Abstract
The human visual system can easily recognize objects across motion and different states of articulation. However it is still unclear how and where dynamic objects are encoded in the brain. Here we use multivariate pattern analysis (MVPA) to investigate the neural representations of dynamic objects. Subjects viewed moving, novel, articulating objects (Pyles & Grossman, 2009) in a slow event-related fMRI experiment. Two different exemplar animations of three moving objects were shown. We employed both region of interest (ROI), and whole-brain "searchlight" approaches in our MVPA. A novel localizer scan presenting moving objects and phase scrambled moving objects was used to identify brain areas selective for dynamic objects. These were found in regions similar to those for static objects, concentrated in the LOC, but also extending to more dorsal areas in the vicinity of MT complex. First, the MVPA classifier was trained on individual animations, and tested using a "leave one run out" approach. Classification accuracy was significantly above chance in the ROIs identified by the localizer. In an assumption-free analysis, a whole-brain support vector machine searchlight identified additional regions of high classification accuracy in early visual cortex focused around the occipital pole, extending dorsally to the middle occipital gyrus. Second, the MVPA classifier was trained on one exemplar of each object, but tested on the other, revealing that the majority of ROIs were not above chance classification. Again using an assumption-free method, a whole-brain searchlight analysis revealed above chance classification across exemplars in an area on the middle occipital gyrus, slightly posterior to MT complex. In contrast, early visual cortex areas were not above chance across exemplars. In sum, much of object-selective cortex is likewise recruited in the perception of dynamic objects. In addition, a dorsal area bordering on more typical object selective cortex appears recruited by moving, articulating objects.
Meeting abstract presented at VSS 2012