Subcortical Discrimination of Unperceived Objects ... - Semantic Scholar

Neuron, Vol. 42, 163–172, April 8, 2004, Copyright 2004 by Cell Press

Subcortical Discrimination of Unperceived Objects during Binocular Rivalry Brian N. Pasley,1,2 Linda C. Mayes,1 and Robert T. Schultz1,* 1 Developmental Neuroimaging Laboratory Child Study Center Yale University New Haven, Connecticut 06520 2 Helen Wills Neuroscience Institute University of California, Berkeley Berkeley, California 94720

Summary Rapid identification of behaviorally relevant objects is important for survival. In humans, the neural computations for visually discriminating complex objects involve inferior temporal cortex (IT). However, less detailed but faster form processing may also occur in a phylogenetically older subcortical visual system that terminates in the amygdala. We used binocular rivalry to present stimuli without conscious awareness, thereby eliminating the IT object representation and isolating subcortical visual input to the amygdala. Functional magnetic resonance imaging revealed significant brain activation in the left amygdala but not in object-selective IT in response to unperceived fearful faces compared to unperceived nonface objects. These findings indicate that, for certain behaviorally relevant stimuli, a high-level cortical representation in IT is not required for object discrimination in the amygdala. Introduction When presented with a variety of visual objects in the environment, the brain must be able to quickly and reliably detect those that are behaviorally relevant (e.g., those that signal potential dangers). The amygdala is believed to be a critical component of this perceptual process and has been implicated in many forms of affective processing, especially fear-related (LeDoux, 1996). There are both cortical and subcortical amygdalar inputs (LeDoux, 1996), but the relative and perhaps unique contribution of each to visually driven, affective responses in the amygdala is not yet well understood. In nonhuman primates, the amygdala receives input from IT (Stefanacci and Amaral, 2002), the last purely visual area in the ventral object recognition pathway (Gallant, 2000), but not from earlier cortical visual areas (Iwai and Yukie, 1987; Webster et al., 1991). Although input from IT may provide the amygdala with a hierarchically processed and detailed object representation, this processing strategy necessarily sacrifices speed for enhanced detail. Visual responses in monkey IT typically occur between 100 and 140 milliseconds (ms) after stimulus presentation (Rolls, 1999) (in humans, IT latencies are a bit longer, i.e., 150–200 ms; Allison et al., 1999). *Correspondence: [email protected]

When the perceived object is behaviorally relevant and requires a rapid adaptive response, this extended processing time may put the animal at a survival disadvantage. However, amygdala neurons selective for complex objects have been found in monkeys with response latencies as short as 60 ms, which is roughly twice as fast as IT response latencies, indicating the presence of a more rapid alternate visual route in primates (Nakamura et al., 1992). Several sources of data, including anatomical (Linke et al., 1999; Romanski et al., 1997), lesion (Miller et al., 1980), and functional brain imaging (Morris et al., 1999, 2001; Vuilleumier et al., 2003) studies, have been used to identify a series of subcortical visual structures that plausibly comprise a subcortical pathway terminating in the amygdala. This pathway, proceeding from the retina to the superior colliculus (SC) to posterior nuclei of the thalamus and onto the amygdala, bypasses detailed cortical processing and is thought to provide the amygdala with lower-resolution but more rapidly processed visual input (LeDoux, 1996). Some accounts emphasize the bias of this system for stimuli that are informative about potential dangers (e.g., fear-relevant stimuli) (LeDoux, 1996; Morris et al., 1999, 2001). Although the anatomical connections underlying this subcortical visual system clearly exist in nonhuman primates (Benevento and Fallon, 1975; Jones and Burton, 1976; Schiller and Malpeli, 1977; Stepniewska et al., 2000), there is not yet completely convincing empirical data to argue for the functional significance of this pathway. One piece of evidence for its functional significance comes from the study of “blindsight.” In blindsight, a lesion in primary visual cortex (V1) prevents conscious vision in the corresponding portion of the visual field and is believed to disrupt the ventral object recognition pathway extending from V1 through IT. Nevertheless, blindsight patients exhibit residual abilities to detect and localize visual stimuli (Weiskrantz, 1997), suggesting the activation of an alternative visual pathway. In a functional magnetic resonance imaging (fMRI) study, Morris and colleagues (2001) demonstrated amygdala activation to emotionally expressive faces in a blindsight patient who did not consciously perceive the stimuli. However, because the subcortical visual stream is believed to possess relatively primitive pattern vision mechanisms, several investigators have argued that highly processed input from IT may be required to support visual discriminatory responses in the amygdala (Pessoa et al., 2002a, 2002b; Rolls, 1999, 2000). In fact, in a study by Goebel and colleagues (2001), this same blindsight patient exhibited increased activity in IT when presented with images of complex objects in his blind visual field, suggesting that IT inputs to the amygdala could have driven the observed amygdala response. Visual backward masking is another phenomenon that has been used to disrupt processing in IT (Morris et al., 1998; Whalen et al., 1998) in order to isolate the subcortical pathway. These studies briefly presented target images of faces with emotional or neutral expressions (ⱕ30 ms) followed immediately by a sustained masking stimulus that rendered the target image invisi-

Neuron 164

ble. In macaques, the mask is known to reduce the amount of information regarding the target image by interrupting or overriding the firing of neurons in IT (Kovacs et al., 1995; Rolls et al., 1994). Importantly, the amygdala receives its major visual input from IT, but not from earlier visual areas (Iwai and Yukie, 1987; Stefanacci and Amaral, 2000; Webster et al., 1991). Because masking degrades the information available in this late stage of cortical visual processing, it has been argued that the differential amygdala response is derived exclusively from subcortical visual input (Morris et al., 1999, 2001). Interpretation of the origin of this visual input to the amygdala is difficult, however, since neurons in IT retain substantial information about the target image despite the masking procedure (Rolls et al., 1999). In fact, under experimental conditions where the object is not “seen,” increased target-related neural activity in IT has been observed consistently across a variety of functional imaging (Bar and Biederman, 1999; Dehaene et al., 2001; Grill-Spector et al., 1999) and single-unit (Kovacs et al., 1995; Rolls and Tovee, 1994) experiments that have employed backward masking. Furthermore, IT activity related to the masked target is subject to repetition suppression, a phenomenon observed in positron emission tomography (PET) and fMRI studies in which the hemodynamic response is reduced for repeated compared to novel stimuli (Dehaene et al., 2001; Henson and Rugg, 2003). Such selective modification is evidence for the sustained presence of a neural representation that is sufficiently intact to differentiate the repeated from the unrepeated stimulus. A related and consistent finding concerns behavioral priming of the masked stimulus (Bar and Biederman, 1998; Buckner et al., 1998; Koutstaal et al., 2001). Priming is a phenomenon that occurs when a prior presentation of a stimulus improves or “primes” performance during subsequent presentations. Like repetition suppression, behavioral priming appears to depend on the presence of a functional neural representation formed late in the ventral visual stream (Bar and Biederman, 1998; Buckner et al., 1998; Koutstaal et al., 2001), suggesting that the high-level representation of the masked stimulus persists despite the masking procedure. Thus, it remains unclear whether or not the discrimination observed in the amygdala during masking and blindsight occurs as a consequence of a degraded but intact neural representation in IT or, alternatively, originates exclusively from a subcortical source of input. Consequently, the ability of the subcortical visual pathway to discriminate certain behaviorally relevant stimuli in isolation from the IT object representation remains controversial (Pessoa et al., 2002a, 2002b; Rolls, 1999, 2000). To address this controversy, we used fMRI to measure brain activity during a binocular rivalry paradigm as a way to probe the ability of the amygdala to discriminate complex objects in isolation from the IT neural representation. During binocular rivalry, each eye is presented with a different, incompatible image. Rather than experience fusion of these two distinct images, the observer experiences alternating perceptual dominance of one image or the other. Thus, while one image is perceptually dominant, the other is completely suppressed, or unperceived. The neural mechanisms underlying rivalry are poorly understood and are a subject of current debate

(Blake and Logothetis, 2002). Some evidence in humans, however, suggests that rivalry may be mediated by a process in V1 referred to as interocular competition (Tong and Engel, 2001). Under this particular model, lateral inhibition among monocular neurons in V1 is believed to suppress input from one eye, effectively blocking further visual processing for the suppressed stimulus while the dominant monocular stimulus continues on to higher-level visual areas and eventually reaches visual awareness (Blake, 1989; Tong, 2001). Although there is some evidence that neurons in V1 and V2 still respond to suppressed stimuli (Leopold and Logothetis, 1996), data from both monkey neurophysiology (Sheinberg and Logothetis, 1997) and human fMRI (Tong et al., 1998) converge to suggest that suppressed visual information is blocked from reaching IT where neural responses reflect the perceptually dominant and not the suppressed stimulus. These physiological data agree well with behavioral studies exploring the effects of rivalry suppression on behavioral priming. These studies demonstrate that rivalry suppression, unlike backward masking, abolishes behavioral priming for suppressed words (Zimba and Blake, 1983) and pictorial stimuli (Cave et al., 1998), providing additional support for the notion that suppression blocks visual information from reaching the late stages of the ventral visual pathway. Perceptual suppression during binocular rivalry therefore appears well suited to test the subcortical visual pathway to the amygdala independently of input from IT. Based on this evidence, we used a binocular rivalry paradigm in which the image of either a face with a fearful expression or a chair with no apparent emotional value was presented under complete perceptual suppression in order to observe differential neural responses that occur in isolation from cortical input. While chairs are neutral stimuli with no obvious emotional significance, fMRI studies have found with notable consistency that fearful faces activate the amygdala, particularly on the left (Calder et al., 2001). Thus, the current study was designed to interrogate subcortical visual regions for increased neural activity to suppressed fearful faces as compared to suppressed neutral chairs. We reasoned that a significant differential response in the amygdala, but not in object-selective IT, would provide strong evidence that rudimentary discrimination of certain complex visual patterns does not require a high-level cortical representation but instead may be mediated by a subcortical visual pathway that can operate independently and in parallel to cortical visual processing. Results In binocular rivalry, perceptual dominance is promoted by “stimulus strength,” as defined by characteristics such as high luminance and high contrast, transient motion, and complexity (Blake, 2001). We therefore presented to one eye a complex image of a house that moved back and forth sharply in order to constrain conscious perception to this image. During the movement of the house, a target image of a fearful face or a neutral chair gradually faded into view of the other eye, remaining in view for approximately 1.5 s before fading

Subcortical Processing of Unperceived Stimuli 165

Figure 1. Binocular Rivalry Stimulus Presentation Procedure for a Suppressed Face Trial Participants viewed blocks of four consecutive trials lasting 10 s and were instructed to indicate by button press if at any point they perceived anything besides a house, no matter how fleeting, in order to verify the successful suppression of the target image. Across all participants, unsuccessful rivalry suppression trials occurred in 2% of these blocks, which were subsequently excluded from further analysis.

out again. This presentation scheme (Figure 1) maintained the perceptual suppression of the target image during the entire trial. In addition to presentations of suppressed chairs and suppressed fearful faces, we also included a third condition in which a moving house was presented to both eyes, in order to compare responses to this nonrivalry condition. These three conditions were indistinguishable to the participant, appearing as a moving house in every case. To assess the presence of discrimination of suppressed objects, we compared activity during suppressed face presentations to activity during suppressed chair presentations within several a priori regions of interest (ROIs). These ROIs included three nodes of the subcortical visual pathway—the SC, the thalamus, and the amygdala—as well as face- and chair-selective IT cortex. To identify these category-selective areas of IT (Epstein and Kanwisher, 1998; Ishai et al., 1999; Kanwisher et al., 1997), we conducted an independent object localizer experiment during the same scan session, in which participants viewed images of fearful and neutral faces or chairs. This allowed us to functionally define the areas of IT involved in encoding high-level face and chair neural representations. These face and chair ROIs overlapped with those reported by other groups (Ishai et al., 1999) and were localized in the lateral and medial fusiform gyrus for faces and chairs, respectively. After transformation of all 12 individual data sets into standard stereotaxic space (Talairach and Tournoux, 1988), a voxel-wise random effects analysis of the group data yielded significant left amygdala activation for a suppressed face versus suppressed chair contrast [t(11) ⫽ 4.063, p ⬍ 0.005, corrected for 5 multiple ROI comparisons; Figure 2]. In addition, inspection of the individual data sets showed that this effect was present in 9 of the 12 participants at a threshold of p ⬍ 0.05, uncorrected. Left-lateralized activation is consistent with previous studies finding a bias toward the left amygdala for fear-related stimuli (Calder et al., 2001). Importantly, in the group data, the same suppressed face versus suppressed chair contrast produced no significant differences in face and chair ROIs within IT cortex (tested at a less stringent threshold of p ⬍ 0.01 to strengthen the claim of a null result), suggesting that visual information regarding the suppressed objects did

not reach these higher-level areas of visual cortex. However, because the exact location of IT object-selective areas can vary somewhat between individuals and may not be accurately reflected in a group average, we also functionally defined face and chair ROIs on each individual’s data. This provided a more rigorous test to determine if the discrimination in the amygdala was paralleled by IT activation that may have in turn driven the observed amygdala activation. Mean signal changes within these ROIs during the rivalry experiment were used to test for condition-specific activations of face- and chair-selective IT. An ANOVA contrasting the peak mean signal change in face- and chair-selective IT cortex revealed no significant main effects for suppressed image type [suppressed face/chair, F(1,44) ⫽ 0.70, p ⫽ .41] or ROI [F(1,44) ⫽ 0.85, p ⫽ .36], nor was there a significant image type by ROI interaction [F(1,44) ⫽ 0.71, p ⫽ .41]. Thus, as shown in Figures 3C and 3D, during rivalrous presentations, the average response function in IT follows the dominant house stimulus, with no significant differences between the three conditions [repeated measures ANOVA, face-selective IT: F(2,20) ⫽ 0.759, p ⫽ .48; chair-selective IT: F(2,22) ⫽ 1.48, p ⫽ .25]. However, when the amygdala ROI is included in this analysis, there are significant effects of image type [F(1,66) ⫽ 10.49, p ⬍ .002], ROI [F(2,66) ⫽ 3.85, p ⬍ .03], and the image type by ROI interaction [F(2,66) ⫽ 5.32, p ⬍ .008]. This result is accounted for by the significantly greater suppressed face activation of the left amygdala, compared to suppressed chair, without any significant effect of image type on the cortical ROIs. Although we did not observe significant activation in any areas outside of our a priori ROIs (tested at p ⬍ .05, corrected), our slice selection primarily covered only the visual cortex and midbrain. We were therefore unable to assess the possibility that other nonvisual cortical regions may have exhibited differential responses to suppressed objects, conceivably providing the amygdala with visual input through an indirect cortical route. However, prior studies of rivalry have found that activity in nonvisual areas (e.g., parietal and prefrontal cortex) is associated with conscious perception rather than the unperceived stimulus (Lumer et al., 1998; Lumer and Rees, 1999), suggesting that suppressed visual information fails to reach these later processing stages. Simi-

Neuron 166

Figure 2. Differential Amygdala Activation for Perceptually Suppressed Fearful Faces and Suppressed Chairs (A) Left amygdala fMRI activation for suppressed faces contrasted with suppressed chairs across all 12 participants (displayed at a threshold of p ⫽ .01, uncorrected). Talairach coordinates of the peak voxel are x ⫽ ⫺18, y ⫽ ⫺2, z ⫽ ⫺21 (p ⬍ .005, corrected for five multiple ROI comparisons). (B) fMRI responses (in standard deviation units ⫾ SEM) averaged across the nine participants who showed an object-selective effect in the left amygdala. Time courses were taken from the five most significant voxels in the left amygdala surviving a threshold of p ⬍ .05, uncorrected. The gray box represents the block of stimulus presentation (the fMRI signal is characterized by a hemodynamic lag of about 4–6 s to peak).

larly, our fMRI data strongly indicate that rivalry suppression blocked face- and chair-related visual information from reaching IT and, presumably, all other downstream nonvisual regions. In monkeys, IT (including areas TE and TEO) is the only area in the cortical ventral visual pathway that sends efferent projections to the amygdala (Iwai and Yukie, 1987; Stefanacci and Amaral, 2000; Webster et al., 1991), implying that this is a critical node for providing cortical visual input to the amygdala. To identify alternative subcortical brain regions that might have provided visual input to the amygdala, we conducted a psychophysiological interaction (PPI) analysis (Friston et al., 1997). The PPI analysis locates other

Figure 3. fMRI Responses in Face- and Chair-Selective IT during Normal Viewing and Rivalry Suppression Face- (A) and chair- (B) selective IT during the localizer (normal viewing) show stronger responses (in standard deviation units ⫾ SEM) for the preferred category. Responses in the same face- (C) and chair- (D) selective regions during rivalry did not differentiate suppressed stimuli from a nonrivalry condition. Note that all three conditions during rivalry shared the same percept, indicating that neural responses here followed the perceptually dominant stimulus (a moving house). For each ROI, time courses were extracted from the five most significant pixels in both left and right IT for each individual, normalized, and then averaged. The gray box represents the block of stimulus presentation.

areas that exhibit condition-specific covariation with a given reference ROI (i.e., the amygdala in the present analysis). In this way, functional connections between different brain regions can be assessed in a conditionspecific manner. In partial accordance with our a priori hypothesis of a subcortical visual pathway, we found that the suppressed image type (i.e., a face or a chair) modulated the correlation between activity in the left amygdala and the left SC [t(6) ⫽ 4.31, p ⬍ .01, corrected; Figures 4A and 4B]. A significant positive correlation between these two regions was evident during suppressed face presentations [r(249) ⫽ .194, p ⬍ .01, corrected for three condition comparisons], while no significant correlations were found during the suppressed chair [r(249) ⫽ ⫺.021, p ⬎ .05, corrected] or the dominant house [r(248) ⫽ .123, p ⬎ .05, corrected] presentations. This pattern of condition-specific covariation suggests that these two regions are involved in a functional circuit that is sensitive to the suppressed image type. We found no other regions that exhibited this significant conditionspecific covariation. Based on this finding, we also performed a second PPI analysis using the SC as the source ROI. This second analysis revealed a single significant area located in a dorsal posterior lateral region of the thalamus [t(5) ⫽ 8.04, p ⬍ .005, corrected; Figures 4C and 4D]. This region was positively correlated with the SC during suppressed face presentations [r(213) ⫽ .163, p ⫽ .05, corrected] but not during the suppressed chair [r(214) ⫽ ⫺.069, p ⬎ .05, corrected] or the dominant house [r(213) ⫽ .14, p ⬎ .05, corrected] presentations. The observed pattern of connectivity is consistent with the hypothesis that without input from IT, subcortical visual input via a pathway involving the SC and visual thalamic nuclei is sufficient to drive a differential amygdala response to two complex objects. Discussion We found that neural activity in the left amygdala discriminated two complex object classes that were rendered invisible by perceptual suppression during binocular rivalry. In contrast, neural responses in objectselective IT, the sole visual area in the ventral stream


Figure 4. ROI Connectivity (A) A region of the left SC (⫺4, ⫺29, ⫺3) showed a significant condition-specific covariation with the left amygdala (p ⬍ .01, corrected). (B) The scatterplot displays the correlation between this left SC ROI and the left amygdala ROI, coded by condition. The significant PPI analyses described in the text are a function of a significant difference between the suppressed face [r(249) ⫽ .194, p ⬍ .01, corrected for three condition comparisons] and the suppressed chair [r(249) ⫽ ⫺.021, p ⬎ .05, corrected] correlations . The correlation during house presentations was also nonsignificant [not displayed; r(248) ⫽ .123, p ⬎ .05, corrected]. (C) A region of the left dorsal posterior lateral thalamus (⫺10, ⫺14, 12) showed a significant condition-specific covariation with the SC (p ⬍ .005, corrected). (D) The SC and thalamic region correlated positively during suppressed face presentations [r(213) ⫽ .163, p ⫽ .05, corrected] but were not significantly correlated during suppressed chair [r(214) ⫽ ⫺.069, p ⬎ .05, corrected] or house [not displayed; r(213) ⫽ .14, p ⬎ .05, corrected] presentations. The fMRI images are displayed at a threshold of p ⫽ .01, uncorrected.

that projects to the amygdala, were identical between rivalrous conditions (suppressed face/chair) and a nonrivalrous control condition (dominant house). This suggests that IT cortex did not receive suppressed visual information and thus could not contribute input to the amygdala to trigger its differential response to suppressed objects. In addition, we found significant correlations between different subcortical visual structures during suppressed face presentations (but not during suppressed chair presentations): (1) the left amygdala correlated positively with the SC and (2) the SC correlated positively with a dorsal posterior lateral region of the thalamus. As shown in Figures 4A and 4C, these correlations were highly specific to these structures. Our results support prior studies suggesting that rivalry suppression blocks suppressed visual signals sometime prior to IT in the ventral processing stream (Tong and Engel, 2001; Tong et al., 1998). On the other hand, our data indicate that information regarding the suppressed image successfully reaches the amygdala, apparently through a subcortical pathway. One possible caveat to this conclusion is that fMRI may lack the sensitivity to detect weak signals in IT during rivalry suppression and that these weak signals may yet provide the amygdala with cortical visual input. However, this possibility seems unlikely in view of past studies demonstrating that fMRI has sufficient sensitivity to detect weak extrastriate signals related to unperceived stimuli during backward masking (Bar and Biederman, 1998; Dehaene et al., 2001). Moreover, a related study that employed “flash suppression” (Kreiman et al., 2002) provides more direct evidence that perceptual suppression blocks the transmission of cortical visual input to the amygdala. Flash suppression is closely related to binocular rivalry and occurs when the perception of a baseline image,

viewed monocularly, is disrupted by a new “flashed” stimulus to the other eye. Kreiman and colleagues, recording from single cells in the amygdalae of patients undergoing brain surgery, identified several amygdala neurons that responded selectively to complex objects at latencies of around 200 ms, suggesting transmission of visual input from IT (where latencies are typically ⵑ170 ms; Allison et al., 1999). However, when these preferred images were perceptually suppressed by a flashed stimulus (that was ineffective in stimulating amygdala activity), the increased firing rate of the amygdala neurons returned immediately to the lower baseline rate. This finding indicates that perceptual suppression of the preferred image blocked the transmission of suppressed visual input to the amygdala via the cortical IT pathway. While Kreiman et al.’s study clearly demonstrates the blocking effect of suppression on cortical visual input, it is unclear why the suppressed stimuli did not continue to engage the amygdala through the subcortical pathway described in our current results. There are several possibilities that may explain this apparent discrepancy. First, the single neurons in this study were chosen for analysis based on their strong selectivity for different complex images or object categories. However, this subsample of amygdala neurons may not overlap with a less selective population of neurons driven by relatively low-resolution subcortical visual inputs. It is therefore possible that amygdala neurons responsive to subcortical inputs may be a distinct population and, consequently, were excluded from Kreiman et al.’s analysis because they did not meet selectivity requirements. In addition, this study investigated only a very small sample of amygdala neurons (25 selective neurons were analyzed, many of which were in the right hemisphere,

Neuron 168

not the left as highlighted by the current study). The fMRI blood oxygenation level-dependent (BOLD) signal, on the other hand, reflects the simultaneous neural activity of large, distributed populations of neurons (typically around 107 per voxel). If amygdala neurons responsive to subcortical inputs are fewer than those responsive to cortical inputs, detection of these relatively uncommon sets of neurons may be easier with fMRI than with single unit recording. Another possible difference between the current study and that of Kreiman et al. concerns the neurophysiological basis of the BOLD signal commonly measured in fMRI. Currently under debate is whether the BOLD signal is more closely linked to the spiking output of neurons or to energy requirements associated with cumulative activity at the synapses (Logothetis, 2003). The amygdala activation that we observed may therefore have been related more to increased input to the amygdala rather than to actual neuronal spikes. This interpretation would be consistent with the absence of single neuron responses observed by Kreiman et al. However, even if these fMRI results reflect an input rather than an output process, either possibility entails the existence of an intact visual pathway that successfully transmits information to the amygdala during perceptual suppression. Given the convergent evidence for cortical blocking during suppression, the increased amygdalar response to the suppressed face evidently resulted from visual input whose origin was other than IT. This conclusion is also supported by our functional connectivity analyses. The observed pattern of connectivity coincides with the hypothesis of a feedforward subcortical visual pathway to the amygdala involving the SC and visual thalamus (LeDoux, 1996; Morris et al., 1999, 2001). However, because this connectivity analysis is based on correlation, we cannot conclusively determine the directionality or the full extent of this pathway. For example, it is possible that the correlation between the amygdala and SC is representative of a shared source of cortical input rather than a feedforward subcortical pathway. In this sense, responses in these two subcortical regions may be modulated by a third region (possibly V1 or V2, which have been found to process aspects of suppressed stimuli; Leopold and Logothetis, 1996). However, these cortical areas appear to have no direct projections to the amygdala (Iwai and Yukie, 1987; Stefanacci and Amaral, 2000; Webster et al., 1991) and are not known to support the discrimination of complex objects. The SC, on the other hand, is an ancient midbrain structure that serves as the primary object processing structure in avians and amphibians and receives direct visual input from the retina in primates (Sewards and Sewards, 2002). It is therefore well positioned to receive direct visual information before incongruent signals during rivalry have been altered by binocular convergence. In addition, there is evidence that the ocular dominance of neurons in the visually responsive superficial layer of the SC is stronger for the contralateral eye and that this dominance becomes even stronger after a visual cortex lesion (Rosa and Schmid, 1994). Compared to the robustness of suppression in binocular extrastriate cortex, suppressive mechanisms in the SC may be weaker for neurons that receive input predominantly from only one eye. Consequently, ocular

dominance in SC neurons may preserve monocular visual input, permitting subsequent subcortical processing of suppressed stimuli. Based on this physiological evidence and our connectivity analysis, it therefore seems reasonable that the SC played at least some role in the form processing that sustained the differential amygdala response to suppressed complex objects. In primates, the SC is usually implicated in visual orienting responses and directional encoding. Although the visual cortex has superseded it as the primary form processing center in the primate brain, the SC still possesses the ability to process simple forms, but only within a narrow range of spatial frequencies (Lomber, 2002; Miller et al., 1980). A critical problem for the hypothesis of a subcortical pathway is to determine how primitive pattern vision in the SC might successfully code and propagate complex form information to other subcortical structures. Several aspects of our results provide indications of what visual information is passed to the amygdala through this putative subcortical system. One potentially important observation is that the differential amygdala activity for suppressed objects was due to both increased signal for faces and decreased signal for chairs. This bidirectional differential response to suppressed objects would seem to indicate that information about both the face and the chair succeeded in reaching the amygdala and consequently modulated activity there. In view of this apparent nonselectivity, the SC is probably not involved in any type of “gating” mechanism (i.e., all or none output) whereby irrelevant stimuli are completely filtered out from further processing. Because we did not observe a differential fMRI response in the SC to suppressed objects, we postulate that the processing of each suppressed object type generated similar neuroenergetic requirements within this structure. In this sense, the SC may act as a general filter for form information, processing each pattern without regard to behavioral relevance and passing on a similar range of visual information for both suppressed object types. If in fact the SC is engaged in such a filtering process, then it is crucial to consider the precise nature and level of detail of information that ultimately reaches the amygdala through this filter. In the absence of visual cortex, the SC is sensitive to the low but not high spatial frequency components of patterns (Lomber, 2002; Miller et al., 1980; Vuilleumier et al., 2003). While high spatial frequencies encode individual facial features and fine details, low spatial frequencies encode the global configural structure of the face (Schyns and Oliva, 1999). Low spatial frequency information about the coarse structure of a face is particularly important for face perception, and it is thought that an efficient face detection mechanism would rely more strongly upon low rather than high spatial frequencies (Dailey and Cottrell, 1999). Thus, the SC appears to encode a range of spatial information that is particularly valuable for detecting a complex face stimulus. This view is consistent with the observation that newborn infants exhibit longer looking times for face-like patterns than for other patterns, even in the first minutes of life (Goren et al., 1975; Simion et al., 1998). At birth, the visual cortex is functionally immature (Kraemer and Sjostrom, 1998), and it is thought that this visual preference is instead supported


by a subcortical mechanism involving the SC (Morton and Johnson, 1991). Thus, the positive correlation we observed between the amygdala and the SC during presentations of suppressed faces may relate to a subcortical pattern vision mechanism that has been configured, possibly via evolutionary pressure, to preserve the most essential spatial information needed to support rudimentary detection of face-like patterns, although the range of preferred patterns is likely to be quite crude. While the SC appears to encode a range of visual information valuable for face detection, another important question concerns how the amygdala actually processes this input to achieve the pattern discrimination that underlies its role in forming responses to salient stimuli. Amygdala neurons involved in a system for detecting behaviorally relevant stimuli are believed to represent the conjunction of a stimulus representation with its behavioral relevance. This system may perform a type of “lookup” operation, searching available memory for the incoming stimulus representation and then deciding if this stimulus has a previously associated behavioral value (Rolls, 1999). The incoming, high-level stimulus representation provided by IT is unlikely to resemble the relatively crude representation provided by a subcortical pathway. It is therefore unclear if this system can use two different source representations (i.e., a detailed cortical versus a crude subcortical representation) to identify the same value association (i.e., the behavioral value associated with the stimulus representation). There are at least two possible models to explain how vastly dissimilar subcortical and cortical representations could both trigger the same value association. First, it may be that when the amygdala receives relatively unprocessed subcortical input, it performs something analogous to simple template matching using certain representations that are tuned to specific views of behaviorally relevant stimuli. Evidence for templatebased pattern vision has been found in other animals that lack a visual cortex and whose pattern vision is mediated by the thalamus and the optic tectum, the homolog to the primate SC. In avians, these structures appear to accomplish pattern recognition through “active vision,” a template-matching mechanism wherein an animal actively changes its position in order to match its view with one of a limited number of stored pattern templates (Dawkins and Woodington, 2000). Second, it might be that subcortical visual structures preceding the amygdala perform a hierarchical construction of an object representation, akin to processes within cortical object recognition pathway. Our results are consistent with the results of blindsight and backward masking studies, suggesting that the SC and a visual posterior thalamic nucleus may be involved in such perceptual processes. Although more rudimentary and less detailed, this subcortical representation might sufficiently approximate the IT representation such that the same amygdala neurons can effectively use either representation to perform the necessary lookup. The pulvinar, which has direct connections to both the SC (Benevento and Fallon, 1975; Stepniewska et al., 2000) and the amygdala (Jones and Burton, 1976), is a likely candidate for an intermediate stage in a subcortical visual hierarchy, given its involvement in basic pattern discrimination (Chalupa et al., 1976). For a subcortical pathway orga-

nized for fast detection of salient stimuli, this hierarchical strategy would presumably be slower than a templatematching mechanism but also considerably faster than the multistage cortical processing stream. Determining if either of these two mechanisms is employed by the subcortical visual pathway is an important line of future work. Our results have demonstrated that differential activity for a behaviorally relevant and a neutral object in the amygdala can occur without input from IT and that this discrimination appears to be mediated by a subcortical pathway involving the SC and a dorsal posterior lateral region within the thalamus. Our method should also provide a means to further characterize the effects of rivalry suppression within subcortical visual structures and to identify the range of visual patterns to which this pathway is responsive in order to understand the analysis principles that support subcortical form discrimination. The computational mechanisms that underlie subcortical pattern vision may then yield insights into both the ontogenetic and evolutionary development of vision. Experimental Procedures Participants Twelve healthy participants (4 men, 4 left-handed) were paid for their participation and ranged in age from 18 to 34. All had normal or corrected-to-normal vision and gave written, informed consent in accordance with procedures and protocols approved by the Institutional Review Board of the Yale University School of Medicine. fMRI Data Acquisition In the scanner, participants viewed stimuli (subtending ⵑ5 visual degrees) through red/blue anaglyph glasses. Images were backprojected onto a translucent screen mounted near the end of the MRI gantry and were viewed through a periscopic prism system on the head coil. Behavioral response data were collected with a fiberoptic button box, and the participant’s head was immobilized using foam wedges and tape across the forehead. T2*-weighted images sensitive to BOLD contrast were acquired on a GE Sigma 1.5 Tesla scanner (LX operating system) with a standard quadrature head coil, using a gradient echo, single-shot echo planar sequence and an oblique axial orientation determined for each participant (TR ⫽ 1500 ms, TE ⫽ 40, flip angle ⫽ 60, NEX ⫽ 1, in-plane voxel size ⫽ 3.125 mm ⫻ 3.125 mm, 18 oblique slices, 3.3 mm thick with no gap). In all participants, these slices covered extrastriate cortex, the superior colliculus, the amygdala, and inferior aspects of the thalamus, and in six participants these slices also covered the entire thalamus. T1-weighted structural images of the same thickness (TR ⫽ 500, TE ⫽ 14, field of view ⫽ 200 mm, 256 mm ⫻ 192 mm matrix, 2 NEX) and a high-resolution 3D structural data set (3D SPGR, contiguous, sagittal acquisition, 124 images with 1.2 mm isotropic voxels, TR ⫽ 24, TE ⫽ 5, flip angle ⫽ 45, matrix ⫽ 192 ⫻ 256, NEX ⫽ 2, FOV ⫽ 30 cm) were collected in the same session. Rivalry and Localizer Experiments The rivalry run consisted of six repetitions of three block types corresponding to one nonrivalry condition (a moving house) and two rivalrous conditions (suppressed fearful face and suppressed chair). Stimuli were 10 s long movies created using Adobe Premiere (Adobe Systems, Inc., San Jose, CA, http://www.adobe.com). To allow the anaglyph glasses to filter different images to the two eyes for the rivalrous stimuli, images of houses defined by red luminance were overlaid on the image of a simple disc defined by blue luminance. The position of the red and blue lens was randomly alternated across participants between the right and left eye. Each trial began with the disc projected to one eye for 0.5 s. At this point, the house image appeared in the other eye and began to move sharply back and forth while the blue disc gradually faded into a target image of a fearful face or neutral chair also defined by blue luminance. The

Neuron 170

target image then faded back into the blue disc after approximately 1.5 s. In the nonrivalrous moving house condition, a house was presented binocularly with movement identical to the monocular house in the rivalrous stimuli. Four trials (2.5 s each) presented consecutively constituted one 10 s stimulus block with each block followed by a 10 s rest period. Participants performed a 1-back behavioral task indicating by button press if the current image was the same as or different than the previous image. We instructed participants to press a separate button if at any point they perceived anything besides a house, even small portions of a face or chair, in order to identify and to exclude from our analysis the blocks where perceptual suppression failed. Participants performed at greater than 90% accuracy with no difference between suppressed image trials (p ⬎ .5). All trials for which the participant did not respond were treated as unsuccessful suppression trials (since they could not be used to verify successful suppression). Across all participants, unsuccessful rivalry suppression trials occurred in 2% of these blocks, which were subsequently excluded from further analysis. The object localizer protocol was administered after the rivalry protocol. It consisted of three runs with four blocks each of fearful faces, neutral faces, and chairs, randomly intermixed and separated by a 10 s block of rest. Stimuli were presented as movies in a manner similar to the rivalry experiment except that each object type was presented binocularly with eight consecutive trials constituting a 20 s long block.

Data Analysis Image analyses and tests of statistical significance were done using Brainvoyager 2000 (Brain Innovation, Maastricht, The Netherlands, http://www.BrainVoyager.com), and locally developed software was used to inspect time courses and correlations. Motion-corrected images were high pass filtered at two cycles per time course and spatially smoothed using a Gaussian filter with a full-width halfmaximum value of 7 mm. The individual data were coregistered with high-resolution 2D anatomical images for display and localization. For group analysis, these data sets were then transformed into a proportional three-dimensional grid defined by Talairach and Tournoux (1988) and coregistered with the high-resolution 3D data set that was resampled to give 1 mm3 voxels. In the averaged group data set, ROIs corresponding to face- and chair-selective IT areas were functionally defined based on a fixed effects general linear model (GLM) analysis contrasting fearful faces with chairs which, after convolution with a standard hemodynamic response function (Boynton et al., 1996), were included as predictors in the GLM. Each ROI included contiguous voxels within ventral temporal cortex surviving a statistical threshold of p ⬍ .01, uncorrected (center Talairach coordinates for each cluster were: face-selective, [right] x ⫽ 47,y ⫽ ⫺43, z ⫽ ⫺18; [left] x ⫽ ⫺38, y ⫽ ⫺53, z ⫽ ⫺17; chairselective, [right] x ⫽ 28, y ⫽ ⫺35, z ⫽ ⫺10; [left] x ⫽ ⫺27, y ⫽ ⫺34, z ⫽ ⫺13). The subcortical ROIs were defined anatomically rather than functionally because (1) we did not have an independent functional task that could reliably activate the SC or the visual thalamus, and (2) we did not want to assume that the localizer task (i.e., dominant face/chair presentations) would activate the same subregions of the amygdala as the rivalry task (i.e., suppressed face/chair presentations), based on the possibility that subregions receiving cortical or subcortical input do not overlap. In the rivalry task, specific effects of each condition in our a priori group ROIs were evaluated by contrasts defined using a random effects GLM with three predictors corresponding to the suppressed face, suppressed chair, and dominant house conditions (separate predictors were defined for those blocks in which rivalry suppression failed). Predictors were convolved with a standard hemodynamic response function (Boynton et al., 1996). Statistical maps of specific contrasts were generated with a resampled voxel resolution of 1 mm3 corresponding to the resolution of the coregistered 3D anatomical. Because of strong a priori hypotheses regarding our ROIs, significant voxels within these regions were identified using a statistical threshold of p ⬍ .005, uncorrected with a spatial extent greater than five contiguous 1 mm3 voxels (reported p values are corrected for five multiple ROI comparisons). This cluster threshold was chosen in an effort to minimize the possibility of excluding any small subcor-

tical activation clusters, while at the same time eliminating single voxel activations that may be due to noise. Significant voxels outside of these ROIs for which we had no a priori hypotheses were corrected for multiple comparisons across the whole brain. We also inspected the individual activation maps at a less stringent threshold of p ⬍ .05, uncorrected, in order to determine the proportion of participants who exhibited activations for our main suppressed face versus suppressed chair contrast. Object-selective cortex was functionally defined on each individual’s 2D anatomy using a fearful face versus chair contrast derived from each individual’s localizer data. Time courses from the rivalry run were extracted from the five most significantly activated pixels in the left and right hemispheres in both face- and chair-selective IT. After normalization, the time courses were averaged by condition and the peak signal amplitudes (positive or negative) for each individual were entered into a two-way ANOVA (cortical ROI by suppressed image type) to test for possible differential effects of the suppressed image on the cortical ROIs. A second two-way ANOVA was performed that included the amygdala as an additional ROI factor in order to test for a significant differential effect of suppressed image type on the amygdala and cortical ROIs. Finally, a repeated measures two-way ANOVA was performed on the condition-specific response functions within each cortical ROI to test for significant differences between the two rivalrous and the nonrivalrous conditions (image type by time; the eight time points corresponding to an increased hemodynamic response were included, which took into account a 4.5 s lag for the peak hemodynamic response). We used a PPI analysis (Friston et al., 1997) to evaluate conditionspecific interactions between the left amygdala and our other a priori ROIs. In this separate analysis, we included the seven individuals for whom our slice selection covered the dorsal posterior thalamus and for whom we observed individually significant left amygdala activation (p ⬍ .05, uncorrected) resulting from a suppressed face versus suppressed chair contrast. The time courses from the maximally activated voxel in the left amygdala in each individual were extracted, normalized, and entered into a separate random effects analysis as a covariate of interest to identify significant, conditionspecific correlations within each ROI taken from the group averaged data. This analysis technique assesses whether correlations between regions during one condition differ significantly from those correlations during another condition. We also performed the same PPI analysis using the superior colliculus as the source data. For this we used ROIs that included the left and right superior colliculi defined on each individual’s 2D anatomical images of the six individuals for whom our slice selection covered the entire thalamus. The same thresholds used in the main effects analysis were also used in these PPI analyses, both within and outside the ROIs. Acknowledgments This work was supported by grants from the National Institute of Child Health and Human Development (grants PO1 HD 03008 and PO1 HD/DC35482). We wish to thank Hedy Sarofin, Terry Hickey, and Larry Win for technical assistance and members of the Perceptual Expertise Network for helpful discussion. Portions of this work were previously presented at Cognitive Neuroscience Society, 2003 (B.N.P. and R.T.S.; fMRI evidence for emotional processing of perceptually suppressed faces in the amygdala during binocular rivalry) and at the 9th annual Human Brain Mapping conference, New York, New York, June 18–22, 2003 (B.N.P. and R.T.S.; Differential amygdala but not extrastriate activity is evidence for categorization of objects not seen during binocular rivalry). Received: June 30, 2003 Revised: December 8, 2003 Accepted: February 13, 2004 Published: April 7, 2004 References Allison, T., Puce, A., Spencer, D.D., and McCarthy, G. (1999). Electrophysiological studies of human face perception. I: Potentials gener-


ated in occipitotemporal cortex by face and non-face stimuli. Cereb. Cortex 9, 415–430.

Jones, E.G., and Burton, H. (1976). A projection from the medial pulvinar to the amygdala in primates. Brain Res. 104, 142–147.

Bar, M., and Biederman, I. (1998). Subliminal visual priming. Psychol. Sci. 9, 464–469.

Kanwisher, N., McDermott, J., and Chun, M.M. (1997). The fusiform face area: a module in human extrastriate cortex specialized for face perception. J. Neurosci. 17, 4302–4311.

Bar, M., and Biederman, I. (1999). Localizing the cortical region mediating visual awareness of object identity. Proc. Natl. Acad. Sci. USA 96, 1790–1793. Benevento, L.A., and Fallon, J.H. (1975). The ascending projections of the superior colliculus in the rhesus monkey (Macaca mulatta). J. Comp. Neurol. 160, 339–361. Blake, R. (1989). A neural theory of binocular rivalry. Psychol. Rev. 96, 145–167. Blake, R. (2001). Primer on binocular rivalry, including controversial issues. Brain Mind 2, 5–38. Blake, R., and Logothetis, N.K. (2002). Visual competition. Nat. Rev. Neurosci. 3, 13–21. Boynton, G.M., Engel, S.A., Glover, G.H., and Heeger, D.J. (1996). Linear systems analysis of functional magnetic resonance imaging in human V1. J. Neurosci. 16, 4207–4221. Buckner, R.L., Goodman, J., Burock, M., Rotte, M., Koutstaal, W., Schacter, D., Rosen, B., and Dale, A.M. (1998). Functional-anatomic correlates of object priming in humans revealed by rapid presentation event-related fMRI. Neuron 20, 285–296. Calder, A.J., Lawrence, A.D., and Young, A.W. (2001). Neuropsychology of fear and loathing. Nat. Rev. Neurosci. 2, 352–363. Cave, C.B., Blake, R., and McNamara, T.P. (1998). Binocular rivalry disrupts visual priming. Psychol. Sci. 9, 299–302. Chalupa, L.M., Coyle, R.S., and Lindsley, D.B. (1976). Effect of pulvinar lesions on visual pattern discrimination in monkeys. J. Neurophysiol. 39, 354–369. Dailey, M.N., and Cottrell, G.W. (1999). Organization of face and object recognition in modular neural network models. Neural Netw. 12, 1053–1073. Dawkins, M.S., and Woodington, A. (2000). Pattern recognition and active vision in chickens. Nature 403, 652–655. Dehaene, S., Naccache, L., Cohen, L., Bihan, D.L., Mangin, J.F., Poline, J.B., and Riviere, D. (2001). Cerebral mechanisms of word masking and unconscious repetition priming. Nat. Neurosci. 4, 752–758. Epstein, R., and Kanwisher, N. (1998). A cortical representation of the local visual environment. Nature 392, 598–601. Friston, K.J., Buechel, C., Fink, G.R., Morris, J., Rolls, E., and Dolan, R.J. (1997). Psychophysiological and modulatory interactions in neuroimaging. Neuroimage 6, 218–229. Gallant, J. (2000). The neural representation of shape. In Seeing, K.K. DeValois and R.L. DeValois, eds. (San Diego, CA: Academic Press), pp. 311–333. Goebel, R., Muckli, L., Zanella, F.E., Singer, W., and Stoerig, P. (2001). Sustained extrastriate cortical activation without visual awareness revealed by fMRI studies of hemianopic patients. Vision Res. 41, 1459–1474. Goren, C.C., Sarty, M., and Wu, P.Y. (1975). Visual following and pattern discrimination of face-like stimuli by newborn infants. Pediatrics 56, 544–549. Grill-Spector, K., Kushnir, T., Edelman, S., Avidan, G., Itzchak, Y., and Malach, R. (1999). Differential processing of objects under various viewing conditions in the human lateral occipital complex. Neuron 24, 187–203.

Koutstaal, W., Wagner, A.D., Rotte, M., Maril, A., Buckner, R.L., and Schacter, D.L. (2001). Perceptual specificity in visual object priming: functional magnetic resonance imaging evidence for a laterality difference in fusiform cortex. Neuropsychologia 39, 184–199. Kovacs, G., Vogels, R., and Orban, G.A. (1995). Cortical correlate of pattern backward masking. Proc. Natl. Acad. Sci. USA 92, 5587– 5591. Kraemer, M., and Sjostrom, A. (1998). Lack of short-latency-potentials in the VEP reflects immature extra geniculate visual function in delayed visual maturation (DVM). Doc. Ophthalmol. 97, 189–201. Kreiman, G., Fried, I., and Koch, C. (2002). Single-neuron correlates of subjective vision in the human medial temporal lobe. Proc. Natl. Acad. Sci. USA 99, 8378–8383. LeDoux, J.E. (1996). The Emotional Brain: The Mysterious Underpinnings of Emotional Life (New York: Simon & Schuster). Leopold, D.A., and Logothetis, N.K. (1996). Activity changes in early visual cortex reflect monkeys’ percepts during binocular rivalry. Nature 379, 549–553. Linke, R., De Lima, A.D., Schwegler, H., and Pape, H.C. (1999). Direct synaptic connections of axons from superior colliculus with identified thalamo-amygdaloid projection neurons in the rat: possible substrates of a subcortical visual pathway to the amygdala. J. Comp. Neurol. 403, 158–170. Logothetis, N.K. (2003). The underpinnings of the BOLD functional magnetic resonance imaging signal. J. Neurosci. 23, 3963–3971. Lomber, S.G. (2002). Learning to see the trees before the forest: reversible deactivation of the superior colliculus during learning of local and global visual features. Proc. Natl. Acad. Sci. USA 99, 4049– 4054. Lumer, E.D., and Rees, G. (1999). Covariation of activity in visual and prefrontal cortex associated with subjective visual perception. Proc. Natl. Acad. Sci. USA 96, 1669–1673. Lumer, E.D., Friston, K.J., and Rees, G. (1998). Neural correlates of perceptual rivalry in the human brain. Science 280, 1930–1934. Miller, M., Pasik, P., and Pasik, T. (1980). Extrageniculostriate vision in the monkey. VII. Contrast sensitivity functions. J. Neurophysiol. 43, 1510–1526. Morris, J.S., Ohman, A., and Dolan, R.J. (1998). Conscious and unconscious emotional learning in the human amygdala. Nature 393, 467–470. Morris, J.S., Ohman, A., and Dolan, R.J. (1999). A subcortical pathway to the right amygdala mediating “unseen” fear. Proc. Natl. Acad. Sci. USA 96, 1680–1685. Morris, J.S., DeGelder, B., Weiskrantz, L., and Dolan, R.J. (2001). Differential extrageniculostriate and amygdala responses to presentation of emotional faces in a cortically blind field. Brain 124, 1241– 1252. Morton, J., and Johnson, M.H. (1991). CONSPEC and CONLERN: a two-process theory of infant face recognition. Psychol. Rev. 98, 164–181. Nakamura, K., Mikami, A., and Kubota, K. (1992). Activity of single neurons in the monkey amygdala during performance of a visual discrimination task. J. Neurophysiol. 67, 1447–1463.

Henson, R.N., and Rugg, M.D. (2003). Neural response suppression, haemodynamic repetition effects, and behavioural priming. Neuropsychologia 41, 263–270.

Pessoa, L., Kastner, S., and Ungerleider, L.G. (2002a). Attentional control of the processing of neutral and emotional stimuli. Cogn. Brain Res. 15, 31–45.

Ishai, A., Ungerleider, L.G., Martin, A., Schouten, J.L., and Haxby, J.V. (1999). Distributed representation of objects in the human ventral visual pathway. Proc. Natl. Acad. Sci. USA 96, 9379–9384.

Pessoa, L., McKenna, M., Gutierrez, E., and Ungerleider, L.G. (2002b). Neural processing of emotional faces requires attention. Proc. Natl. Acad. Sci. USA 99, 11458–11463.

Iwai, E., and Yukie, M. (1987). Amygdalofugal and amygdalopetal connections with modality-specific visual cortical areas in macaques (Macaca fuscata, M. mulatta, and M. fascicularis). J. Comp. Neurol. 261, 362–387.

Rolls, E.T. (1999). Brain and Emotion (Oxford: Oxford University Press). Rolls, E.T. (2000). Neurophysiology and functions of the primate amygdala, and the neural basis of emotion. In The Amygdala: A

Neuron 172

Functional Analysis, J.P. Aggleton, ed. (New York: Oxford University Press), pp. 447–478. Rolls, E.T., and Tovee, M.J. (1994). Processing speed in the cerebral cortex and the neurophysiology of visual masking. Proc. R. Soc. Lond. B Biol. Sci. 257, 9–15. Rolls, E.T., Tovee, M.J., Purcell, D.G., Stewart, A.L., and Azzopardi, P. (1994). The responses of neurons in the temporal cortex of primates, and face identification and detection. Exp. Brain Res. 101, 473–484. Rolls, E.T., Tovee, M.J., and Panzeri, S. (1999). The neurophysiology of backward visual masking: information analysis. J. Cogn. Neurosci. 11, 300–311. Romanski, L.M., Giguere, M., Bates, J.F., and Goldman-Rakic, P.S. (1997). Topographic organization of medial pulvinar connections with the prefrontal cortex in the rhesus monkey. J. Comp. Neurol. 379, 313–332. Rosa, M.G., and Schmid, L.M. (1994). Topography and extent of visual-field representation in the superior colliculus of the megachiropteran Pteropus. Vis. Neurosci. 11, 1037–1057. Schiller, P.H., and Malpeli, J.G. (1977). Properties and tectal projections of monkey retinal ganglion cells. J. Neurophysiol. 40, 428–445. Schyns, P.G., and Oliva, A. (1999). Dr. Angry and Mr. Smile: when categorization flexibly modifies the perception of faces in rapid visual presentations. Cognition 69, 243–265. Sewards, T.V., and Sewards, M.A. (2002). Innate visual object recognition in vertebrates: some proposed pathways and mechanisms. Comp. Biochem. Physiol. A Mol. Integr. Physiol. 132, 861–891. Sheinberg, D.L., and Logothetis, N.K. (1997). The role of temporal cortical areas in perceptual organization. Proc. Natl. Acad. Sci. USA 94, 3408–3413. Simion, F., Valenza, E., Umilta, C., and Dalla Barba, B. (1998). Preferential orienting to faces in newborns: a temporal-nasal asymmetry. J. Exp. Psychol. Hum. Percept. Perform. 24, 1399–1405. Stefanacci, L., and Amaral, D.G. (2000). Topographic organization of cortical inputs to the lateral nucleus of the macaque monkey amygdala: a retrograde tracing study. J. Comp. Neurol. 421, 52–79. Stefanacci, L., and Amaral, D.G. (2002). Some observations on cortical inputs to the macaque monkey amygdala: an anterograde tracing study. J. Comp. Neurol. 451, 301–323. Stepniewska, I., Ql, H.X., and Kaas, J.H. (2000). Projections of the superior colliculus to subdivisions of the inferior pulvinar in New World and Old World monkeys. Vis. Neurosci. 17, 529–549. Talairach, J., and Tournoux, P. (1988). Co-planar Stereotaxic Atlas of the Human Brain (New York: Thieme). Tong, F. (2001). Competing theories of binocular rivalry: a possible resolution. Brain Mind 2, 55–83. Tong, F., and Engel, S.A. (2001). Interocular rivalry revealed in the human cortical blind-spot representation. Nature 411, 195–199. Tong, F., Nakayama, K., Vaughan, J.T., and Kanwisher, N. (1998). Binocular rivalry and visual awareness in human extrastriate cortex. Neuron 21, 753–759. Vuilleumier, P., Armony, J.L., Driver, J., and Dolan, R.J. (2003). Distinct spatial frequency sensitivities for processing faces and emotional expressions. Nat. Neurosci. 6, 624–631. Webster, M.J., Ungerleider, L.G., and Bachevalier, J. (1991). Connections of inferior temporal areas TE and TEO with medial temporallobe structures in infant and adult monkeys. J. Neurosci. 11, 1095– 1116. Weiskrantz, L. (1997). Consciousness Lost and Found (Oxford: Oxford University Press). Whalen, P.J., Rauch, S.L., Etcoff, N.L., McInerney, S.C., Lee, M.B., and Jenike, M.A. (1998). Masked presentations of emotional facial expressions modulate amygdala activity without explicit knowledge. J. Neurosci. 18, 411–418. Zimba, L.D., and Blake, R. (1983). Binocular rivalry and semantic processing: out of sight, out of mind. J. Exp. Psychol. Hum. Percept. Perform. 9, 807–815.