Figures and data
Schematic diagram of stimuli and procedure in fMRI experiment.
(a) Schematic diagram of stimuli for the 6 conditions in the fMRI experiment. The images from left to right are example for images of male face, female face, male object, female object, male body and female body, respectively. (b) Schematic diagram of a block in fMRI experiment. Participants viewed the stimuli in 12 s blocks, where each block contained images from one condition (female faces shown here). Each image with fixation was presented for 900 ms, followed by a 100 ms fixation between images. The red arrow indicates a repeated image, where the participant should press button. (c) Top panel: Searchlight regression-based RSA was conducted to predict neural RDM structure using a linear combination of theoretical category and gender model RDMs within each searchlight sphere. Bottom panel: RSA linking activation in brain and CNN. Correlations were calculated between RDMs from each layer of a fine-tuned AlexNet and neural RDMs from each region of interest. The example illustrates the correlation calculation between Conv4 layer RDM and neural RDM. For privacy reasons, the example faces are the authors’ own.
Results of whole-brain searchlight analysis for the classification of gender and category.
(a) Brain regions able to classify gender from faces (left panel), bodies (middle panel), and objects (right panel). (b) Brain regions able to classify gender across faces and bodies (left panel), across bodies and objects (middle panel), and across objects and faces (right panel). The statistical maps in (a) and (b) show significant brain regions after Monte Carlo simulations (p = 0.001 at the initial voxel-wise, 5000 iterations; the color scale bar show the z-scores above 1.64, corresponding cluster level p < 0.05). (c) The yellow color depicting overlap areas of three maps for within-category gender classification in (a). The red color depicting overlap areas of three maps (rMTG) for three cross-category gender classification maps (b). (d) Brain regions for category classification, identified by the conjunction analysis of significant maps of category classification on each pairwise categories. (e) The cyan color shows the regions able to classify category. The orange and yellow show the overlapping regions for category classification and within- category, cross-category gender classification, respectively. The blue color shows a few regions able to classify within-category gender, but not category.
Results of RSA.
(a) Regression-based RSA results depicting areas whose activation were significant predicted by category and gender theoretical RDMs. These statistical maps show the significant brain regions after Monte Carlo simulations (p = 0.001 at the initial voxel-wise, 5000 iterations; the color scale bar show the z-scores above 1.64, corresponding cluster level p < 0.05). (b) RSA linking neural responses in rMTG (left panel), V1(middle panel) and PFC (right panel) to CNN layers. * p < 0.05, ** p < 0.01 (FDR- corrected). The data predicted by the linear and quadratic model were plotted using red and black dashed lines.
Results of the PPI analysis.
(a) – (c) The significant functional connectivity (lines in each figure) changes between seeds (blue dots) for the gender processing of face (a), body (b) and object (c). The color scale bar shows −log10(p values) ranging between 1.301 (p =0.05) and 3 (p =0.001), FDR-corrected. (d) PPI matrix for gender processing of face, body and object, and the correlation between these marix. Each PPI value in matrix showed in figure was the mean PPI from all participants.
