fMRI studies have characterized several "object category specific" areas in human inferotemporal (IT) cortex. However, these areas also show robust responses to non-preferred categories (Spiridon 2006). In contrast, single unit recordings in monkeys indicate almost pure selectivity for object categories (Tsao 2006). In standard fMRI experiments, data are typically acquired at a spatial resolution of 3x3x3 mm³, causing averaging of responses across functionally distinct units. With special techniques it is possible to bring fMRI resolution to the scale of cortical columns (Cheng 2001). Here, we apply high-resolution fMRI to assess selectivity profiles and spatial distribution of face responsive areas in human IT cortex.

A 4 Tesla MRI system, together with a 16 channel receive and 4 channel transmit array coil covering the posterior half of the head, was used to acquire gradient-echo EPI images (volume TR 4.05 s, TE 0.015 s, 3 mm slice thickness). In-plane resolution was either 3x3 mm² (low-resolution; with a 4-segment acquisition) or 0.75x0.75 mm² (high-resolution; with a 16-segment acquisition). Head motion was minimized with a bite bar and a 2D motion correction procedure.

Face-selective areas were first identified in occipital and temporal lobes with low resolution fMRI while the subjects viewed images of faces, buildings, bodies, objects, and noisy patterns (block design, 1 image/s, 1-back task). These areas were then targeted in high-resolution studies with four slices aligned parallel to the local cortical surface, yielding optimal resolution along the cortical mantle.

High-resolution fMRI revealed robust visual and face specific responses. Overall, high- and low-resolution data were in good agreement, but high-resolution data exhibited a clear sharpening in specificity: face-selective areas appeared strictly confined to gray matter, and more focal and more sharply tuned than suggested by low-resolution data.

Our study demonstrates that columnar resolution imaging of human IT cortex is feasible and can provide detailed information on how various categories of objects are represented in those areas. It can also help in bridging knowledge from electrophysiological and fMRI studies.