Abstract
The aim of this study was to measure quantitatively which facial characteristics determine the signal in the human fusiform face area (FFA). Methods: We presented synthetic faces and modified i) the amount of “face information” (internal features only, head shape only, entire face), ii) their contrast and iii) their facial geometry (defined by how much a face differed from the population mean). The FFA was localised individually in each observer by contrasting the activation to grey-scale photographs of faces from houses. Subsequent measurements were focussed within this region of interest. The BOLD signal was measured using a 4T, whole body MRI system. Voxel volume was 3×3×6mm collected using a T2*-weighted, interleaved EPI acquisition. Subjects engaged in a modified one-back same-different task during scanning. Results: i) Synthetic faces elicited, on average, 85% of the activation of grey-scale face photographs in the FFA. ii) The BOLD signal for isolated face features and isolated head shapes was significantly above baseline (mid-grey background or textured patterns). The signal for isolated features was about the same as the signal for the full face. Both yielded a stronger signal (about 25%) than the head shape alone. iii) The signal in the FFA was independent of the contrast of the face for contrasts above 40% but activation was reduced (to about 70%) for lower contrasts (20%). iv) The BOLD signal showed a dependence on facial geometry: the higher the geometric distance from the mean face, the stronger the activation. Conclusions: These results support the notion of Valentine's face space, in which the distance from the mean is a measure of distinctiveness. They have strong implications for how faces are computed in the FFA and, more generally, how they are represented in human cortex.
Supported by an EPSRC grant to GL.