Abstract
The adult human ventral temporal cortex (VTC) contains regions with strong category selective responses. When and how do these arise in development? fMRI studies in awake human and macaque infants (Deen et al., 2017; Livingstone et al., 2017) found adult-like regions with reliable preferences for faces>scenes or scenes>faces that lacked full category selectivity (i.e., these regions did not show faces>objects or scenes>objects). In Experiment 1, we replicated these results with a new group of awake infants (n=16; 2.5-8.7 months) using the same infant coil and quiet pulse sequence. Infants watched videos of faces, objects, bodies, and scenes. High head-motion data were excluded as in Deen et al (2017). Voxels were selected for each contrast in each infant using a subset of the data, and response magnitudes in those voxels were quantified in held-out data (see Figures 1 and 2); these response magnitudes were then compared across subjects in a paired t-test for each region and contrast. We replicated previous findings that infants have reliable face>scene and scene>face responses in VTC that are not fully category selective (i.e., these regions do not respond more to faces> objects and scenes>objects respectively). In a second experiment (n=8; 4.6-9.3 months), we optimized infant data acquisition with a variety of technical innovations: 1) an infant coil that increases the signal-to-noise ratio by 25%, 2) novel MR-safe headphones, and 3) an improved pulse sequence. The analysis pipeline was identical for both experiments. This experiment replicated face and scene preferences and the lack of selectivity for scenes>objects. Crucially, however, we found regions in infant VTC that responded significantly more to faces than to scenes, bodies, and objects. These data suggest that fMRI noise was a major limitation in previous infant fMRI experiments and show the first fMRI evidence for truly face-selective cortical regions in human infants.