Object parsing in the left lateral occipitotemporal cortex: Whole shape, part shape, and graspability

Small and manipulable objects (tools) preferentially evoke a network of brain regions relative to other objects, including the lateral occipitotemporal cortex (LOTC), which is assumed to process tool shape information. Given the correlation between various object properties, the exact type of information being represented in the LOTC remains debated. In three fMRI experiments, we examined the effects of multiple levels of shape (whole vs. object parts) and motor-related (grasping; manipulation) information. Combining representational similarity analysis and commonality analysis allowed us to partition the unique and shared effects of correlated dimensions. We found that grasping manner (for pickup), not the overall object shape or manner of manipulation, uniquely explained the LOTC neural activity pattern (Experiments 1 and 2). Experiment 3 tested tools composed of two parts to understand better how grasping manner was computed from object visual inputs. Support vector machine analysis revealed that the LOTC activity could decode different shapes of the tools' handle parts but not the tools' head parts. Together, these results suggest that the LOTC parses tool shapes by how it maps onto grasping programs; such parsing is not fully based on the whole-object shape but rather an interaction between the whole (where to grasp) and its parts (distinguishing the shape for the grasping part for specific grasping manners).