Abstract
A recent study reports that face-selective regions respond more strongly to faces of increasing distinctiveness, defined as deviation from the mean or most typical face (Loffler et al., 2005). In their study, physical variability was smaller among typical faces than among distinctive faces, likely resulting in more fMRI-adaptation, and in turn a lower response, to typical faces. We posit that if physical variability were equated across typical and distinctive faces, face-selective regions might respond more strongly to typical faces, as they are more frequently encountered exemplars of the face category. Here we manipulated face distinctiveness while equating physical variability by using parameterized two-tone silhouetted face profiles (Davidenko, 2007). Nine subjects participated in several high-resolution (1.5mm isotropic voxels) fMRI sessions. Face-selective regions in the fusiform gyrus (defined by gray-scale faces [[gt]] gray-scale objects, p[[lt]]0.001) responded preferentially to upright face silhouettes compared to similarly constructed nonface shape silhouettes or upside-down face silhouettes, demonstrating that face silhouettes preferentially activate face-selective regions. Next, we manipulated distinctiveness by parametrically varying face silhouettes along a principal dimension of silhouette face space highly correlated with distinctiveness ratings. We defined 9 groups of face silhouettes: one centered at the mean face silhouette, and 4 centered at increasing steps from the mean in each direction along this dimension. At each distinctiveness level, subjects viewed different exemplars with the same physical variability. Responses in face-selective regions decreased as distinctiveness increased, while responses in object-selective regions were not affected by distinctiveness. Further, responses in face-selective regions were highly correlated with how “face-like” subjects rated the silhouettes in a post-scan behavioral task. We conclude that controlling stimulus variability is critical for characterizing response properties of face-selective regions. By controlling face variability, we show that face-selective regions respond preferentially to typical faces, likely reflecting the dense distribution of face exemplars near the mean face.
This research was funded by Whitehall grant 2005-05-111-RES & Klingenstein fellowship to KGS and NSF grants 0547775 & 0624345 to MR.