Abstract
Developmental prosopagnosia (DP) is a neurodevelopmental condition characterized by face recognition deficits. fMRI studies comparing groups of neurotypical (NTs) and DP participants (DPs) have found differences across the face network. However, it has not been possible to confidently interpret the fMRI results from individual DPs using standard category-level localizer studies. Here we attempt a different strategy by comparing the image-level responses between NTs and DPs. We performed extensive scanning (~10h, over 5 fMRI sessions) of 4 neurotypical subjects and 1 DP participant with naturalistic images (faces, scenes, bodies, objects, etc.). As previously reported, we find that the DP has all the category-responsive regions in similar locations as NTs albeit with lower selectivity. We next scrutinized the response in FFA to the naturalistic stimuli. We observe that the pattern of response was more similar between any two NTs than between any NT and the DP participant (median R, within NTs = 0.80, between NTs and DP = 0.69, P = 0.010 ranksum test). Importantly, this difference in tuning correlation was specific to faces only (median R, within-NTs = 0.49, between NT-DP = 0.20, P = 0.001), and not present for bodies, scenes, or objects (each P > 0.2, ranksum test). These results indicate the FFA in the DP is not encoding faces normally. We probed this result further by building encoding models for the FFAs of each NT and the DP. These subject-specific encoding models (based on the CLIP-ResNet50 model) showed comparable accuracy for all participants at predicting the responses to stimuli (cross-validated) and also reproduced differences observed in the data. Together, these findings suggest that intensive image-level scanning can allow investigation of the neural characteristics of individual DPs and enable computational modeling efforts to further understand DP.