Author:

Abstract:

The visual system is thought to be composed of two parallel pathways, a dorsal pathway that is concerned with motion and position, and a ventral pathway concerned with object representation. Recent evidence suggests that many regions in the dorsal pathway also encode 3-D depth cues, such as 3-D structure-from-motion (SFM). Interestingly, SFM cues can also be used to define any object and humans can accurately recognize objects defined by SFM cues, but only if the ventral stream is intact--thus the recognition of SFM objects requires dorsal-ventral integration. We were interested to know whether the inferior temporal gyrus, a region in the macaque cortex thought to be critical for object recognition, has an enhanced response to whole objects defined by 3-D SFM as compared to depth-motion controls. During fMRI imaging with MION, awake-fixating monkeys passively viewed videos of SFM defined objects or their depth-motion matched controls. The stimuli consisted of white dots on a black background that defined a 3-D shape rotating left to right and then up and down. The 2-D projected dot density was calculated on each frame and regions with increased or decreased dot density were adjusted so as to remove 2-D projected dot density as a cue of depth or shape. The stimuli were only visible when in motion--pausing the video caused the percept to cease. Two factors were manipulated: object category (faces vs. hands) and surface coherence (coherent vs. randomized depth). The depth controls were made of the same original objects, but the depth ordering of the surface segments was randomized such that the rotating pattern did not result in any discernible object, but maintained similar motion and depth profiles. We find that 3-D SFM defined monkey faces and human hands do indeed activate a number of regions of the macaque inferior temporal gyrus as compared to the control stimuli. These results suggest that the ventral object recognition regions are sensitive to information derived from dorsal stream regions, such as surfaces defined by motion via SFM. Furthermore, the results suggest that the ventral stream representations are likely invariant to depth-cues.