Individual differences in prefrontal coding of visual features

The lateral prefrontal cortex (LPFC) is commonly associated with high-level cognition, such as attention, language and cognitive control. However, recent work has demonstrated that it is also critical for basic perceptual functions including object recognition. Here we characterize the role of LPFC in visual processing with computational models. Using a dataset of human fMRI data at 7T, we built encoding models relating visual features extracted from a deep neural network (the image encoder of a CLIP [Contrastive Language–Image Pre-training] network) to brain responses to thousands of natural images. Within each of the eight subjects, we were able to robustly predict responses in patches of LPFC, most notably in FEF (frontal eye field) and vlPFC (ventrolateral PFC) regions. Leveraging these robust encoding models, we then explored the representational structures and screened for images with high predicted responses in LPFC. We found striking individual differences in the coding schemes of LPFC. In contrast, the coding scheme of the ventral visual stream remains more consistent across individuals. Overall, our study demonstrates the under-appreciated role of LPFC in visual processing and suggests that LPFC may underlie the idiosyncrasies in how different individuals experience the visual world. Methodologically, these findings may also explain why previous group studies have often failed to observe robust visual functions in LPFC, as subjects’ responses may need to be calibrated individually.