Abstract
Neuroimaging studies have found reliable patterns of response to different object categories in the ventral visual pathway. This has been interpreted as evidence for a categorical representation of objects in this region. However, in addition to their semantic content, categories also differ in terms of their image properties. The aim of this study was to determine if image properties could explain category-selective patterns of neural response in the ventral visual pathway. We hypothesized that, if patterns of response in this region are tuned to low-level image properties, similar patterns of activity should also be evident for scrambled images that contain the same low-level properties, but are not perceived as objects. To address this issue, we generated phase-scrambled versions of intact objects in two ways: 1) globally-scrambled – applied to the whole image; 2) locally-scrambled - dividing each image into an 8x8 grid and scrambling the contents of each window independently. A behavioral study revealed that both scrambling processes rendered images unrecognizable. We then used fMRI to measure patterns of ventral response to five object categories (bottles, chairs, faces, houses and shoes) with three image conditions (intact, globally-scrambled, locally-scrambled). Using multivariate pattern analysis, we found distinct and reliable patterns for all five categories in intact and locally-scrambled image types. In contrast, the globally-scrambled images only showed reliable patterns for faces and houses. In addition, we found that the similarity matrices for the intact and locally-scrambled images were significantly correlated (r=0.79, p< 0.001). However, the similarity matrices from the intact and locally-scrambled images were not correlated with the globally-scrambled images. These results suggest that similar patterns of response are elicited by intact and locally scrambled images. Taken together, these data suggest that category-selective patterns of response in the ventral visual pathway can be explained by image properties typical of different object categories.
Meeting abstract presented at VSS 2015