What is the middle face patch?

TitleWhat is the middle face patch?
Publication TypeConference Proceedings
Year of Publication2010
AuthorsAparicio, PL, Issa, EB, DiCarlo, JJ
Conference NameSociety for Neuroscience
Date Published2010
Conference LocationSan Diego, CA, USA

Functional MRI in the monkey has revealed at least six patches of tissue in the ventral visual cortical stream that prefer images of faces over non-face objects. Neuronal recordings targeted to the largest patch (“middle face patch”), hypothesized to be homologous to the human FFA, revealed that nearly all visually driven cells were at least 2x more responsive to faces over non-face images. These results support a view in which this region consists entirely of “face cells” involved exclusively in face processing. However, previous work could not distinguish between a single large cluster, several clusters, or a local, graded enrichment of “face cells”. More deeply, while “face cells” are typically defined as those with a preference for face stimuli on average, it is not clear what model best explains their response properties. A strong form of the “face cell” hypothesis is a semantic response model in which each cell responds only to images that an observer would distinguish as a face. An alternative class of models places these cells at an intermediate processing stage along the ventral visual feature hierarchy in which some face selectivity has been achieved, but is the result of neuronal tuning for complex visual features found in both face and some non-face objects. To explore these questions, we used a custom built high resolution x-ray imaging system to map the spiking selectivity at hundreds of sites in and around the fMRI-identified macaque middle face patch using a large class of face and non-face stimuli. Our current data show a single region of face selectivity, approximately 2-3 mm in diameter, in rough correspondence with the fMRI-identified face patch. We confirmed previous results in that ~90% of neuronal sites in this region are “face selective” when evaluated with simple face vs. object response criteria. However, we found that none of these sites maintained a strict semantic designation by responding significantly more to all face images than to any non face image. Moreover, at least ~60% of sites show a significantly greater response to at least one non-face image, such as images of fruits, cars, and toys, over at least one face image, such as caricatures of human and monkey faces. These results argue against a strong semantic response model of the middle “face patch.” Instead, the most parsimonious interpretation of these data is that neurons in the fMRI-identified middle “face patch” are located at an intermediate stage of visual processing: although they may be tapped at higher areas to build a single-neuron explicit, semantic level representation of faces, their role may be to represent intermediate visual features that are also found in non-face objects.