Abstract
The face has great significance for social interactions and is the most diagnostic information for identifying individuals in humans. Understanding the neural mechanisms involved in face identity recognition (FIR) is critical, particularly for the individuation of unfamiliar faces, which cannot be based on encoded multimodal semantic associations, but on available visual cues only. To investigate the neural basis of FIR, we capitalize on an original fMRI approach based on fast periodic visual stimulation, providing objective, sensitive, and reliable measures of human face recognition. FMRI recordings were performed in ten healthy human subjects. Natural images of a single unfamiliar identity were presented within a rapid 6Hz stream in two conditions: (1) with the same face image across low-level changes (size, luminance, contrast) only, or (2) with different images, introducing higher-level changes (background, head orientation, expression). Every 9s during a 243s run, 7 images of different unfamiliar identities were introduced in bursts. For each participant and each condition, we recorded 3 runs with upright faces and 3 with inverted faces. Analyses were performed within face-selective regions (defined from a frequency-tagging localizer) and in the Fourier domain where individual face discrimination responses were objectively identified and quantified, at the peak of the identity change frequency (0.111Hz). Robust image-based individual face discrimination responses were found across both conditions in core face-selective ventral regions (FFA, OFA) and exhibited inversion effects, invariant to high-level stimulus changes. In contrast, responses in low-level visual regions and in the pSTS were negligible in our second condition, which involved generalization across changes of views. Interestingly, we also found specific responses to FIR in the IFG which were significantly reduced for inverted faces. Overall, our results highlight the cortical network involved in human FIR and suggest that fMRI frequency-tagging provides a valid approach to characterize the cortical network underlying this function.