Shape Interactions in Macaque Inferior Temporal Neurons

Abstract

Missal et al. observed that the responses of inferior temporal (IT) neurons to a shape were reduced markedly when this shape partially overlapped a larger second shape, suggesting that shape interactions determine IT responses. In the present study, we compared the responses of IT neurons with combinations of two shapes which did or did not overlap and studied the effect of the relative and absolute positions of the two shapes. In a first test, a preferred shape (figure) was presented at the fixation point while a second, nonpreferred, shape was displayed either in the background of the figure (overlap) or at one of four peripheral positions (nonoverlap). Controls consisted of presentations of either shape in isolation at each of the five positions. The stimuli were presented during a fixation task. The responses to these combinations of two shapes were, on average, reduced compared with those elicited by the preferred shape presented in isolation. This suppression occurred whether or not the two shapes overlapped, but the degree of suppression in the overlap and nonoverlap conditions did not correlate. These interactions were stronger when the interacting stimulus was located in the contralateral compared with the ipsilateral hemifield. The position of the interacting stimulus within a hemifield significantly affected the suppression associated with combined shapes in some neurons. The strength of the interactions of the two nonoverlapping shapes depended on the shape of the interacting stimulus in half of the neurons. In a second test, the preferred shape and interacting stimulus could appear either at the fixation point or at one eccentric position. Here we found that the suppression was, on average, strongest when the interacting stimulus, rather than the preferred shape, was presented at the fixation position. Also, in 40% of the neurons, the response reduction was similar in overlap and nonoverlap conditions if effects of stimulus position were taken into account. In both tests, we also measured the responses to combinations of a nonpreferred shape and the interacting stimulus and showed that the response to a combination of two nonpreferred shapes was, in general, smaller than the response to a combination of the preferred and nonpreferred shape. Overall the results indicate that stimulus interactions in the receptive fields of IT neurons can be position and shape selective; this can contribute to the coding for the relationships between object parts.

Multiple Object Response Normalization in Monkey Inferotemporal Cortex

“…This trend is consistent with the reduction of response suppression produced by the less effective shape in a pair as a function of its distance from the more effective shape (Missal et al, 1999).…”

“…However, this idealized notion of IT is inconsistent with available data showing that IT responses are altered in cluttered scenes (Sheinberg and Logothetis, 2001;Rolls et al, 2003), and responses to object pairs are typically weaker than responses to isolated, preferred objects (Sato, 1989;Miller et al, 1993;Rolls and Tovee, 1995;Chelazzi et al, 1998;Missal et al, 1999). Moreover, although some contemporary models of the ventral stream use MAX operations (Riesenhuber and Poggio, 1999a), even these models do not predict complete clutter tolerance in IT (Riesenhuber and Poggio, 1999b).…”

“…However, because IT neuronal RFs are not all centered at the same retinal position, have a broad range of sizes (Op De Beeck and Vogels, 2000), and can often be small relative to the separation of our objects (Op De Beeck and Vogels, 2000;DiCarlo and Maunsell, 2003), we used these RF variations to ask whether there was any relationship between the averaging behavior described above and position in the RF. In particular, we might not expect the response to a pair of objects to be the average of the responses to the constituent objects if one of those objects was presented very far outside the RF (Missal et al, 1999), but we wondered whether we might detect some breakdown in averaging behavior when one of the objects was near the edge of the RF.…”

“…In particular, although several IT studies have touched on the issue of IT responses in clutter (Miller et al, 1993;Rolls and Tovee, 1995;Chelazzi et al, 1998;Missal et al, 1999;Sheinberg and Logothetis, 2001;Rolls et al, 2003), no study has systematically tested the relationship between responses to object pairs and responses to constituent objects (but see Missal et al, 1999), and there has been no attempt to understand how IT responses to multiple objects depend on object shape similarity.…”

A paradoxical improvement of misreaching in optic ataxia: new evidence for two separate neural systems for visual localization

“…Given the present evidence that it can function relatively well in a patient with extensive bilateral parietal damage, this second system may receive information about spatial relationships through occipito-temporal visual areas. In support of this idea, the relative coding of stimulus location within a visual array has recently been physiologically demonstrated in neurons in the monkey's temporal neocortex (Baker et al 1999;Missal et al 1999).…”