Lower-level stimulus features strongly influence responses in the fusiform face area

Abstract

An intriguing region of human visual cortex (the fusiform face area; FFA) responds selectively to faces as a general higher-order stimulus category. However, the potential role of lower-order stimulus properties in FFA remains incompletely understood. To clarify those lower-level influences, we measured FFA responses to independent variation in 4 lower-level stimulus dimensions using standardized face stimuli and functional Magnetic Resonance Imaging (fMRI). These dimensions were size, position, contrast, and rotation in depth (viewpoint). We found that FFA responses were strongly influenced by variations in each of these image dimensions; that is, FFA responses were not "invariant" to any of them. Moreover, all FFA response functions were highly correlated with V1 responses (r = 0.95-0.99). As in V1, FFA responses could be accurately modeled as a combination of responses to 1) local contrast plus 2) the cortical magnification factor. In some measurements (e.g., face size or a combinations of multiple cues), the lower-level variations dominated the range of FFA responses. Manipulation of lower-level stimulus parameters could even change the category preference of FFA from "face selective" to "object selective." Altogether, these results emphasize that a significant portion of the FFA response reflects lower-level visual responses.

Figure 1.

Responses to variations in face size. (a) The hypothesis of size invariance is shown as a solid line. In the alternative hypothesis (dashed line), a population response based on the CMF produces increasing responses as face size is increased as a linear function on a logarithmic scale. (b) Results in FFA. The FFA response is consistent with the CMF hypothesis but not with the size invariance hypothesis. (c) The FFA response correlated highly with the CMF. The CMF function is (log(x)/0.063 + 11.36), where x is size in averaged diameter. (d) The FFA and V1 responses correlated highly. Two y-axes are displayed showing the absolute percent changes in both FFA (left y-axis) and V1 (right y-axis) and their amplitude-normalized correlation. (e) Correlation in individual voxels in V1. Voxels in V1 showed clear retinotopic correlations. (f) Correlation in individual voxels in FFA. Voxels were largely devoid of this retinotopic effect. For voxel-wise correlation analysis, each voxel’s activity within the ROI was extracted based on GLM for each condition and converted to a percent signal change. The correlation was done for each individual subject and each hemisphere.

Figure 2.

Responses to variations in visual field position of face stimuli. (a) The hypothesis of position invariance (solid line) predicts an equivalent response in FFA irrespective of visual field position. Alternatively, the CMF hypothesis (dashed line) predicts a response that decreases according to the CMF variation (e.g., Figure 1). (b) Stimulus configuration. Within a given block, faces were presented in one of the visual field locations outlined by dashed black lines. (c) Results from FFA combined across all 4 axes. FFA activity decreases systematically with increasing stimulus eccentricity consistent with the CMF hypothesis. (d) The FFA response curve (solid line) was near-identical to that in V1 (dashed line). (e) FMRI variations in V1 due to variations in position along 4 major visual field axes. As expected, the steepest decrease resulted from moving the face into the ipsilateral visual field by the most direct path along the horizontal meridian (green). Movement in the opposite direction (contralateral visual field, red) produced the highest overall activity. The slope of these 2 curves was a nonmonotonic function reflecting 2 competing factors: 1) activity decreased with increasing face eccentricity, as in the other measurements, but: 2) activity increased/decreased more at the first offset from zero as the face shifted from half-viewed in a given hemisphere (central position) to whole viewed (all other positions). As expected, activity along the vertical meridian (yellow and cyan) decreased with a slope intermediate to those along the horizontal meridian. (f) Variations in FFA activity as a function of position, otherwise as in panel (e). The changes in activity are qualitatively similar to those in V1 (see e) except for a single point. However, the range of response difference is compressed relative to that in V1, as one would expect from larger receptive fields in FFA.

Figure 3.

Responses to variations in facial contrast. (a) Hypotheses for changes in FFA activity due to variations in contrast level of face stimuli. The “contrast invariance” hypothesis is shown as a solid line. Alternatively, FFA responses might increase with increasing contrast, as in the V1 prediction (dashed line), based on grating stimuli (from Tootell et al. 1995). (b) Stimulus examples. (c) FFA activity to faces increases at progressively higher contrast (relative to the uniform gray baseline stimuli), similar to the V1 prediction based on gratings (dashed line in panel a). (d) Based on the face stimuli, the contrast gain response in FFA (solid line) was similar to that in V1 (dashed line). The contrast levels shown here are not exact due to lack of control over publication displays.

Figure 4.

Response to variations in face viewpoint (rotation in depth). (a) Two hypotheses for FFA activity. The solid line shows the rotation-limited invariance hypothesis. Since the face is occluded beyond 0 ± 135°, a baseline response is predicted to those stimuli. The second hypothesis (dashed line) is the prediction from the lower-order model incorporating the contrast and CMF influence in FFA. (b) Stimulus examples. Half of the tested rotations are illustrated; the remaining stimuli are mirror symmetrical. We also tested the responses to a sphere similar in size to the back of the head (bottom right). (c) fMRI responses in FFA. Results from left and right rotation angles have been combined. Normalized responses to the sphere are shown in dashed line. (d) The variation in FFA response (solid line) correlated well with the variation produced by the model (dashed line). (e) An analogous result was found in V1. (f) Responses from FFA (solid line) and V1 (dashed line) correlated highly.

Figure 5.

Sensitivity to stimulus category and parameters. (a) Examples of the stimuli from left to right: 1) optimized faces (12.7° diameter, 100% contrast); 2) computer-generated objects (“blobs”), with lower-level features matched to the optimized faces; 3) nonoptimized faces (1.02° diameter; 14.14% contrast). The white scale bar represents 1° for the nonoptimized face and 6.3° for the blob and optimized face. (b) Average FFA response to each stimulus type.