task activates a region anterior and ventral to hMT+. Together, these results describe a discrete organization of feature and task selectivity surrounding hMT+.