Building of object view invariance in a newly-discovered network in inferior temporal cortex

[摘要] Object recognition in primates is mediated by hierarchical, multi-stage processing of visual information within occipital and inferior temporal (IT) cortex. It is known that IT contains several networks that process specific categories or stimulus dimensions. Furthermore, at least in the case of face processing network, the nodes appear to be organized hierarchically, e.g., neurons in the middle faces patches are tuned for specific facial views, while those in the most anterior patch are tuned for the identity of faces in a view-invariant way. However, there remains a fair amount of IT cortex that doesn't belong to any known network, raising the question: are there any new, undiscovered networks not yet accounted for by existing functional parcellation studies? If so, what are these networks processing and how are they organized? To address this question, we exploited the technique of electrical microstimulation combined with simultaneous functional magnetic resonance imaging. Electrical microstimulation of a region of macaque IT cortex not belonging to any known network produced strong activation in three patches that also didn't overlap with any known networks. We targeted single-unit recordings to these three patches, while monkeys passively viewed an image set consisting of 51 objects with 24 views for each object; the objects included faces, animals, houses, vegifruit, vehicle, and man-made objects. Average responses across neurons from the three patches revealed high similarity in object preferences between the patches, further confirming these patches belong to a common network; for example, all three patches showed the smallest response to faces. Representational similarity analysis on population single-unit responses in each of the three patches revealed the most view-invariant representation in the most anterior patch, and the least view-invariant representation in the posterior patch, suggesting that, analogous to the face patch network, view-invariant object representation is built up hierarchically within this new network.

[学科分类] 眼科学