Abstract

Processing spatial configuration is a fundamental requirement for object recognition. Using fMRI, the neural basis underlying this ability was examined while human participants viewed possible and visually similar, but spatially impossible, objects presented for either long or short exposure duration. Response profiles in object-selective cortical regions exhibited sensitivity to object possibility, but only for the long exposure duration. Contrary, functional connectivity, indexed by the pairwise correlations between activation profiles across ROIs, revealed sensitivity to possibility, evident in enhanced correlations for impossible compared with possible objects. Such sensitivity was found even following a brief exposure duration, which allowed only minimal awareness of possibility. Importantly, this sensitivity was correlated with participants' general spatial ability as assessed by an independent neuropsychological test. These results suggest that the visual system is highly susceptible to objects' 3-D structural information even with minimal perceptual awareness. Such sensitivity is captured at the level of functional connectivity between object-selective regions, rather than the absolute level of within-region activity, implicating the role of interregional synchronization in the representation of objects' 3-D structure.

INTRODUCTION

Objects in the visual environment have diverse visual structure and can appear in numerous viewpoints; hence, rendering the representation of 3-D information is critical for object recognition. Consistently, neurophysiological evidence suggest that a set of cortical regions, along the dorsal and ventral visual streams, is susceptible to 3-D cues (e.g., Georgieva, Todd, Peeters, & Orban, 2008; Welchman, Deubelius, Conrad, Bulthoff, & Kourtzi, 2005), particularly if they carry information about objects' 3-D structure (Kornblith, Cheng, Ohayon, & Tsao, 2013; Orban, 2011; Kourtzi, Erb, Grodd, & Bulthoff, 2003; Faillenot, Decety, & Jeannerod, 1999). Collectively, these studies indicate that objects' 3-D structure is processed by a number of cortical regions, rather than being localized to one particular brain area. This notion is consistent with recent literature emphasizing the role of distributed cortical networks in various complex cognitive functions (Behrmann & Plaut, 2013; Haxby et al., 2001). Nevertheless, the nature of the neural representation of objects' 3-D structure in specific cortical regions and at the network level is still largely unknown (see Orban, 2011). The goal of the current study is to examine the extent to which 3-D object structure is represented in a distrusted fashion by quantitatively measuring both the within-region activity and the synchronization between object-selective regions in the visual cortex, often referred to as “functional connectivity” (Friston, 2011; Biswal, Zerrin Yetkin, Haughton, & Hyde, 1995).

To that end, we utilized a depth illusion, known as “impossible objects” (Penrose & Penrose, 1958). These 2-D stimuli represent objects that violate fundamental rules of spatial organization and therefore could not exist in real 3-D space. Importantly, at the 2-D level, the pixel-wise physical differences between object categories are minor; however, the interpretation of the 3-D structure of these two object types is markedly different and elicits remarkable differences at the phenomenological–perceptual experience (Figure 1). Thus, impossible objects provide unusual circumstances in which the 2-D features comprising an object are well defined whereas the overall 3-D structural interpretation of this object is incoherent. This unique object category was previously used to shed light on various domains such as long-term memory representation (e.g., Schacter, Cooper, & Delaney, 1990), development of perceptual processes in infants (e.g., Shuwairi, Albert, & Johnson, 2007), and perceptual processes that mediate the representation of 3-D information (e.g., Freud, Avidan, & Ganel, 2015). In the current study, we utilized possible and impossible objects in a multisession fMRI experimental design that allows probing the nature of the representation of objects' 3-D structure in the visual cortex at the level of specific regional activity as well as at the level of distinct patterns of functional connectivity across regions.

Figure 1. 

Examples of possible (left) and their matched counterpart impossible objects (right). Note that there are minimal physical differences between object types, whereas the perceptual experience in viewing these two object sets is substantially different.

Figure 1. 

Examples of possible (left) and their matched counterpart impossible objects (right). Note that there are minimal physical differences between object types, whereas the perceptual experience in viewing these two object sets is substantially different.

To date, the underlying cognitive and neural mechanisms that support the representation of object possibility/impossibility are still largely unknown. Particularly, several studies demonstrated that the visual system of both animals (newly hatched chicks) and humans is highly susceptible to the distortions of objects' 3-D structure, which underlie impossible objects (Regolin, Rugani, Stancher, & Vallortigara, 2011; Shuwairi, 2009; Shuwairi et al., 2007). This sensitivity further emphasizes the importance of 3-D interpretation for object recognition. Nevertheless, we recently showed that, despite their inherent incoherency, impossible objects may still be represented in the human visual system, at least initially, by similar perceptual and neural principles as possible objects (Freud, Avidan, & Ganel, 2013; Freud, Ganel, & Avidan, 2013). Hence, it is still unclear how the visual system encodes the differences between the two object categories that support the sensitivity to objects' 3-D structure and how such incoherency in 3-D structure is represented at the neural level.

Previous research on the representation of objects' 3-D structure was mainly focused on within-region activity and revealed sensitivity to 3-D structure in several cortical regions such as lateral occipital complex (LOC) and intraparietal sulcus (IPS; Georgieva et al., 2008; Murray & He, 2006). Moreover, it was also found that the mere perceived experience of a stimulus as a 3-D object elicited greater fMRI activation compared with the response to the same stimulus when it was perceived as a 2-D object (Moore & Engel, 2001). Note that the 3-D structure of impossible objects contains irregularities that may enhance the processing demands in object-selective regions. Hence, in line with the above findings, we predicted that impossible objects would elicit greater activation within object-selective regions. Importantly, the activation level in high-order visual cortex is also known to be correlated to the conscious perceptual experience (e.g., Grill-Spector et al., 2000). Therefore, in the current study, we manipulated the exposure duration to minimize explicit awareness of object possibility, and we predicted that under this condition no marked differences would be observed in the pattern of regional activation between object categories.

Finally, although most previous studies focused on the role of within-region activity in the representation of objects' 3-D structure, the role of functional connectivity between object-selective regions is still largely unknown. Nonetheless, the sensitivity of multiple cortical regions, in both the dorsal and ventral visual streams, to depth information strongly suggests that the perception of 3-D structure relies on a distributed representation (Orban, 2011; Welchman et al., 2005). The concept of large-scale connectivity has been extensively investigated in the last several years in the context of fMRI studies using diverse approaches (for a review, see Friston, 2011). One of the leading approaches, termed functional connectivity, which is also used in this study, is based on measuring pairwise correlations between activation profiles of specific ROIs to assess synchronous patterns across regions. Although the usage of simple correlations does not enable to infer causality or the existence of direct connections between two regions (in contrast to other approaches such as effective connectivity, see Friston, 2011), this method was shown to be very useful for describing the topology of brain networks produced by fMRI (Smith et al., 2011). Moreover, such measures could account for a wide range of cognitive phenomena (e.g., Dinstein et al., 2011; Wang et al., 2006, 2007) and are consistent with patterns of anatomical organization (e.g., Honey et al., 2009). Thus, in this study, in addition to the within-region activity, we also measured the pairwise correlations between object-selective regions (i.e., functional connectivity) in the visual cortex. Given the predicted distributed nature of 3-D representation in the visual cortex, our main hypothesis was that impossibility would modulate not only the within-region activation profile but also the pattern of functional connectivity between regions.

METHODS

Participants

Fifteen healthy, right-handed individuals (eight women, age = 22–29 years) with normal or corrected-to-normal vision participated in the experiment (8 in the 2000 msec exposure duration group, 7 in the 30 msec exposure duration group). The results of additional five participants were discarded (three participants with excessive head movements, one participant that fell asleep during the scan, and one that performed below 50% in the 1-back perceptual discrimination task used during the scan). The experiments were approved by the Helsinki committee of the Soroka Medical Center, Beer Sheva, Israel.

Stimuli

Stimuli consisted of 42 pairs of grayscale line-drawings of possible and matched impossible objects adapted with permission from the impossible world website (im-possible.info/english/index.html; Figure 1). Matched objects are identical objects, except for a single or a few features that transform the object's global 3-D structure from being possible to being impossible or vice versa. All stimuli were used in a previous fMRI study (Freud, Ganel, et al., 2013). To ensure that low-level features were equal between the two object categories, for each object, the overall number of pixels and the number of pixels that defined edges were calculated. Paired t tests validated that no significant differences were found between object categories in these two measures (mean number of overall pixels: 37,323 vs. 37,365 for possible vs. impossible objects, respectively; mean number of pixels defining edges: 3738 vs. 3730 for possible vs. impossible objects, respectively; ts < 1).

General Procedure

Participants completed two separate sessions of fMRI scans and an additional session of behavioral experiments. The first scanning session included a 3-D anatomical scan, localizer run, and four runs, which included possible/impossible object stimuli optimally designed for a functional connectivity analysis (see details below). The second session, which took place several weeks following the first session, included an anatomical 3-D scan used for registration to the first scan and four runs of block design presentations of possible/impossible objects optimally designed for a univariate analysis. The order of the experimental runs was counterbalanced across participants. Several weeks following the completion of both imaging sessions, participants also took part in a behavioral session, which included two experiments: a possible/impossible classification task and the Money Road-Map Test (Vingerhoets, Lannoo, & Bauwens, 1996; Money, Walker, & Alexander, 1965), a neuropsychological task that assesses general spatial ability.

Imaging Experiments

MRI Setup

Participants were scanned in a 3T Philips Ingenia scanner (Amsterdam, The Netherlands) equipped with a standard head coil, located at the Soroka Medical Center, Beer Sheva, Israel. fMRI BOLD contrast was acquired using the gradient-echo echo-planner imaging sequence with parallel acquisition (SENSE: factor 2.8). Specific scanning parameters were as follows: whole brain coverage 35 slices, transverse orientation, 3 mm thickness, no gap, repetition time = 2000 msec, echo time = 35 msec, flip angle = 90°, field of view = 256 × 256, voxel size = 3 × 3 × 3 mm3, and matrix size = 96 × 96. High-resolution anatomical volumes were acquired with a T1-weighted 3-D pulse sequence (1 × 1 × 1 mm3, 170 slices).

Visual Stimulation

Stimuli were presented using the E-prime 2.0 software (Psychology Software Tools, Inc., Pittsburgh, PA) and projected to an LCD screen located at the back of the scanner bore behind the participant's head. Participants viewed the stimuli through a tilted mirror mounted above their eyes on the head coil; before scanning, they all completed a short training session to acquaint them with the experimental tasks and stimuli.

Localizer Scan

A standard block design localizer experiment was used to define object- and face-selective regions. Stimuli were presented in 10-sec blocks composed of nine images of faces, houses, daily objects, novel possible objects, or scrambled objects. Participants performed a 1-back task (indicate an immediate repeat of the same object) to maintain attention throughout the experiment, and there was one image repetition per block. Accuracy and RTs were recorded.

Experimental Runs

Participants completed four scans optimized for univariate analysis and four scans optimized for functional connectivity analysis in different sessions. Before the experiments, they were told that novel objects would be presented during the scan but the nature of object impossibility was not explicitly referred to in order to avoid attentional and expectation confounds. Possible and impossible objects were presented following an initial fixation period of 20 sec. To ensure that participants maintained their attention, they were asked to perform a 1-back task during the experimental runs; one of eight stimuli was repeated in all experiments.

Experiment 1—Univariate Design

Participants completed four runs in which stimuli were presented in 20-sec blocks, with each block containing eight stimuli. Each run was composed of 15 blocks of either possible or impossible objects (total of four runs). Experimental blocks were separated by 6-sec-long fixation periods. In Experiment 1a, stimuli were presented for 2000 msec followed by 500-msec fixation. In Experiment 1b, stimuli were presented for 30 msec followed by 2470-msec fixation (total trial duration = 2500 msec in both experiments). Behavioral pilot experiments indicated that after the long exposure duration, participants are well aware of object impossibility whereas after the short presentation duration, explicit recognition of object impossibility is close to chance level.

Experiment 2—Functional Connectivity Design

Participants completed four runs in which either possible or impossible objects were presented (two runs per each stimulus category). After an initial fixation period of 20 sec, stimuli were presented continuously. Total trial duration varied between 2200 and 2800 msec, and it was composed of a stimulus presentation followed by a varying fixation duration. In Experiment 2a stimuli were presented for 2000 msec with a random fixation interval between 200 and 800 msec. In Experiment 2b, stimuli were presented for 30 msec with a random fixation interval between 2170 and 2770 msec. In each scan, a total of 21 stimuli were repeated five to six times (total of 120 trials per scan). Across the four scans, stimuli were identical to the stimuli of Experiment 1, which was used for the univariate analysis. The absence of long rest periods in this design was aimed to maximize the differences between scans elicited by the visual stimuli (given that rest periods are equal between scans) and to allow optimal analysis of functional connectivity. Note that, in the absence of a baseline condition, this design does not enable estimation of the BOLD signal in response to particular stimuli.

Behavioral Experiments

Object Classification Task

Participants were asked to classify objects as possible or impossible (Schacter et al., 1990). Before the experiment, they were told that possible objects were defined as objects that could be created in 3-D space whereas impossible objects were defined as objects that could not be created in 3-D space because they violate fundamental rules of spatial organization. Possible and impossible objects were randomly intermixed, and exposure durations were similar to those used during the fMRI scans (2000 msec/30 msec). Two participants (one from each experiment) that took part at the fMRI sessions were unavailable for the behavioral experiments.

The Money Road-Map Test

To assess the general spatial ability of the participants, they performed a computerized version of the Money Road-Map Test (Vingerhoets et al., 1996; Money et al., 1965), which is commonly used in the neuropsychological literature (Mapstone, Steffenella, & Duffy, 2003; Flament et al., 1990). This test is composed of 32 steps and measures the performance of left–right orientation decisions that participants have to make to reach their target end point, 10 of these decisions involve egocentric mental rotation in space of more than 90°, 13 decisions require mental rotation of approximately 90°, and 9 decisions do not require mental rotation. Participants were instructed to follow a dashed line as if they were tracing a map. In each junction, they were asked to indicate whether they turned to the right or left by pressing a keyboard button. Head movements were restricted using a chin rest, and no body movements were allowed. A short practice of three turns was conducted before the experiments. Accuracy and RTs served as the dependent variables.

Data Analysis

Imaging Experiments

fMRI data were processed using the BrainVoyager QX software (BrainInnovations, Maastricht, the Netherlands, version 2.6, RRID: nif-0000-00274) and complementary in-house software written in Matlab (The MathWorks, Inc., Natick, MA, RRID: nlx_153890). Preprocessing included 3-D motion correction, slice time correction, regression of auto correlation (AR2), and filtering of low temporal frequencies (slow drift).

ROI Selection

The localizer experiment was analyzed using a general linear model (GLM). For each participant, ROIs were defined at a significance level of at least q(FDR) < .05. Table 1 includes the mean Talairach coordinates of the selected ROIs, the mean cluster size, and the number of participants exhibiting each ROI.

Table 1. 

Talairach Coordinates, Cluster Size of ROIs, and Number of Participants Exhibiting Activation in Each ROI

ROIHemispherexyzSize ± SD (voxels)n
LOC 34.2 ± 3 −81.5 ± 3.6 1.3 ± 3.8 1463.8 ± 973 12 
−35 ± 3.3 −81.5 ± 3.3 −1 ± 3.8 1639.8 ± 1310 14 
ITS 44.2 ± 3.8 −61.6 ± 5.4 −7.4 ± 4.3 893.1 ± 758 14 
−46.8 ± 3.2 −61.3 ± 4 −6.8 ± 3.4 2580.3 ± 1082 14 
TOS 29.7 ± 3.5 −81.0 ± 4.1 13.0 ± 4.8 1281.5 ± 839 15 
−30.2 ± 4.3 −83.2 ± 3.5 12.5 ± 4.4 936.2 ± 524 15 
IPS 24.2 ± 3 −68.9 ± 7 37.8 ± 6.7 2603.2 ± 1648 15 
−25.9 ± 4.2 −70.9 ± 6.2 36.1 ± 5.7 3026.6 ± 1902 15 
pFs 28.3 ± 3.9 −39.4 ± 4.8 −13.5 ± 3.3 266.6 ± 118 10 
−31.1 ± 4.4 −41.4 ± 5 −13.5 ± 3.4 348.9 ± 116 10 
FFA 38.1 ± 3.2 −49.4 ± 4.8 −16.8 ± 4 1696.5 ± 1581 15 
−41.2 ± 2.8 −46.9 ± 5.2 −18.2 ± 3.4 1293.0 ± 1015 14 
OFA 35.0 ± 3.8 −76.2 ± 5.8 −14.2 ± 3.2 927.8 ± 765 13 
−38.4 ± 3.5 −74.2 ± 7.8 −14.3 ± 4.6 875.8 ± 746 13 
pSTS 47.6 ± 5 −45.3 ± 7.8 10.0 ± 5.4 853.4 ± 512 13 
−51.3 ± 6.7 −45.4 ± 7.5 10.0 ± 6 602.4 ± 355 
EVC 8.1 ± 5.7 −93.6 ± 2.9 −4.1 ± 5.7 69.9 ± 19 15 
−7.4 ± 6.7 −95.1 ± 2.6 −5.5 ± 8.3 67.5 ± 27 15 
ROIHemispherexyzSize ± SD (voxels)n
LOC 34.2 ± 3 −81.5 ± 3.6 1.3 ± 3.8 1463.8 ± 973 12 
−35 ± 3.3 −81.5 ± 3.3 −1 ± 3.8 1639.8 ± 1310 14 
ITS 44.2 ± 3.8 −61.6 ± 5.4 −7.4 ± 4.3 893.1 ± 758 14 
−46.8 ± 3.2 −61.3 ± 4 −6.8 ± 3.4 2580.3 ± 1082 14 
TOS 29.7 ± 3.5 −81.0 ± 4.1 13.0 ± 4.8 1281.5 ± 839 15 
−30.2 ± 4.3 −83.2 ± 3.5 12.5 ± 4.4 936.2 ± 524 15 
IPS 24.2 ± 3 −68.9 ± 7 37.8 ± 6.7 2603.2 ± 1648 15 
−25.9 ± 4.2 −70.9 ± 6.2 36.1 ± 5.7 3026.6 ± 1902 15 
pFs 28.3 ± 3.9 −39.4 ± 4.8 −13.5 ± 3.3 266.6 ± 118 10 
−31.1 ± 4.4 −41.4 ± 5 −13.5 ± 3.4 348.9 ± 116 10 
FFA 38.1 ± 3.2 −49.4 ± 4.8 −16.8 ± 4 1696.5 ± 1581 15 
−41.2 ± 2.8 −46.9 ± 5.2 −18.2 ± 3.4 1293.0 ± 1015 14 
OFA 35.0 ± 3.8 −76.2 ± 5.8 −14.2 ± 3.2 927.8 ± 765 13 
−38.4 ± 3.5 −74.2 ± 7.8 −14.3 ± 4.6 875.8 ± 746 13 
pSTS 47.6 ± 5 −45.3 ± 7.8 10.0 ± 5.4 853.4 ± 512 13 
−51.3 ± 6.7 −45.4 ± 7.5 10.0 ± 6 602.4 ± 355 
EVC 8.1 ± 5.7 −93.6 ± 2.9 −4.1 ± 5.7 69.9 ± 19 15 
−7.4 ± 6.7 −95.1 ± 2.6 −5.5 ± 8.3 67.5 ± 27 15 

Talairach coordinates and cluster size of each ROI as sampled from the localizer experiment. LOC = lateral occipital complex; ITS = inferior temporal sulcus; TOS = transverse occipital sulcus; IPS = intraparietal sulcus; pFs = posterior fusiform; FFA = fusiform face area; OFA = occipital face area; pSTS = posterior superior temporal sulcus; EVC = early visual cortex. (The same abbreviations are used in all figures.)

Object-selective regions were defined by the contrast of novel objects and daily objects > faces, and these regions were used for both the univariate and network analysis. These regions included the LOC (Malach et al., 1995), the inferior temporal sulcus (Hasson, Harel, Levy, & Malach, 2003), and the posterior fusiform (pFs; Hasson et al., 2003). To demarcate this latter ROI from the adjacent parahippocampal gyrus, typically activated by buildings, any voxels that were more selective to buildings compared with novel and daily objects were excluded from further analysis using a conjunction analysis. Additionally, two dorsal regions were defined as ROIs, the transverse occipital sulcus, which is known as a scene-selective region (Hasson et al., 2003) but is also responsive to spatial properties of objects (Troiani, Stigliani, Smith, & Epstein, 2014), and the IPS, which is suggested to be involved in the perception of objects' spatial features (Faillenot et al., 1999), also exhibited stronger activation for novel and daily objects compared with faces, and is therefore also included in the object-selective ROIs.

The early visual cortex (EVC) was defined to serve as a control region for the univariate analysis and was delineated based on a combination of anatomical and functional landmarks. Particularly, we defined voxels at the posterior part of the calcarine sulcus that were significantly activated by the contrast of scrambled objects > rest. The size of this ROI was restricted to 125 anatomical voxels.

For the network analysis we also defined the face-selective network, which served as a control for the object-selective network, based on the reverse contrast (i.e., faces > objects and daily objects). This network included the fusiform face area (FFA; Kanwisher, McDermott, & Chun, 1997), the occipital face area (OFA), and the posterior STS (pSTS; Gauthier et al., 2000), which are all part of the core face network (Haxby, Hoffman, & Gobbini, 2000).

Univariate Analysis

The four runs of each participant were concatenated and analyzed using a GLM. Because there were no significant hemispheric differences in any of the predefined object-selective ROIs, we describe the results below after pooling together the signal in each ROI across the two hemispheres while taking into account the size of the ROI (in terms of the number of voxels) in each hemisphere as in previous studies (e.g., Freud, Ganel, et al., 2013). The signal in each ROI was calculated as follows:
formula
These values were then also used for calculating mean beta weights across all object-selective regions.
For each participant, we used the mean beta weights obtained from all object-selective regions for impossible and possible objects to calculate a ratio measuring the selectivity to possible/impossible objects:
formula
Positive values indicate that impossible objects elicit greater activation compared with possible ones whereas negative values indicate the opposite. Zero indicates similar activation for the two object categories.

Functional Connectivity Analysis

To calculate functional connectivity, the time course of each ROI was extracted and a pairwise correlation coefficient was calculated between each pair of ROIs. The level of correlation between homologous regions in the two hemispheres is known to be particularly strong (Biswal et al., 1995) and have unique functional significance (Konen, Behrmann, Nishimura, & Kastner, 2011); therefore, we did not collapse the functional connectivity analysis across the two hemispheres. To enable the usage of statistical parametric tests on the correlation coefficients, the Fisher's z transformation was first applied to these values. Next, for each participant, we calculated the mean correlation between all predefined object-selective ROIs and the mean correlation between all predefined face-selective ROIs. Finally, we performed a two-way ANOVA with object type (possible/impossible) and network (objects/faces) as independent variables. For each participant, we used the mean correlation between object-selective ROIs for impossible and possible objects to calculate the functional connectivity (FC) selectivity ratio:
formula
Positive values indicate higher correlations for impossible compared with possible objects whereas negative values indicate the opposite. Zero indicates similar correlation levels for the two object categories.

An additional analysis was aimed to examine the contribution of specific ROIs to the overall functional connectivity pattern. Basically, this analysis could be performed by averaging the correlations of each ROI with all the remaining ROIs; however, such a procedure would lead to inflated correlations, which are, by definition, intercomposed of each other. That is, each correlation is being calculated more than once, and with 10 ROIs, as in this study, each correlation is being used 10 times. To avoid this caveat, we conducted a bootstrapping analysis. Because no interaction was found between object possibility and exposure duration in Experiment 2a and Experiment 2b, the analysis was collapsed across these experiments. Each iteration, out of a 10,000 bootstrap iterations, was composed of the following steps: (a) shuffling the order of the 10 ROIs, (b) calculating the mean correlation between the first ROI (within a particular order) with the remaining ROIs separately for possible and impossible objects, (c) calculating the difference between the mean correlation for possible and impossible objects for the first ROI, (d) Removing the first ROI from the correlation matrix such that it would not be included in the mean correlations calculated for the following ROIs, and (e) repeating Steps b to d for the remaining ROIs. To establish the statistical significance of this analysis, for each ROI, we calculated the 95% confidence interval of the obtained bootstrap distribution of the mean differences between impossible and possible objects.

Behavioral Experiments

For the possible/impossible classification task, accuracy and RTs were measured. Additionally, because each stimulus was presented five to six times along the experiment, the repetition number served as a within-participant independent variable that was used to assess whether awareness to object possibility increased during the course of the experiment.

The Money Road-Map Test consisted of 32 trials and accuracy (i.e., number of correct “turns” of all trials) as well as RTs were calculated for each participant.

RESULTS

Behavioral Results during fMRI Scanning

Accuracy was high in both experiments (Experiment 1: 99%, SD = 1%; Experiment 2: 93%, SD = 6%), and no differences in accuracy were observed between possible and impossible objects in the 1-back task (ps > .3). Additionally, RTs in both experiments were similar across object categories (Experiment 1: possible 563 msec [SD = 76 msec], impossible 576 msec [SD = 93 msec]; Experiment 2: possible 618 msec [SD = 110 msec], impossible 602 msec [SD = 115 msec]; all ps > .5). These results are in-line with our previous study in which behavioral performance in a same/different classification task was similar across the two object categories (Freud, Ganel, et al., 2013).

fMRI Univariate Analysis

The goal of this analysis was to assess the sensitivity to possible versus impossible objects as reflected in the BOLD regional responses. The analysis was first conducted separately for each object-selective region as well as for the EVC extracted for each participant. Next, we calculated the mean of the beta weights for all the object-selective regions (see Methods). An ANOVA was conducted for the results of Experiment 1a (exposure duration of 2000 msec). This analysis revealed a two-way interaction between Region (object-selective/EVC) and Object type (possible/impossible) [F(1, 7) = 18.15, p < .01, ηp2 = 0.72], which stemmed from greater activation for impossible objects [F(1, 7) = 22.9, p < .01] in object-selective regions. Conversely, in EVC, a similar BOLD signal was observed for the two object types [F(1, 7) = 2.1, p > .15] (Figure 2A).

Figure 2. 

Univariate analysis results. In Experiment 1a, when stimulus exposure duration was 2000 msec, all object-selective regions exhibited stronger activation for impossible compared with possible objects. (A) Activation in EVC did not differ between the two object categories. (B) In Experiment 1b, when stimulus exposure duration was 30 msec, no differences were observed in the activation patterns of possible and impossible objects. (C) The average beta weights of object-selective regions in Experiments 1a and 1b reflect stronger activation for impossible objects, but only in Experiment 1a. Error bars in all figures denote ICIs (Tryon & Lewis, 2008; Tryon, 2001). (D) A whole-brain group analysis of Experiment 1a. Significant voxels (p < .0005, uncorrected) with greater activation for impossible objects are overlaid on an axial slice from a representative participant. The Talairach coordinates were as follows: right hemisphere: x = 45.8, y = −64.9, z = −5.1; left hemisphere: x = −43.9, y = −66.8, z = −2.3. Mean activation profiles (percent signal change) for possible and impossible objects obtained from the left and right significant activated voxels shown on the axial slice. Zero on the x axis indicates time of block on-set. A similar analysis conducted for Experiment 1b did not reveal any significant voxels.

Figure 2. 

Univariate analysis results. In Experiment 1a, when stimulus exposure duration was 2000 msec, all object-selective regions exhibited stronger activation for impossible compared with possible objects. (A) Activation in EVC did not differ between the two object categories. (B) In Experiment 1b, when stimulus exposure duration was 30 msec, no differences were observed in the activation patterns of possible and impossible objects. (C) The average beta weights of object-selective regions in Experiments 1a and 1b reflect stronger activation for impossible objects, but only in Experiment 1a. Error bars in all figures denote ICIs (Tryon & Lewis, 2008; Tryon, 2001). (D) A whole-brain group analysis of Experiment 1a. Significant voxels (p < .0005, uncorrected) with greater activation for impossible objects are overlaid on an axial slice from a representative participant. The Talairach coordinates were as follows: right hemisphere: x = 45.8, y = −64.9, z = −5.1; left hemisphere: x = −43.9, y = −66.8, z = −2.3. Mean activation profiles (percent signal change) for possible and impossible objects obtained from the left and right significant activated voxels shown on the axial slice. Zero on the x axis indicates time of block on-set. A similar analysis conducted for Experiment 1b did not reveal any significant voxels.

To further confirm the patterns of differences and similarities in each region in response to possible and impossible objects, we adapted a statistical method for calculating inferential confidence intervals (ICIs; Tryon & Lewis, 2008; Tryon, 2001; see also Avidan et al., 2014). This method addresses some of the difficulties typical to traditional null hypothesis testing and enables to infer statistical differences, but also, and even more critically for the present purposes, statistical equivalence. We separately calculated the 95% ICI for the response to possible and impossible objects in each ROI and then compared the obtained values across the two conditions. This analysis revealed that each of the object-selective ROIs exhibited significantly greater activation for impossible compared with possible objects whereas a statistical equivalence across these conditions was observed in EVC.

Contrary to these results, in Experiment 1b (exposure duration = 30 msec), no interaction was found between object-selective regions and the EVC [F(1, 6) = 1.3, p > .2]. ICI analysis revealed statistical equivalence for the activation pattern for the two object categories in each of the ROIs (Figure 2B). To directly compare Experiments 1a and 1b, we performed a repeated-measures ANOVA on the mean beta weights averaged across all object-selective regions, as obtained in each experiment. This analysis revealed a two-way interaction between Exposure duration and Object type [F(1, 13) = 19.08, p < .001, ηp2 = 0.59] and a main effect of Experiment [F(1, 13) = 41.2, p < .001, ηp2 = 0.76] (Figure 2C).

Note that the similar pattern of response to possible and impossible objects found in Experiment 1b could be attributed to the minimal behavioral awareness to object possibility under the brief exposure duration, and we relate to this option below. However, an alternative account for this finding could be that the short presentation duration did not enable sufficient modulation of the fMRI BOLD signal because of a floor effect. To rule out this explanation, we analyzed the average fMRI BOLD signal change in this experiment for possible and impossible objects. The results of this analysis suggest that object-selective regions were sufficiently responsive to the stimuli, thus arguing against the possibility of a floor effect. Particularly, in all object-selective regions (except for the pFs), the range of the average BOLD percent signal change compared with baseline was between 0.5% and 1%.

Lastly, to test whether other cortical regions, outside the preselected object-selective ROIs, were more activated for impossible compared with possible objects, we conducted an exploratory whole-brain analysis. This analysis was done after concatenating all the experimental runs of all the participants (separately for Experiments 1a and 1b). The results of this analysis did not reveal any additional brain regions in which the signal was selective to object possibility. Particularly, in Experiment 1a, no significant voxels were found at the level of q(FDR) < .05. Lowering the threshold to p < .0005, uncorrected, revealed two foci that correspond to the location of the inferior temporal sulcus as defined by the independent localizer, in which significant differences were observed between impossible and possible objects (Figure 2D). In agreement with the ROI analysis of Experiment 1b (30 msec), in which no differences were found for possible versus impossible objects, the whole-brain analysis of this experiment revealed no significant voxels at the level of q(FDR) < .05 as well in a more lenient, uncorrected threshold of p < .0005.

Functional Connectivity

Given that several studies proposed that 3-D structure is represented in a number of cortical regions (e.g., Orban, 2011; Welchman et al., 2005), we set to examine whether and in what way the differences between possible and impossible objects would be reflected beyond the level of the regional activation. Because we had no a priori reason to assume that some object-selective ROIs would be sensitive to object possibility while others would not, we first focused on the functional connectivity pattern in the object-selective network as a whole (and see Figure 4 for further verification of this issue). To examine the specificity of the observed effects for the object-selective network, we also examined the pattern of connectivity for these stimuli within the core face network (Haxby et al., 2000), which served as a control network.

Functional connectivity was assessed by correlating the time course of each ROI with the time courses of all other ROIs. The average correlation level observed for each pair of ROIs is plotted in separate matrices for possible and impossible objects (Figure 3A, B, D, E). The level of the correlation is indicated by the color of the cell, ranging from blue (r = 0) to red (r = 1). For clarity, the regions that constitute the object-selective network are demarcated by a black rectangle, whereas the face-selective network is demarcated by a red rectangle. To statistically compare correlations between object types, the within-network correlations were averaged and the mean correlation coefficient was subjected to a repeated-measures ANOVA with object type (possible /impossible) and network (faces/objects) as independent variables.

Figure 3. 

Functional connectivity matrices for impossible (A, D) and possible (B, E) objects. Matrices show all pairwise correlations between regions within the object network (black rectangle) and core–face network (red rectangle). The color code indicates the correlation coefficient between each pair of regions. Note that, because there were no negative correlations between ROIs, a scale of 0–1 is used for presentation purposes. A mean correlation coefficient was calculated separately for each network and each object type (C, F). Object-selective regions were more correlated with each other for impossible compared with possible objects regardless of exposure durations.

Figure 3. 

Functional connectivity matrices for impossible (A, D) and possible (B, E) objects. Matrices show all pairwise correlations between regions within the object network (black rectangle) and core–face network (red rectangle). The color code indicates the correlation coefficient between each pair of regions. Note that, because there were no negative correlations between ROIs, a scale of 0–1 is used for presentation purposes. A mean correlation coefficient was calculated separately for each network and each object type (C, F). Object-selective regions were more correlated with each other for impossible compared with possible objects regardless of exposure durations.

In Experiment 2a (2000 msec), greater correlations were observed in the object-selective network for impossible compared with possible objects (Figure 3A, B). A two-way interaction was found between network and object type [F(1, 7) = 5.57, p = .05, ηp2 = 0.44]. Planned comparisons showed that the mean correlation coefficient was higher for impossible compared with possible objects in the object-selective network [F(1, 7) = 7.78, p < .05] whereas no differences were observed in the face-selective network [F(1, 7) < 1] (Figure 3C). This result suggests that, when stimuli are presented for 2000 msec, sensitivity to the objects' 3-D structure is captured not only at a regional level (as indicated by the univariate analysis; Figure 3A) but also at the network level. Importantly, such sensitivity only exists within the object network, but not in the face network, implying that this effect was specific and related to the experimental stimuli and not to a more general attentional effect.

Next, a similar analysis was applied to the results of Experiment 2b, in which objects were presented for 30 msec. Interestingly, despite this brief exposure duration, which limits the ability to explicitly process object possibility (see the behavioral analysis below), a two-way interaction between object type and network was found [F(1, 6) = 8.41, p < .05, ηp2 = 0.58]. Planned comparisons showed that impossible objects elicited higher correlations between object-selective regions compared with possible objects [F(1, 6) = 6.14, p < .05] whereas no differences between object types were observed in the face network [F(1, 6) < 1] (Figure 3DF).

To directly compare the results of Experiments 2a and 2b, an additional analysis was conducted, in which exposure duration served as a between-subject variable. As expected, a two-way interaction between object type and network selectivity (i.e., objects vs. faces) was found [F(1, 13) = 11.9, p < .01, ηp2 = 0.47]. Importantly, no interaction was observed between exposure duration and all other factors (all ps > .2), suggesting that the differences in the connectivity patterns between object types in the object-selective network were the same across the two experiments. A main effect of exposure duration was also found, which stemmed from overall higher correlation values in the long, compared with the short, exposure duration [F(1, 13) = 6.5, p < .05, ηp2 = 0.33]. This finding is in-line with the univariate analysis that showed an overall greater activation in Experiment 1a compared with Experiment 1b.

Note that in the above analyses in which the object-selective and face-selective networks were compared, the number of ROIs was not equal across networks. Particularly, the object-selective network included five ROIs per hemisphere compared with three ROIs per hemisphere in the face-selective network. To ensure that the differences between the two networks were not related to the reduced number of ROIs in the face network, we conducted an additional bootstrapping analysis. The bootstrap consisted of 10,000 iterations in which six face-selective ROIs and six object-selective ROIs (across both hemispheres) were randomly sampled with replacement. Next, the differences in the functional connectivity between possible and impossible objects were tested within the face-selective and object-selective networks, and the differences between the networks were compared. This analysis replicated the results of the previous analyses and revealed greater functional connectivity for impossible objects compared with possible objects only in the object-selective network (rpossible = 0.73; rimpossible = 0.81, mean difference = 0.08) and not in the face-selective network (rpossible = 0.46; rimpossible = 0.47, mean difference = 0.01). To establish the statistical significance of these findings, we calculated the 95% confidence interval of the obtained bootstrap distribution of the mean differences between the face and object networks (mean differences = 0.07, p = .05, 95% CI [0.001, 0.13]). These results imply that the obtained difference between the object- and face-selective networks in terms of their functional connectivity patterns are not related to the reduced number of ROIs in the face network.

The statistical approach we have taken so far, in which we averaged the correlation coefficients across all ROIs in a given network, might conceal possible differences in the pattern of connectivity of specific ROIs comprising the object-selective network. To examine the specific correlation pattern of each ROI and their contribution to the network, we conducted a bootstrap analysis (see Methods for details). This analysis yielded a distribution of the mean correlation for each ROI. To establish the statistical significance of these findings, we calculated the 95% confidence interval of the obtained bootstrap distribution of the mean differences between impossible and possible objects. The results of this analysis are presented in Figure 4 and reveal that most ROIs (excluding the right pFs) exhibited larger correlations for impossible compared with possible objects. These findings further suggest that the greater correlations between object-selective regions for impossible compared with possible objects reflect a property of the whole network and are not driven by the correlation pattern of specific ROIs.

Figure 4. 

Bootstrap analysis of ROI connectivity. For each ROI, a distribution of the mean differences between the correlations for impossible and possible objects was created (see text for details). The red vertical lines represent the 95% confidence interval of the mean differences. The confidence interval of all ROIs (excluding right pFs) did not include 0, suggesting significant enhanced correlations for impossible objects.

Figure 4. 

Bootstrap analysis of ROI connectivity. For each ROI, a distribution of the mean differences between the correlations for impossible and possible objects was created (see text for details). The red vertical lines represent the 95% confidence interval of the mean differences. The confidence interval of all ROIs (excluding right pFs) did not include 0, suggesting significant enhanced correlations for impossible objects.

Finally, note that two regions exhibited somewhat different functional connectivity profiles compared with other regions. In the object network, the pFs was correlated to a lesser degree compared with other regions in this network. Particularly, the mean correlation coefficient of this region with other object-selective regions was r = .45 and r = .35 (for Experiments 2a and 2b, respectively), whereas the mean correlations between other object-selective regions were r = .78 and r = .68, respectively, and these differences were significant in both studies [t(5) = 7.92, p < .05; t(4) = 15.86, p < .05]. Similarly, in the face network, the pSTS was correlated to a lesser degree with the FFA and the OFA (r = .32 and r = .27 for Experiments 2a and 2b, respectively), compared with the mean correlation between the FFA and the OFA (Experiment 2a: r = .66, Experiment 2b: r = .55) [t(4) = 3.5, p < .05; t(6) = 8.02, p < .05]. We further elaborate on this finding in the Discussion.

The Relation between Univariate and Functional Connectivity Analyses

An important issue that emerges from the findings described so far has to do with the relation between the univariate and the functional connectivity results and, particularly, whether these two measures convey similar information. The results of Experiment 2b suggest otherwise: Although a clear distinction was found in the pattern of functional connectivity between object categories in both the long and short exposure durations (Experiments 2a and 2b), in the univariate analysis (Experiments 1a and 1b), such differences were only observed in the long exposure duration. To directly compare the results of the two analyses, we calculated a ratio reflecting the response difference at the level of beta weights (univariate analysis) or connectivity for the impossible compared with possible objects in object-selective network (see Methods for more details). Importantly, there was no correlation between these two measures in any of the experiments (Experiments 1a–2a, r = .02, p > .9; Experiments 1b–2b, r = .07, p > .8) implicating that these two measures are indeed distinct.

In summary, the main finding of the functional connectivity analysis is that the object-selective network exhibits sensitivity to objects' 3-D structure even following a brief exposure duration. Critically, this susceptibility was not evident in the activation profile within each ROI obtained using standard GLM univariate analysis, where the differences in the response to the two object categories were observed only during the long, but not the short, exposure duration.

Correspondence between the Behavioral and Imaging Findings

Possible–Impossible Classification

In a separate session outside the scanner, participants preformed a possible/impossible classification task (Schacter et al., 1990) either for long (n = 7) or short exposure duration (n = 6) to examine their extent of awareness to object impossibility under these two conditions. Repeated-measures ANOVA with Exposure duration (2000 msec/30 msec) and Repetition number of each stimulus along the experiment (6 levels) as independent variables was conducted (Figure 5A). As expected, this analysis revealed a main effect of Exposure duration [F(1, 11) = 15.59, p < .01, ηp2 = 0.58] (average accuracy: 83.5%, SD = 14% vs. 58.8%, SD = 7%, for exposure duration of 2000 and 30 msec, respectively). These results suggest that, during the short presentation duration, participants had minimal explicit awareness of object possibility; note however that performance level was still significantly greater than chance [t(5) = 2.92, p < .05]. Additionally, a two-way interaction was found between Exposure duration and Repetition number [F(5, 55) = 3.35, p < .05, ηp2 = 0.23]. This interaction stemmed from a significant linear trend obtained for the long-exposure duration [F(1, 11) = 19.64, p < .01] and the absence of such a trend for the short exposure duration [F(1, 11) < 1]. This analysis indicates that, although for the long presentation duration performance has improved with repeated exposures along the experiment, no such improvement occurred for the short presentation duration.

Figure 5. 

(A) Accuracy in the possible/impossible classification task as a function of the number of stimulus repetitions. Accuracy was higher in Experiment 1a (black rectangle, exposure duration = 2000 msec) compared with Experiment 1b (gray circle, exposure duration = 30 msec), regardless of stimulus repetition. Furthermore, in Experiment 1a, performance was significantly above chance level and improved throughout the experiment. In contrast, in Experiment 1b, performance was marginally above chance level (58%). Moreover, in this experiment, performance did not improve with stimulus exposure (linear fit is represented by the straight lines). (B) Accuracy in the possible/impossible classification task as a function of the number of stimulus repetition and order of presentation among naive participants. Accuracy was higher for long exposure duration (black rectangle) compared with the short exposure duration. Importantly, no differences were observed between Group 2 for which possible and impossible objects were intermixed (gray circle) and Group 3 for which objects were presented separately (white triangle). Similar to the results obtained in participants of the fMRI experiment, no improvement was observed for the short exposure duration with stimulus exposure (linear fit is represented by the straight lines). (C) Correlation between functional connectivity distinction ratio and the accuracy in the Money Road-Map Test. Greater selectivity for object category was correlated with better spatial ability.

Figure 5. 

(A) Accuracy in the possible/impossible classification task as a function of the number of stimulus repetitions. Accuracy was higher in Experiment 1a (black rectangle, exposure duration = 2000 msec) compared with Experiment 1b (gray circle, exposure duration = 30 msec), regardless of stimulus repetition. Furthermore, in Experiment 1a, performance was significantly above chance level and improved throughout the experiment. In contrast, in Experiment 1b, performance was marginally above chance level (58%). Moreover, in this experiment, performance did not improve with stimulus exposure (linear fit is represented by the straight lines). (B) Accuracy in the possible/impossible classification task as a function of the number of stimulus repetition and order of presentation among naive participants. Accuracy was higher for long exposure duration (black rectangle) compared with the short exposure duration. Importantly, no differences were observed between Group 2 for which possible and impossible objects were intermixed (gray circle) and Group 3 for which objects were presented separately (white triangle). Similar to the results obtained in participants of the fMRI experiment, no improvement was observed for the short exposure duration with stimulus exposure (linear fit is represented by the straight lines). (C) Correlation between functional connectivity distinction ratio and the accuracy in the Money Road-Map Test. Greater selectivity for object category was correlated with better spatial ability.

Note that the nature of the experimental design employed here limits the ability to directly correlate the fMRI results and the behavioral ability to classify objects as possible or impossible as assessed outside the scanner. Particularly, the behavioral experiment employed a between-group design to avoid possible carryover in performance between the two conditions, which, as expected, elicited marked differences in accuracy. Thus, any within-group correlation analysis is bound to be performed on a relatively small number of participants (7 or 8); hence, we avoid such analysis, which is prone to spurious correlations.

Note that one important difference between the behavioral experiments described above and the fMRI sessions is that in the former possible and impossible objects were presented in an intermixed fashion whereas in the latter possible and impossible objects were presented in separate blocks (GLM session) and even in separate runs (functional connectivity session). One plausible outcome of such repeated presentation of the same object category (possible or impossible) is that it might provide accumulating evidence and could lead to enhanced awareness to object possibility even for the short exposure duration. To rule out this possible account that might have affected the fMRI findings, we conducted two additional behavioral experiments outside the scanner in which participants were asked to perform a possible/impossible classification task. In the first experiment, 22 naive participants were assigned to three groups: in Group 1 (8 participants) possible and impossible objects were intermixed and were presented for 2000 msec, in Group 2 (8 participants) possible and impossible objects were again intermixed but presented for 30 msec, and in Group 3 (6 participants) possible and impossible objects were separated and were presented for 30 msec (similar to the design of the functional connectivity sessions). Importantly, the results of the possible/impossible classification task resembled the behavioral performance obtained from the fMRI participants. Specifically, we found superior performance for Group 1 compared with the two other groups [F(1, 19) = 26.4, p < .01], whereas no differences between Groups 2 and 3 were obtained [F(1, 19) < 1]. Most importantly, the interaction between exposure duration and repetition number was replicated [F(10, 95) = 2.4, p < .05, ηp2 = 0.2] such that linear improvement was found for Group 1 [F(1, 19) = 11.9, p < .01] but not for Group 2 [F(1, 19) = 1, p > .2] or Group 3 in which possible and impossible objects were presented in an intermixed fashion [F(1, 19) = 1.7, p > .2] (Figure 5B).

Note that one limitation of these results is that they are based on between-subject designs. Thus, the goal of the second behavioral experiment was to provide further support for the notion that repeated presentation of one object category did not facilitate the awareness to object possibility by utilizing a within-subject design to enhance the experimental statistical power. Twelve naive participants performed the possible/impossible classification task under two conditions: the block design—alternating blocks of possible and impossible objects (similarly to Experiment 1a), and the functional connectivity design—separate runs of possible and impossible objects (similarly to Experiment 2b). The order of the conditions was counterbalanced across participants, and in both conditions, stimuli were presented for 30 msec. The results showed that participants had limited awareness for object impossibility (blocks: 53%, separate runs: 55%), with no statistical differences between these conditions [t(11) < 1, p = .5].

Taken together, these results undermine the notion that repeated presentations of objects from specific category provide accumulating evidence for object possibility and might underlie the pattern of the fMRI results. Rather, these findings indicate that exposure duration is the critical factor in determining object possibility.

Assessment of Participants' General Spatial Ability

Finally, we explored whether general spatial ability, as assessed by the Money Road-Map Test (Money et al., 1965), is related to the pattern of fMRI response obtained in our experiments. Because there were no significant differences in the pattern of functional connectivity between Experiments 2a and 2b in the distinction between impossible and possible objects, we conducted the correlation analysis between the behavioral scores on the Money Road-Map Test and the collapsed results of these two fMRI experiments. This analysis revealed a significant positive correlation between spatial ability and functional connectivity selectivity ratio (r = .61, p < .05; Figure 5C). In particular, higher scores in the Money Road-Map Test, indicating better spatial ability, were correlated with greater selectivity for impossible compared with possible objects as reflected in functional connectivity.

DISCUSSION

This study was aimed to examine how the visual system represents the 3-D structure of novel objects and, in particular, to reveal the extent to which the visual system is sensitive to distortions in spatial information. To this end, we utilized possible and impossible objects and analyzed the activation profiles within object-selective regions as well as the interregional correlations between these regions. Our findings show that spatial incoherency modulated the activation profile within these ROIs as well as the functional connectivity between ROIs, but in a distinct manner. Hence, we present evidence for dissociable effects of spatial incoherency at the within-region level and at the between-region level of neural connectivity.

Within-region Activation

The univariate analysis revealed that impossible objects elicited greater fMRI activation but only during the long exposure duration, which presumably allowed sufficient processing of object possibility. One source for this enhanced activation could be related to the incoherency of the 3-D interpretation associated with impossible objects (e.g., Schacter et al., 1990). Particularly, it has been demonstrated that several cortical regions are sensitive to the 3-D structure of objects and to the perception of stimuli as 3-D (Georgieva et al., 2008; Murray & He, 2006; Moore & Engel, 2001). Hence, the current results may extend these findings and suggest that object-selective regions may be highly sensitive to objects' 3-D structure and that irregularity of this structure may require more processing resources that would be manifested in enhanced activation.

Interestingly, similar levels of fMRI BOLD signal were found for the two object categories when stimuli were briefly presented. As is evident from the behavioral experiments, in this exposure duration, participants only had limited explicit awareness of object possibility. However, the experimental design did not allow us to directly correlate the behavioral performance in the possible/impossible classification task with the fMRI response. Nonetheless, our findings may indirectly point to an association between the neural activation and the behavioral outcome. This assumption is in-line with previous fMRI studies, which demonstrated that activation in object-selective regions is related to conscious perceptual experience (Avidan et al., 2002; Grill-Spector, et al., 2000). Importantly, in EVC no response differences were obtained for the two object types, which are physically very similar, in either the short or long exposure duration, further suggesting that the signal in high-level visual regions is related to the perception of these stimuli and not merely to their physical attributes.

Interregional Correlations

The functional connectivity analysis revealed a differential pattern of responses for possible and impossible objects. This distinct response is suggested to be associated with the processing of the impossibility embedded in the 3-D structure. Importantly, this differential response was evident regardless of exposure duration and awareness to object possibility. Particularly, higher correlations between object-selective regions were found for impossible objects compared with possible ones, even following short exposure duration of 30 msec. Note that under this condition participants gained only minimal explicit knowledge about object possibility, as evident from the behavioral session. Nevertheless, this minimal awareness might have been sufficient to mediate the changes in the network dynamics, which preceded changes in the within ROI level.

This result provides a novel quantitative measure indicating the distributed fashion in which 3-D information is represented in the visual cortex. Particularly, previous studies already showed that different regions in the dorsal and ventral visual streams are sensitive to objects' 3-D structure. The critical finding in this study is that the distributed nature of 3-D representation is not merely deduced from the sensitivity of each ROI to 3-D information, but rather from the synchronous activation pattern of these ROIs. Moreover, the absence of correlation between functional connectivity and within-region activity in the current investigation may suggest that these two measures convey different information about objects' 3-D structure.

But what is the significance of such enhanced functional connectivity for impossible objects? One plausible account is that such enhanced connectivity is related to the spatial incoherency of the 3-D structure of impossible objects, which requires further processing within the object-selective network. Note that despite the growing fMRI literature on network analysis, the interpretation of enhanced or reduced functional connectivity is still largely unclear. For example, Santangelo and Macaluso (2013) showed that memory performance was correlated with object saliency and that this advantage was accompanied by increased functional connectivity between posterior parietal cortex and medial-temporal lobe. On the other hand, Ricciardi and colleagues have shown that administration of cholinergic muscarinic agonist improved selective attention by reducing the functional connectivity between ventral visual regions and multiple cortical sites (Ricciardi, Handjaras, Bernardi, Pietrini, & Furey, 2012). Thus, the enhancement versus reduction in functional connectivity could not be considered as isolated findings, but rather should be thought of as relative effects that interact with behavioral performance. In this study, enhanced functional connectivity for impossible compared with possible objects is suggested to be related to spatial abilities. Specifically, the extent to which the correlation between object-selective regions was dissociated between object types was correlated with participant's general spatial abilities. Hence, the ability of a given network to change its dynamics in response to specific stimuli is suggested to have behavioral significance.

Two ROIs exhibited a distinct correlation pattern. Specifically, the pFs, an object-selective region that is located laterally to the PPA and medially to FFA (Hasson et al., 2003), was less correlated with other object-selective regions and exhibited less sensitivity to object possibility. Several studies demonstrated that this region may have different functional properties compared with other object-selective regions such as the LOC. For example, Drucker and Aguirre (2009) argued that information on visual similarity is encoded in a within-voxel scale in the ventral LOC (i.e., pFs) whereas the lateral component of the LOC demonstrated a coarse (across-voxel) sensitivity to object similarity. It is not clear, however, why, in the context of the current study, this region was not as correlated as other object-selective regions. The second ROI that exhibited a distinct correlation pattern is the pSTS, which is part of the face–core network. This region did not exhibit a correlation with the face-selective regions as well as the object-selective regions, a pattern that may reflect the distinct anatomical connections of this region relative to other face-selective regions (Pyles, Verstynen, Schneider, & Tarr, 2013; Ethofer, Gschwind, & Vuilleumier, 2011).

The Association between the Univariate Analysis and the Pattern of Functional Connectivity

The current study characterizes the neural response for impossible objects at two distinct levels, the within-region activity profile and the interregional synchronization. Nonetheless, the association between these two measures is still unclear. Critically, it has been suggested that modulation of functional connectivity could result from modification of the neural response (Friston, 2011). Thus, one could argue that the findings of stronger correlations between object-selective regions in Experiment 2a were induced by the greater activation elicited by this object category. However, two main findings counteract this potential account. First, the results of Experiment 2b show that impossible objects elicited higher correlation even in the absence of differences in the univariate analysis. Second, the distinction between the two object types in the univariate analysis (Experiment 1a) was not correlated with the distinction between the object types in the functional connectivity analysis (Experiment 2a). These findings suggest that, in the current study, the univariate analysis and the functional connectivity analysis reflect dissociable aspects of the neural response.

This conclusion is also in-line with a recent study that used electrocorticographic signals to decode visual object categories. In this study, the decoding performance, based on the spectral power and phase of individual electrode, was compared with the decoding performance that was based on the temporal correlations between electrodes. The results showed that the temporal correlations served as better predictors compared with other features, but that the best prediction was achieved by a combination of the features. These findings suggest that the between-region connectivity and the within-region activation may carry different aspects of neural information about object category. Additionally, this study showed that the decoding performance based on the between-electrode correlations began to rise earlier in time compared with the decoding based on the spectral power of each electrode (Majima et al., 2014). This finding may account for the differences observed here between functional connectivity and univariate analysis for exposure duration of 30 msec. Nonetheless, the temporal resolution of fMRI does not allow us to directly assess the temporal sequence of the neural sensitivity to objects' 3-D structure. Future studies, using time-sensitive imaging tools such as ERP or MEG should investigate whether, as predicted from the present results, sensitivity to objects' 3-D structure would emerge first at the level of connectivity and only later would be expressed at the level of the isolated response within object-selective regions.

Conclusions

The current study was focused on the decoding of 3-D information of object structure, one of the most fundamental aspects of visual perception. Our results show that a cortical object-selective network in the human brain is susceptible to this type of information, as evident by modulations of the fMRI responses at both the within-region and between-region levels. Importantly, our data strongly suggest that these two levels of processing can be dissociated and, therefore, imply that the network connectivity pattern and the overall pattern of activation within each of the brain regions that constitute the network may have different roles in the representation of objects' 3-D structure.

Reprint requests should be sent to Galia Avidan, Department of Psychology, Ben-Gurion University, The Abraham Ben David Ohayon Behavioral Sciences Complex, P.O.B. 653, Beer Sheva, Israel, 8410501, or via e-mail: galiaa@bgu.ac.il or to Erez Freud via e-mail: erezfreud@gmail.com.

REFERENCES

REFERENCES
Avidan
,
G.
,
Harel
,
M.
,
Hendler
,
T.
,
Ben-Bashat
,
D.
,
Zohary
,
E.
, &
Malach
,
R.
(
2002
).
Contrast sensitivity in human visual areas and its relationship to object recognition.
Journal of Neurophysiology
,
87
,
3102
3116
.
Avidan
,
G.
,
Tanzer
,
M.
,
Hadj-Bouziane
,
F.
,
Liu
,
N.
,
Ungerleider
,
L. G.
, &
Behrmann
,
M.
(
2014
).
Selective dissociation between core and extended regions of the face processing network in congenital prosopagnosia.
Cerebral Cortex
,
24
,
1565
1578
.
Behrmann
,
M.
, &
Plaut
,
D. C.
(
2013
).
Distributed circuits, not circumscribed centers, mediate visual recognition.
Trends in Cognitive Sciences
,
17
,
210
219
.
Biswal
,
B.
,
Zerrin Yetkin
,
F.
,
Haughton
,
V. M.
, &
Hyde
,
J. S.
(
1995
).
Functional connectivity in the motor cortex of resting human brain using echo-planar MRI.
Magnetic Resonance in Medicine
,
34
,
537
541
.
Dinstein
,
I.
,
Pierce
,
K.
,
Eyler
,
L.
,
Solso
,
S.
,
Malach
,
R.
,
Behrmann
,
M.
,
et al
(
2011
).
Disrupted neural synchronization in toddlers with autism.
Neuron
,
70
,
1218
1225
.
Drucker
,
D. M.
, &
Aguirre
,
G. K.
(
2009
).
Different spatial scales of shape similarity representation in lateral and ventral LOC.
Cerebral Cortex
,
19
,
2269
2280
.
Ethofer
,
T.
,
Gschwind
,
M.
, &
Vuilleumier
,
P.
(
2011
).
Processing social aspects of human gaze: A combined fMRI-DTI study.
Neuroimage
,
55
,
411
419
.
Faillenot
,
I.
,
Decety
,
J.
, &
Jeannerod
,
M.
(
1999
).
Human brain activity related to the perception of spatial features of objects.
Neuroimage
,
10
,
114
124
.
Flament
,
M. F.
,
Koby
,
E.
,
Rapoport
,
J. L.
,
Berg
,
C. J.
,
Zahn
,
T.
,
Cox
,
C.
,
et al
(
1990
).
Childhood obsessive-compulsive disorder: A prospective follow-up study.
Journal of Child Psychology and Psychiatry
,
31
,
363
380
.
Freud
,
E.
,
Avidan
,
G.
, &
Ganel
,
T.
(
2013
).
Holistic processing of impossible objects: Evidence from Garner's speeded-classification task.
Vision Research
,
93
,
10
18
.
Freud
,
E.
,
Avidan
,
G.
, &
Ganel
,
T.
(
2015
).
The highs and lows of object impossibility: Effects of spatial frequency on holistic processing of impossible objects.
Psychonomic Bulletin & Review
,
22
,
297
306
.
Freud
,
E.
,
Ganel
,
T.
, &
Avidan
,
G.
(
2013
).
Representation of possible and impossible objects in the human visual cortex: Evidence from fMRI adaptation.
Neuroimage
,
64
,
685
692
.
Friston
,
K. J.
(
2011
).
Functional and effective connectivity: A review.
Brain Connectivity
,
1
,
13
36
.
Gauthier
,
I.
,
Tarr
,
M. J.
,
Moylan
,
J.
,
Skudlarski
,
P.
,
Gore
,
J. C.
, &
Anderson
,
A. W.
(
2000
).
The fusiform “face area” is part of a network that processes faces at the individual level.
Journal of Cognitive Neuroscience
,
12
,
495
504
.
Georgieva
,
S. S.
,
Todd
,
J. T.
,
Peeters
,
R.
, &
Orban
,
G. A.
(
2008
).
The extraction of 3-D shape from texture and shading in the human brain.
Cerebral Cortex
,
18
,
2416
2438
.
Grill-Spector
,
K.
,
Kushnir
,
T.
,
Hendler
,
T.
, &
Malach
,
R.
(
2000
).
The dynamics of object-selective activation correlate with recognition performance in humans.
Nature Neuroscience
,
3
,
837
843
.
Hasson
,
U.
,
Harel
,
M.
,
Levy
,
I.
, &
Malach
,
R.
(
2003
).
Large-scale mirror-symmetry organization of human occipito-temporal object areas.
Neuron
,
37
,
1027
1041
.
Haxby
,
J. V.
,
Gobbini
,
M. I.
,
Furey
,
M. L.
,
Ishai
,
A.
,
Schouten
,
J. L.
, &
Pietrini
,
P.
(
2001
).
Distributed and overlapping representations of faces and objects in ventral temporal cortex.
Science
,
293
,
2425
2430
.
Haxby
,
J. V.
,
Hoffman
,
E. A.
, &
Gobbini
,
M. I.
(
2000
).
The distributed human neural system for face perception.
Trends in Cognitive Sciences
,
4
,
223
233
.
Honey
,
C. J.
,
Sporns
,
O.
,
Cammoun
,
L.
,
Gigandet
,
X.
,
Thiran
,
J. P.
,
Meuli
,
R.
,
et al
(
2009
).
Predicting human resting-state functional connectivity from structural connectivity.
Proceedings of the National Academy of Sciences, U.S.A.
,
106
,
2035
2040
.
Kanwisher
,
N.
,
McDermott
,
J.
, &
Chun
,
M. M.
(
1997
).
The fusiform face area: A module in human extrastriate cortex specialized for face perception.
Journal of Neuroscience
,
17
,
4302
4311
.
Konen
,
C. S.
,
Behrmann
,
M.
,
Nishimura
,
M.
, &
Kastner
,
S.
(
2011
).
The functional neuroanatomy of object agnosia: A case study.
Neuron
,
71
,
49
60
.
Kornblith
,
S.
,
Cheng
,
X.
,
Ohayon
,
S.
, &
Tsao
,
D. Y.
(
2013
).
A network for scene processing in the macaque temporal lobe.
Neuron
,
79
,
766
781
.
Kourtzi
,
Z.
,
Erb
,
M.
,
Grodd
,
W.
, &
Bulthoff
,
H. H.
(
2003
).
Representation of the perceived 3-D object shape in the human lateral occipital complex.
Cerebral Cortex
,
13
,
911
920
.
Majima
,
K.
,
Matsuo
,
T.
,
Kawasaki
,
K.
,
Kawai
,
K.
,
Saito
,
N.
,
Hasegawa
,
I.
,
et al
(
2014
).
Decoding visual object categories from temporal correlations of ECoG signals.
Neuroimage
,
90c
,
74
83
.
Malach
,
R.
,
Reppas
,
J. B.
,
Benson
,
R. R.
,
Kwong
,
K. K.
,
Jiang
,
H.
,
Kennedy
,
W. A.
,
et al
(
1995
).
Object-related activity revealed by functional magnetic resonance imaging in human occipital cortex.
Proceedings of the National Academy of Sciences, U.S.A.
,
92
,
8135
8139
.
Mapstone
,
M.
,
Steffenella
,
T. M.
, &
Duffy
,
C. J.
(
2003
).
A visuospatial variant of mild cognitive impairment getting lost between aging and AD.
Neurology
,
60
,
802
808
.
Money
,
J.
,
Walker
,
H. T.
, &
Alexander
,
D.
(
1965
).
A standardized road-map test of direction sense: Manual
.
Baltimore, MD
:
Johns Hopkins Press
.
Moore
,
C.
, &
Engel
,
S. A.
(
2001
).
Neural response to perception of volume in the lateral occipital complex.
Neuron
,
29
,
277
286
.
Murray
,
S. O.
, &
He
,
S.
(
2006
).
Contrast invariance in the human lateral occipital complex depends on attention.
Current Biology
,
16
,
606
611
.
Orban
,
G. A.
(
2011
).
The extraction of 3D shape in the visual system of human and nonhuman primates.
Annual Review of Neuroscience
,
34
,
361
388
.
Penrose
,
L. S.
, &
Penrose
,
R.
(
1958
).
Impossible objects: A special type of visual illusion.
British Journal of Psychology
,
49
,
31
33
.
Pyles
,
J. A.
,
Verstynen
,
T. D.
,
Schneider
,
W.
, &
Tarr
,
M. J.
(
2013
).
Explicating the face perception network with white matter connectivity.
PloS One
,
8
,
e61611
.
Regolin
,
L.
,
Rugani
,
R.
,
Stancher
,
G.
, &
Vallortigara
,
G.
(
2011
).
Spontaneous discrimination of possible and impossible objects by newly hatched chicks.
Biology Letters
,
7
,
654
657
.
Ricciardi
,
E.
,
Handjaras
,
G.
,
Bernardi
,
G.
,
Pietrini
,
P.
, &
Furey
,
M. L.
(
2012
).
Cholinergic enhancement reduces functional connectivity and BOLD variability in visual extrastriate cortex during selective attention.
Neuropharmacology
,
64
,
305
313
.
Santangelo
,
V.
, &
Macaluso
,
E.
(
2013
).
Visual salience improves spatial working memory via enhanced parieto-temporal functional connectivity.
Journal of Neuroscience
,
33
,
4110
4117
.
Schacter
,
D. L.
,
Cooper
,
L. A.
, &
Delaney
,
S. M.
(
1990
).
Implicit memory for unfamiliar objects depends on access to structural descriptions.
Journal of Experimental Psychology: General
,
119
,
5
24
.
Shuwairi
,
S. M.
(
2009
).
Preference for impossible figures in 4-month-olds.
Journal of Experimental Child Psychology
,
104
,
115
123
.
Shuwairi
,
S. M.
,
Albert
,
M. K.
, &
Johnson
,
S. P.
(
2007
).
Discrimination of possible and impossible objects in infancy.
Psychological Science
,
18
,
303
307
.
Smith
,
S. M.
,
Miller
,
K. L.
,
Salimi-Khorshidi
,
G.
,
Webster
,
M.
,
Beckmann
,
C. F.
,
Nichols
,
T. E.
,
et al
(
2011
).
Network modelling methods for fMRI.
Neuroimage
,
54
,
875
891
.
Troiani
,
V.
,
Stigliani
,
A.
,
Smith
,
M. E.
, &
Epstein
,
R. A.
(
2014
).
Multiple object properties drive scene-selective regions.
Cerebral Cortex
,
24
,
883
897
.
Tryon
,
W. W.
(
2001
).
Evaluating statistical difference, equivalence, and indeterminacy using inferential confidence intervals: An integrated alternative method of conducting null hypothesis statistical tests.
Psychological Methods
,
6
,
371
386
.
Tryon
,
W. W.
, &
Lewis
,
C.
(
2008
).
An inferential confidence interval method of establishing statistical equivalence that corrects Tryon's (2001) reduction factor.
Psychological Methods
,
13
,
272
277
.
Vingerhoets
,
G.
,
Lannoo
,
E.
, &
Bauwens
,
S.
(
1996
).
Analysis of the Money Road-Map Test performance in normal and brain-damaged subjects.
Archives of Clinical Neuropsychology
,
11
,
1
9
.
Wang
,
K.
,
Liang
,
M.
,
Wang
,
L.
,
Tian
,
L.
,
Zhang
,
X.
,
Li
,
K.
,
et al
(
2007
).
Altered functional connectivity in early Alzheimer's disease: A resting-state fMRI study.
Human Brain Mapping
,
28
,
967
978
.
Wang
,
L.
,
Zang
,
Y.
,
He
,
Y.
,
Liang
,
M.
,
Zhang
,
X.
,
Tian
,
L.
,
et al
(
2006
).
Changes in hippocampal connectivity in the early stages of Alzheimer's disease: Evidence from resting state fMRI.
Neuroimage
,
31
,
496
504
.
Welchman
,
A. E.
,
Deubelius
,
A.
,
Conrad
,
V.
,
Bulthoff
,
H. H.
, &
Kourtzi
,
Z.
(
2005
).
3D shape perception from combined depth cues in human visual cortex.
Nature Neuroscience
,
8
,
820
827
.