Abstract

Object individuation and identification are 2 key processes involved in representing visual information in short-term memory (VSTM). Individuation involves the use of spatial and temporal cues to register an object as a distinct perceptual event relative to other stimuli, whereas object identification involves extraction of featural and related conceptual properties of a stimulus. Together, individuation and identification provide the “what,” “where,” and “when” of visual perception. In the current study, we asked whether individuation and identification processes are underpinned by distinct neural substrates, and to what extent brain regions that reflect these 2 operations are consistent across encoding, maintenance, and retrieval stages of VSTM. We used functional magnetic resonance imaging to identify brain regions that represent the number of objects (individuation) and/or object features (identification) in an array. Using univariate and multivariate analyses, we found substantial overlap between these 2 operations in the brain. Moreover, we show that regions supporting individuation and identification vary across distinct stages of information processing. Our findings challenge influential models of multiple-object encoding in VSTM, which argue that individuation and identification are underpinned by a limited set of nonoverlapping brain regions.

Introduction

Many everyday activities, like searching for a stapler on a cluttered desk, rely on our ability to simultaneously isolate and identify multiple visual objects. To complete such a task, an observer must use spatial and temporal information to register each object as distinct (object individuation) and bind its features into a coherent form (object identification; Kahneman et al. 1992; Chun 1997; Pylyshyn 1989; Xu and Chun 2009). It is currently unclear how these processes are represented in the brain. In their “neural object file theory,” Xu and Chun (2009) predict that the inferior intra-parietal sulcus (iIPS) plays a key role in object individuation, whereas the superior intra-parietal sulcus (sIPS) and lateral occipital complex (LOC) are involved in object identification (Xu 2007; Xu and Chun 2007; Xu 2008, 2009; Jeong and Xu 2013). Here, we test Xu and colleagues' prediction of a strict functional dissociation between the contributions of these and other regions to object individuation and object identification.

In a pioneering study, Xu (2009) provided evidence for dissociation between object individuation and identification within a single experiment. While undergoing functional magnetic resonance (fMRI) imaging, participants were briefly presented with a sample display consisting of 1 object, 4 identical objects, or 4 different objects (see Fig. 1). After a short delay, a single test item appeared centrally, and participants' task was to indicate whether the test identity matched 1 of the sample identities. Xu observed reduced blood-oxygen-level-dependent (BOLD) activity in the iIPS for 1-object displays, relative to both 4-identical-object and 4-different-object displays, which did not differ from each other. Conversely, the sIPS and LOC responded similarly for 1-object and 4-identical-object displays, but both had reduced BOLD activity relative to 4-different-object displays. Consistent with the neural object file theory (Xu and Chun 2009), these findings demonstrated that iIPS is sensitive to the number of objects in a display requiring individuation, whereas the sIPS and LOC respond to the number of distinct object identities requiring identification.

Figure 1.

A schematic representation of the short-term memory paradigm. Each trial began with the auditory presentation of 2 letters, followed by a sample display consisting of 1 object, 4 identical objects, or 4 different objects. After an extended delay, observers identified whether a single test shape was the same as, or different from, 1 of the sample shapes, both in terms of its identity and location. The same judgment was also made for 2 subsequent test letters. The secondary auditory memory task was included to prevent participants from using verbal rehearsal strategies to remember the shape stimuli.

Figure 1.

A schematic representation of the short-term memory paradigm. Each trial began with the auditory presentation of 2 letters, followed by a sample display consisting of 1 object, 4 identical objects, or 4 different objects. After an extended delay, observers identified whether a single test shape was the same as, or different from, 1 of the sample shapes, both in terms of its identity and location. The same judgment was also made for 2 subsequent test letters. The secondary auditory memory task was included to prevent participants from using verbal rehearsal strategies to remember the shape stimuli.

A potential shortcoming of Xu's (2009) study, and related empirical work (Jeong and Xu 2013; e.g., Xu 2007; Xu and Chun 2007; Xu 2008) upon which the neural object file theory is based, is that brain activity was analyzed within occipital and parietal regions only, thus potentially missing contributions from other brain regions. Further, these studies employed univariate analyses of the fMRI data, which are less sensitive to small changes in patterns of BOLD response that can emerge when activity is analyzed across groups of voxels (Haynes and Rees, 2006; Kamitani and Tong 2005). Our recent work using multi-voxel pattern analysis (MVPA) suggests that temporal individuation—registering objects as distinct perceptual events based on when they appear in “time”—recruits a broad set of frontal, parietal, and occipital regions (Naughtin et al. 2014). Based on this work, we predicted that such widespread networks of brain activity might also emerge from MVPA analyses of fMRI data obtained for displays in which objects must be individuated and identified in the “spatial” domain.

Previous work on the neural object file theory has been also limited to paradigms with short retention internals (Xu 2009), meaning that these studies have not been able to examine individuation and identification during distinct visual short-term memory (VSTM) stages. It is important to consider the role of different processing stages, given VSTM has been hypothesized to involve encoding, maintenance, and retrieval operations (e.g., Cohen et al. 1997; Courtney et al. 1997). Although Jeong and Xu (2013; also see Xu and Chun 2006) did use a long-delay VSTM paradigm and suggested that the iIPS, sIPS, and LOC are involved in individuation and identification processes during the encoding phase and beyond, their contrasts differed in both the number of objects presented and the number object identities. Thus, they could not unambiguously compare individuation and identification processes.

Here, we adopted a similar logic and approach to that used by Xu (2009), but with several important modifications, to test whether object individuation and identification can be dissociated in the brain (as suggested in the neural object file theory, Xu 2009; Xu and Chun 2009). We analyzed activity across a broad set of brain regions using both univariate and MVPA approaches and compared the role of each region across encoding, maintenance, and retrieval stages of VSTM. In addition, unlike Xu's (2009) paradigm, our task required participants to remember both the identity and spatial location of all the objects in the display; this task ensured that both identification and individuation operations were engaged. By using both analytic approaches, we could first attempt to conceptually replicate the original univariate results from Xu (2009) under different task conditions and also determine whether other brain regions show evidence of object identification or individuation using the more sensitive MVPA approach.

Materials and Methods

Participants

We recruited 21 volunteers to participate in the experiment (12 females, mean age = 25.7 years, SD = 3.4 years). Data from 1 participant were excluded due to excessive head motion and from another due to a technical error that meant behavioral responses were not recorded. A post hoc power analysis confirmed that a minimum sample size of 10 was sufficient to have an 80% chance of detecting similar effects (Cohen's fs > 1.13) to those reported by Xu (2009). Participants had normal or corrected-to-normal vision, gave informed consent, and were financially reimbursed for their time ($15/h). The University of Queensland ethics committee approved the experimental protocol.

Design and Stimuli

Stimuli consisted of 8 distinct shapes derived from Xu and Chun (2006). Each shape was presented in black on a light gray background (3.1°× 3.1° of visual angle). Stimuli could appear in 1 of 8 dark gray placeholders (3.4° × 3.4°) arranged in a circular array, with each placeholder presented 5.3° from fixation. Placeholders were included to prevent grouping between closely presented objects (see Fig. 1; Xu 2009). The experiment was programed in MATLAB using the Psychophysics Toolbox (Brainard 1997; Pelli 1997).

Procedure

We used an STM paradigm, based on that of Xu (2009), to identify brain areas that selectively code for the number of objects (individuation) and/or number of object features (identification; see Fig. 1). This paradigm had 2 components: a primary VSTM task and a secondary verbal memory task. Each trial began with the auditory presentation of 2 letters via headphones, and participants were asked to rehearse these letters throughout the trial. The verbal memory task was included to prevent participants from using verbal rehearsal strategies for the visual stimuli (Todd and Marois 2004). After a brief inter-stimulus interval, a sample display consisting of 1 object, 4 identical objects, or 4 different objects was presented for 200 ms. Participants were instructed to remember the locations and identities of the sample objects during a subsequent 12.8-s retention interval. We used a slightly longer retention interval than is typically employed in slow event-related VSTM paradigms (e.g., see Harrison and Tong 2009; Serences et al. 2009; Emrich et al. 2013) to ensure that we could clearly distinguish between each stage of VSTM.

A single test shape appeared in 1 of the placeholders after the retention interval. The participants' task was to identify whether this test shape matched or mismatched 1 of the previously presented sample shapes, both in terms of its identity “and” location. It is important to note that a correct response on this task required that participants successfully individuate and identify each of the sample items, rather than merely their identity (cf. Xu and Chun 2007; Xu 2009, 2010; Jeong and Xu 2013). Thus, even though the 4-identical-object displays only contained a single identity, participants still had to individuate where each object appeared to produce a correct response. Two test letters were then presented, and participants made the same match/mismatch response for these letters, relative to the 2 sample letters presented at the beginning of the trial. Only response accuracy was emphasized, but participants had a fixed interval of 2 s in which to make each of their responses. The test shape and letter matched or mismatched 1 of the sample items an equal number of times.

Each trial was followed by a 12-s fixation period. This experimental design allowed us to isolate sources of BOLD activity previously associated with encoding, maintenance, and retrieval of the visual information (e.g., see Harrison and Tong 2009; Serences et al. 2009; Emrich et al. 2013). We chose these labels for the different time periods because previous studies have used them to describe the distinct phases of activity observed across long retention intervals. A more theoretically neutral approach might be to label the stages as “early,” “middle,” and “late,” but we adopt the former set of labels to be consistent with prior VSTM studies (e.g., Todd and Marois 2004; Xu and Chun 2006).

Participants completed 6 practice trials outside the scanner, followed by 8 scanning runs of 12 test trials. We only provided feedback for incorrect responses on practice trials, in which the central fixation square briefly turned red. Each run contained an equal number of trials per trial type. Trial types were presented in a randomized order, and the selection of stimuli and their locations were randomized across trials. Preliminary pilot testing ensured that participants could complete the task with a high level of accuracy (>80%). We made sure participants could perform this task well, as only correct trials were included in the fMRI analyses.

fMRI Acquisition

We acquired anatomical and functional images using a 3T Siemens Trio MRI scanner (Erlangen) and a 12-channel head coil. Participants laid supine in the scanner and viewed the visual display via a rear-projection mirror. Functional T2*-weighted images were acquired parallel to the AC–PC plane using a GRE EPI sequence (TR = 2 s, TE = 25 ms, FA = 90°, FOV = 192 × 192, matrix = 64 × 64, in-plane resolution = 3 × 3 mm). Each volume consisted of 33 slices with a thickness of 3 mm and a 0.3-mm inter-slice gap. In the middle of the session, we collected a T1-weighted anatomical image using an MPRAGE sequence (TR = 1.9 s, TE = 2.32 ms, FA = 9°, FOV = 192 × 230 × 256, resolution = 1 mm3). We synchronized the stimulus presentation with the acquisition of functional volumes. Each run consisted of 184 volumes, including an 8-s dummy fixation block presented at the start of the run.

fMRI Analyses

We conducted preprocessing, univariate analyses, and MVPA with Brain Voyager QX 2.4 (Brain Innovation) and custom MATLAB code. We used both univariate and MVPA techniques to test for overall differences in BOLD amplitude and changes in the spatial patterns of activity across groups of voxels for each condition, respectively. There were 2 key comparisons of interest: First, to identify regions that reflect object individuation, we compared activity between displays containing 1 object and those containing 4 identical objects. Regions exclusively associated with this process should be sensitive to the episodic properties of objects (i.e., the number of objects in the display), but not to object identity (since this remained constant across the 2 display types). Second, to isolate regions that support object identification, we compared activity for displays containing 4 identical objects with those containing 4 different objects. If a given brain region specifically contributes to object identification, it should show a difference in activity in response to the number of distinct identities present, rather than to the number of objects per se. These were the same comparisons previously used by Xu (2009).

Preprocessing

Data preprocessing steps consisted of 3D motion correction (with all functional images aligned to the first image), slice-scan time correction, high-pass temporal filtering (3 cycles per run), and Talairach space transformation (Talairach and Tourmoux 1988). We did not apply spatial smoothing to the data to preserve fine-grained changes in activity for MVPA (as described in detail later).

Regions of Interest

All regions of interest (ROIs) were defined anatomically using mean Talairach coordinates from other relevant published studies. We used coordinates from Xu and Chun (2006) to define the iIPS, sIPS, and LOC, which have previously been implicated in object individuation and identification (e.g., Xu 2009; Xu and Chun 2009). To explore the role of other regions, we also isolated a set of frontal, parietal, and occipital regions involved in higher-level resource-limited processes such as response selection, decision making, and encoding (Szametitat et al. 2002; Heekeren et al. 2004; Dux et al. 2006; Dux et al. 2009; Tombu et al. 2011; Naughtin et al. 2014). Talairach coordinates for these regions were derived from our previous published studies (Dux et al. 2009; Naughtin et al. 2014). Since the sIPS and superior parietal lobule (SPL) ROIs overlapped in the majority of participants, we averaged the results across these 2 regions (referred to as sIPS/SPL).

Regions of interest were defined by a 11-mm3 cube for the univariate analyses (we used the same number of voxels in each ROI as Xu 2009) and a 15-mm3 or 21-mm3 cube for MVPA. In both cases, the ROI cube was centered on the mean Talairach coordinates. We defined each ROI using a larger number of voxels for MVPA, compared with the univariate analyses, as it is conventional to use larger ROIs in the former to provide increased variability across voxels (e.g., Kamitani and Tong 2005; Harrison and Tong 2009; Gallivan et al. 2011; Oosterhof et al. 2012). As the sensitivity of MVPA depends upon the number of voxels included in the analysis, we used 2 different ROI sizes for MVPA—rather than choosing an arbitrary ROI size—to ensure the results were reliable, regardless of the number of voxels included in the classification analysis (Spiridon and Kanwisher 2002; Carp et al. 2011; Naughtin et al. 2014). For simplicity, we report the MVPA results from the 21-mm3 ROIs that were significant across both ROI sizes. In addition to our main experimental ROIs, we also used the left and right primary auditory cortices as control regions (see also Naughtin et al. 2014). As these areas primarily respond to auditory information rather than visual information (e.g., pitch; Hyde et al. 2008), they should not show any evidence of individuation or identification. The auditory cortex control regions were defined in the same way as the other ROIs, with the ROI cube centered on the superior portion of the temporal lobe (Rademacher et al. 2001).

Univariate Analysis

The purpose of the univariate analysis was to identify gross changes in BOLD amplitude that have been hypothesized to reflect the processes of individuation or identification. We extracted time courses for each display condition with percentage signal change at each time point calculated relative 1 volume prior to trial onset. This procedure was performed separately for each ROI and participant. To determine the peak amplitude for encoding, maintenance, and retrieval, we collapsed time courses across all display types, participants, and ROIs and selected the 2 volumes corresponding to the peak at each stage. The time windows for the encoding, maintenance, and retrieval stages of VSTM were 4–8 s, 12–16 s, and 18–22 s after sample display onset, respectively (similar to previous VSTM studies, Todd and Marois 2004; Xu and Chun 2006; Jeong and Xu 2013).

Multivariate Analyses

Multi-voxel pattern analysis was employed to assess for individuation- or identification-related changes in activity that may be present in the ensemble patterns of activity of each ROI (Kamitani and Tong 2005; Haynes and Rees 2006). This type of approach is more sensitive to small, reliable changes in activity that might not be detected in standard univariate analyses. We used custom MATLAB software and a linear support vector machine algorithm (Chang and Lin 2011) for these analyses. We ran separate classification analyses for encoding, maintenance, and retrieval stages for each ROI; data for each voxel were averaged across the respective time windows (4–8, 12–16, and 18–22 s after sample display onset). Prior to MVPA, these data samples were transformed into z-scores and mean-centered to remove any amplitude differences between conditions (Esterman et al. 2009; Tamber-Rosenau et al. 2011). To assess classification performance, we used the leave-one-out cross-validation procedure. On each loop, 1 run was reserved to test the classifier's generalization performance and the remaining 7 runs were used to train the classifier. We averaged classification performance for each ROI across all cross-validation loops and compared this accuracy with chance performance (50%) using a 1-sample t-test (P < 0.05, Bonferroni-corrected for the number of ROIs and number of VSTM stages tested).

Results

Behavioral Performance

Performance on the verbal memory task did not differ between the 3 display types (88.7–90.6% accuracy across all display types; F < 1), which is consistent with the idea that auditory and VSTM systems operate independently of each other (Baddeley 1992; Smith and Jonides 1998). We used Cowan's K formula (Cowan 2001) to estimate participants' VSTM capacity for each display type in the visual memory task. As shown in Figure 2, K estimates significantly differed across display types, F2,36 = 75.21, P < 0.001, ηF2=0.81. Behavioral performance was close to ceiling for 1 object and 4 identical objects, but participants could only hold about 2 objects in memory for the 4-different-object displays.

Figure 2.

Behavioral estimates of VSTM capacity as a function of the 3 display types. These estimates were calculated using Cowan's (2001)K formula. Error bars denote standard error of the mean.

Figure 2.

Behavioral estimates of VSTM capacity as a function of the 3 display types. These estimates were calculated using Cowan's (2001)K formula. Error bars denote standard error of the mean.

Univariate Analyses

We first tested for gross differences in BOLD amplitude associated with individuation and identification processes. For each comparison, we conducted a repeated-measures analysis of variance with factors of display type (1 object, 4 identical objects, and 4 different objects) and stage (encoding, maintenance, and retrieval), separately for each ROI. A significant main effect of display type indicates that a given region is consistently modulated by 1 or both of the processes of interest (i.e., the number of objects—“individuation”; or the number of object features—“identification”) and a significant display type by stage interaction signals that activity associated with the process(es) of interest varies across different VSTM stages. Whenever the analysis for a given region yielded a significant interaction, we ran follow-up t-tests at each VSTM stage to identify the stage(s) at which that region was recruited for the process(es). We present results from the significant regions according to whether they were modulated by the number of objects, the number of object features, or both.

Individuation-Related Activity

Areas associated with object individuation showed either an overall amplitude difference between 1-object versus 4-identical-object displays or an interaction between these 2 display types and stage (see Fig. 3). We found that the left and right iIPS were showed a significant main effect of display type, Fs2,36 > 4.44, Ps < 0.024, ηP2s>0.20, and follow-up t-tests revealed that activity was significantly reduced for displays with 1-object, compared with 4-identical-object displays, ts18 > 2.95, Ps < 0.009. The comparison between 4 identical objects and 4 different objects was not significant, ts18 < 1.08, Ps > 0.294. Consistent with previous work by Xu (2009), we found a profile in the univariate activity that suggests that the iIPS is recruited for object individuation and not object identification. Here, we also show, for the first time, that the involvement of these regions in individuation is consistent across all 3 VSTM stages.

Figure 3.

BOLD time courses during encoding, maintenance, and retrieval stages for regions that were influenced by the number of objects in the display (reflecting object individuation). These areas showed a significant difference between 1-object and 4-identical-object displays at 1 or more VSTM stages. Separate lines denote display types, and the shaded gray areas reflect the time window used for each stage of VSTM. Error bars denote standard error of the mean.

Figure 3.

BOLD time courses during encoding, maintenance, and retrieval stages for regions that were influenced by the number of objects in the display (reflecting object individuation). These areas showed a significant difference between 1-object and 4-identical-object displays at 1 or more VSTM stages. Separate lines denote display types, and the shaded gray areas reflect the time window used for each stage of VSTM. Error bars denote standard error of the mean.

Identification-Related Activity

We identified areas that support object identification as those for which there were significant amplitude differences between 4-identical-object displays and 4-different-object displays, or an interaction between these 2 display types and stage (see Fig. 4). The bilateral anterior cingulate cortex (ACC) and the right dorsolateral prefrontal cortex (DLPFC) showed a significant main effect of display type, Fs2,36 > 3.62, Ps < 0.040, ηP2s>0.17. Follow-up t-tests revealed that this effect was driven by an enhanced response for 4 different objects versus 4 identical objects, ts18 > 2.40, Ps < 0.027.

Figure 4.

BOLD time courses during encoding, maintenance, and retrieval periods for regions that were influenced by the number of object features in the display (reflecting object identification). These regions showed a significant difference between displays containing 4 identical objects and those with 4 different objects at 1 or more VSTM stages. Separate lines denote display types, and the shaded gray areas reflect the time window used for each stage of VSTM. Error bars denote standard error of the mean.

Figure 4.

BOLD time courses during encoding, maintenance, and retrieval periods for regions that were influenced by the number of object features in the display (reflecting object identification). These regions showed a significant difference between displays containing 4 identical objects and those with 4 different objects at 1 or more VSTM stages. Separate lines denote display types, and the shaded gray areas reflect the time window used for each stage of VSTM. Error bars denote standard error of the mean.

Five additional ROIs showed a significant display type × stage interaction, Fs4,72 > 2.87, Ps < 0.047, ηFSs>0.14 (4 of which also showed a significant effect of display type, Fs2,36 > 5.14, Ps < 0.011, ηFSs>0.22): superior medial frontal cortex (SMFC), right premotor cortex (PMC), bilateral insula, and left inferior frontal junction (IFJ). All these regions showed a significant difference in activity between 4 identical and 4 different objects for the encoding and maintenance time windows, ts18 > 2.06, Ps < 0.054, and the right insula also showed a significant difference at retrieval, t18 = 3.17, P = 0.005. Unlike the individuation results, in a task where both identity and location information had to be processed, we failed to observe Xu's (2009) finding that object identification is restricted to the sIPS and LOC. Instead, we found a different set of brain regions was associated with this process, and these regions were active during the encoding and maintenance time windows.

Individuation- and Identification-Related Activity

In addition to those regions that were modulated either by the number of objects (individuation) or object features (identification), we also found some regions that showed evidence of both processes: left PMC, right IFJ, and bilateral sIPS/SPL (see Fig. 5). These 4 regions showed a significant interaction between display type and stage, Fs4,72 > 4.09, Ps < 0.005, ηF2s>0.19. Follow-up t-tests revealed that all regions showed a difference between 1 object and 4 identical objects at encoding, ts18 > 2.01, Ps < 0.059. A significant difference between 1-object and 4-identical-objects displays was also observed during maintenance in the bilateral sIPS/SPL regions, ts18 > 2.15, Ps < 0.045, and in the right hemisphere sIPS/SPL during retrieval, t18 = 2.66, P = 0.016. For the identification comparison, activity was significantly reduced for 4-identical-object displays versus 4-different-object displays for the left PMC, right IFJ, and bilateral sIPS/SPL during encoding and maintenance, ts18 > 2.11, Ps < 0.049. Contrary to what Xu and Chun originally proposed (Xu 2009; Xu and Chun 2009), these findings suggest that individuation and identification cannot be completely dissociated in the brain under conditions where identity and location must be analyzed, as our univariate results revealed a subset of brain areas that are involved in both operations.

Figure 5.

BOLD time courses during encoding, maintenance, and retrieval periods for regions that were influenced by both the number of objects (object individuation) and number of object features (object identification). These regions showed significant differences between both comparisons of interest at 1 or more VSTM stages. Separate lines denote display types, and the shaded gray areas reflect the time window used for each stage of VSTM. Error bars denote standard error of the mean.

Figure 5.

BOLD time courses during encoding, maintenance, and retrieval periods for regions that were influenced by both the number of objects (object individuation) and number of object features (object identification). These regions showed significant differences between both comparisons of interest at 1 or more VSTM stages. Separate lines denote display types, and the shaded gray areas reflect the time window used for each stage of VSTM. Error bars denote standard error of the mean.

Multivariate Analyses

Similar to the univariate analyses, we conducted MVPA on data corresponding to encoding, maintenance, and retrieval stages of VSTM. In separate analyses, we trained a classifier to discriminate between the same 2 key comparisons used in the univariate analyses: 1 object versus 4 identical objects for individuation, and 4 identical objects versus 4 different objects for identification. Classification performance for each ROI was compared with chance (50%) using a one-sample t-test and a significance threshold of P < 0.05, Bonferroni-corrected for the 18 regions tested (i.e., 16 main regions of interest plus 2 control regions) and the 3 VSTM stages (critical P = 0.0009).

Compared with the univariate analyses, the MVPA approach revealed a more extensive neural overlap between individuation and identification processes (see Fig. 6). Six regions displayed evidence of either individuation or identification processes. Specifically, right IFJ, bilateral insula, and left LOC showed significant decoding for the number of objects (ts18 > 4.47, Ps < 0.016 [corrected]), whereas the right DLPFC showed significant decoding for the number of object features (t18 = 4.78, P = 0.008 [corrected]). For 9 of the sixteen regions—the left IFJ, left sIPS/SPL, left LOC, ACC, SMFC, and bilaterally in the PMC and iIPS—classifiers could decode activity between the “number of objects” comparison (1 object vs. 4 identical objects) and the “number of object features” comparison (4 identical vs. 4 different objects) in 1 or more VSTM stages, ts18 > 4.06, Ps < 0.040 (corrected). Of key interest, we found that activity in the bilateral iIPS, left sIPS/SPL, and left LOC could be decoded for the 2 comparisons of interest at single or multiple stages of VSTM, suggesting these regions represent “both” individuation and identification processes, rather than 1 distinct process. In addition, while activity associated with the number of objects or object features could be decoded across all VSTM stages in some regions (e.g., bilateral sIPS/SPL, left PMC, and right LOC), other brain areas were recruited during specific stages (e.g., within the left IFJ, the number of object features presented could be decoded during encoding and maintenance, but the number of objects per se could be decoded at retrieval only).

Figure 6.

Mean classification performance during encoding, maintenance, and retrieval periods across all key regions of interest. Classifiers were trained to discriminate between the 2 key comparisons: 1 object versus 4 identical objects (“number of objects” comparison; object individuation), and 4 identical objects versus 4 different objects (“number of object features” comparison; object identification). Results from each region are displayed on separate plots. Asterisks denote classification performance that is significantly greater than chance across both ROI sizes (Bonferroni-corrected for the number of regions and VSTM stages tested). Error bars reflect standard error of the mean.

Figure 6.

Mean classification performance during encoding, maintenance, and retrieval periods across all key regions of interest. Classifiers were trained to discriminate between the 2 key comparisons: 1 object versus 4 identical objects (“number of objects” comparison; object individuation), and 4 identical objects versus 4 different objects (“number of object features” comparison; object identification). Results from each region are displayed on separate plots. Asterisks denote classification performance that is significantly greater than chance across both ROI sizes (Bonferroni-corrected for the number of regions and VSTM stages tested). Error bars reflect standard error of the mean.

Taken together, these findings suggest an extensive interplay between object individuation and identification in the brain and suggest that signals associated with these processes can be detected across a wide network of regions at each stage of VSTM (see also Naughtin et al. 2014). These findings are inconsistent with the neural object file theory put forward by Xu and Chun (2009), which predicts that individuation and identification are subserved by a limited set of nonoverlapping brain regions. Rather, we show that activity is modulated in a distributed set of common regions when observers are required to identify and individuate a set of items in a display (novel conditions that are not accounted for in the current conceptualization of the neural object file theory).

One caveat to these conclusions is that the differences we observed for individuation and identification could instead reflect some general, unspecified effect of task difficulty. As we elaborate on in the section General Discussion, such task difficulty effects are a potential issue for any study where there are behavioral differences between conditions (including the previous work by Xu and colleagues). It is therefore challenging to operationalize general, unspecified effects of task difficulty as a measure that is independent of the manipulated factors. Nevertheless, as MVPA has the potential to exacerbate such difficulty effects (e.g., see Todd et al. 2013), we addressed this issue by balancing reaction times across our 2 key comparisons of interest. Prior to conducting the MVPAs, we equated reaction times between the “number of objects” and “number of object features” comparisons by removing the slowest trials from the condition with the longest average reaction time and the fastest trials from the condition with the shortest average reaction time, until there was no significant reaction time difference between the conditions, ts14 < 1.56, Ps > 0.141. Data from 4 participants were removed from these analyses, as we could not balance their reaction times between conditions for 1 or both comparisons (leaving the sample at n = 15).

Despite the substantial reduction in power due to fewer trials and subjects included in this analysis, we found 32 of our original 43 significant decoding effects held once reaction time was equated (see Table 1). The number of objects and/or object features could no longer be discriminated in the maintenance or retrieval periods for the left IFJ, right DLPFC, bilateral PMC, left insula, and right sIPS/SPL (Ps > 0.05, corrected, across 1 or both ROI sizes). For the encoding period, the number of objects in the right LOC and the number of object features in the left PMC could not be reliably discriminated (Ps > 0.05, corrected, across 1 or both ROI sizes). It is possible that those effects that were no longer reliable in the reaction time-balanced analysis might have reflected some general effect of task difficulty, rather than the processes of interest. An alternative explanation is that this control analysis simply had less power than the main analysis to detect these effects, due to the reduction in trial numbers and subjects. Although these reaction time-controlled results raise a degree of doubt about the role played by some of our ROIs in individuation and identification, the overall results were largely robust to this balancing procedure. Collectively, our findings demonstrate that individuation and identification processes can be detected in overlapping substrates that are distributed across the brain.

Table 1

Mean classification performance during encoding, maintenance, and retrieval periods for the main decoding analysis (where reaction time [RT] was not equated) and the task difficulty control analysis (where RT was equated)

Region of interest VSTM: encoding
 
VSTM: maintenance
 
VSTM: retrieval
 
Unequal RTs Equal RTs Unequal RTs Equal RTs Unequal RTs Equal RTs 
IFJ (L) 
 1 vs. 4I 57.8 (11) 56.1 (11) 54.9 (10) 56.1 (12) 61.1 (10)* 59.5 (9) 
 4I vs. 4D 64.1 (12)* 68.8 (15)* 61.1 (11)* 62.9 (12) 56.8 (8) 61.5 (12) 
IFJ (R) 
 1 vs. 4I 61.2 (13) 60.3 (15) 59.4 (10) 59.5 (12) 64.0 (8)* 63.2 (9)* 
 4I vs. 4D 61.2 (9) 62.1 (12) 62.8 (14) 63.8 (15) 55.6 (10) 64.7 (15)# 
ACC (Bi) 
 1 vs. 4I 55.7 (8) 55.4 (10) 54.4 (8) 55.2 (10) 72.5 (9)* 73.3 (9)* 
 4I vs. 4D 62.9 (12)* 67.0 (15)* 63.6 (14)* 66.8 (15)* 57.4 (9) 73.1 (11)* 
SMFC (Bi) 
 1 vs. 4I 58.3 (10) 60.0 (12) 49.4 (9) 52.8 (11) 68.6 (14)* 69.2 (12)* 
 4I vs. 4D 62.2 (11)* 67.2 (9)* 62.9 (13)* 66.6 (13)* 54.2 (7) 67.6 (13) 
DLPFC (L) 
 1 vs. 4I 52.9 (11) 54.5 (11) 55.1 (7) 56.9 (11) 58.6 (8) 55.1 (8) 
 4I vs. 4D 60.3 (10) 62.6 (14) 64.8 (15) 64.9 (15) 55.3 (8) 60.3 (12) 
DLPFC (R) 
 1 vs. 4I 55.1 (10) 55.5 (11) 57.3 (10) 58.2 (11) 59.7 (11) 60.1 (12) 
 4I vs. 4D 56.3 (14) 60.3 (14) 62.4 (11)* 65.2 (13) 56.9 (9) 60.7 (10)* 
PMC (L) 
 1 vs. 4I 63.6 (10)* 66.1 (12)* 59.1 (6)* 61.4 (10) 59.7 (10)* 59.7 (11) 
 4I vs. 4D 61.7 (11)* 61.4 (11) 61.3 (12)* 62.5 (12) 53.3 (11) 62.3 (14) 
PMC (R) 
 1 vs. 4I 62.6 (11)* 63.7 (9)* 57.7 (11) 57.9 (13) 61.7 (12)* 58.5 (14) 
 4I vs. 4D 60.9 (11)* 64.8 (13)* 62.2 (13)* 64.0 (14)# 57.9 (9) 62.1 (11) 
Insula (L) 
 1 vs. 4I 51.8 (7) 52.3 (12) 52.4 (8) 54.3 (10) 61.0 (11)* 59.2 (10) 
 4I vs. 4D 58.9 (9) 61.7 (12) 60.4 (9) 63.8 (10) 53.7 (10) 59.5 (14) 
Insula (R) 
 1 vs. 4I 55.0 (8) 54.7 (9) 55.8 (7) 56.9 (10) 65.0 (11)* 63.9 (13)# 
 4I vs. 4D 55.5 (11) 60.3 (16) 57.6 (11) 62.8 (11)* 56.6 (10) 63.0 (12)* 
iIPS (L) 
 1 vs. 4I 69.8 (10)* 66.2 (12)* 68.3 (8)* 67.2 (9)* 57.3 (9) 57.4 (7) 
 4I vs. 4D 57.8 (7)* 67.2 (13)* 61.3 (11)* 65.3 (12)* 51.6 (9) 59.8 (14) 
iIPS (R) 
 1 vs. 4I 66.3 (10)* 65.1 (12)* 66.9 (11)* 69.7 (12)* 61.2 (13) 59.4 (13) 
 4I vs. 4D 56.0 (7) 65.3 (14)* 61.0 (10)* 66.9 (11)* 54.4 (7) 59.0 (12) 
sIPS/SPL (L) 
 1 vs. 4I 67.6 (10)* 67.4 (10)* 64.5 (8)* 63.7 (10)* 63.2 (10)* 61.8 (11)* 
 4I vs. 4D 62.9 (9)* 68.7 (11)* 64.0 (10)* 66.5 (12)* 54.8 (9) 67.0 (10)* 
sIPS/SPL (R) 
 1 vs. 4I 66.0 (13)* 66.7 (13)* 63.4 (9)* 65.1 (10)* 63.6 (11)* 62.7 (12) 
 4I vs. 4D 61.3 (11) 67.0 (14)* 62.3 (10) 66.6 (12)* 56.2 (8) 63.6 (11)* 
LOC (L) 
 1 vs. 4I 60.9 (10) 62.4 (10)* 60.1 (9)* 62.8 (8)* 63.5 (9)* 62.2 (11)* 
 4I vs. 4D 60.0 (12) 64.8 (14) 61.6 (10)* 67.6 (11)* 55.4 (9) 60.6 (13) 
LOC (R) 
 1 vs. 4I 62.7 (11)* 60.1 (12) 66.0 (7)* 66.7 (6)* 67.9 (11)* 67.5 (12)* 
 4I vs. 4D 55.6 (9) 61.2 (12) 57.2 (13) 58.0 (11) 54.2 (4) 68.8 (11)* 
Region of interest VSTM: encoding
 
VSTM: maintenance
 
VSTM: retrieval
 
Unequal RTs Equal RTs Unequal RTs Equal RTs Unequal RTs Equal RTs 
IFJ (L) 
 1 vs. 4I 57.8 (11) 56.1 (11) 54.9 (10) 56.1 (12) 61.1 (10)* 59.5 (9) 
 4I vs. 4D 64.1 (12)* 68.8 (15)* 61.1 (11)* 62.9 (12) 56.8 (8) 61.5 (12) 
IFJ (R) 
 1 vs. 4I 61.2 (13) 60.3 (15) 59.4 (10) 59.5 (12) 64.0 (8)* 63.2 (9)* 
 4I vs. 4D 61.2 (9) 62.1 (12) 62.8 (14) 63.8 (15) 55.6 (10) 64.7 (15)# 
ACC (Bi) 
 1 vs. 4I 55.7 (8) 55.4 (10) 54.4 (8) 55.2 (10) 72.5 (9)* 73.3 (9)* 
 4I vs. 4D 62.9 (12)* 67.0 (15)* 63.6 (14)* 66.8 (15)* 57.4 (9) 73.1 (11)* 
SMFC (Bi) 
 1 vs. 4I 58.3 (10) 60.0 (12) 49.4 (9) 52.8 (11) 68.6 (14)* 69.2 (12)* 
 4I vs. 4D 62.2 (11)* 67.2 (9)* 62.9 (13)* 66.6 (13)* 54.2 (7) 67.6 (13) 
DLPFC (L) 
 1 vs. 4I 52.9 (11) 54.5 (11) 55.1 (7) 56.9 (11) 58.6 (8) 55.1 (8) 
 4I vs. 4D 60.3 (10) 62.6 (14) 64.8 (15) 64.9 (15) 55.3 (8) 60.3 (12) 
DLPFC (R) 
 1 vs. 4I 55.1 (10) 55.5 (11) 57.3 (10) 58.2 (11) 59.7 (11) 60.1 (12) 
 4I vs. 4D 56.3 (14) 60.3 (14) 62.4 (11)* 65.2 (13) 56.9 (9) 60.7 (10)* 
PMC (L) 
 1 vs. 4I 63.6 (10)* 66.1 (12)* 59.1 (6)* 61.4 (10) 59.7 (10)* 59.7 (11) 
 4I vs. 4D 61.7 (11)* 61.4 (11) 61.3 (12)* 62.5 (12) 53.3 (11) 62.3 (14) 
PMC (R) 
 1 vs. 4I 62.6 (11)* 63.7 (9)* 57.7 (11) 57.9 (13) 61.7 (12)* 58.5 (14) 
 4I vs. 4D 60.9 (11)* 64.8 (13)* 62.2 (13)* 64.0 (14)# 57.9 (9) 62.1 (11) 
Insula (L) 
 1 vs. 4I 51.8 (7) 52.3 (12) 52.4 (8) 54.3 (10) 61.0 (11)* 59.2 (10) 
 4I vs. 4D 58.9 (9) 61.7 (12) 60.4 (9) 63.8 (10) 53.7 (10) 59.5 (14) 
Insula (R) 
 1 vs. 4I 55.0 (8) 54.7 (9) 55.8 (7) 56.9 (10) 65.0 (11)* 63.9 (13)# 
 4I vs. 4D 55.5 (11) 60.3 (16) 57.6 (11) 62.8 (11)* 56.6 (10) 63.0 (12)* 
iIPS (L) 
 1 vs. 4I 69.8 (10)* 66.2 (12)* 68.3 (8)* 67.2 (9)* 57.3 (9) 57.4 (7) 
 4I vs. 4D 57.8 (7)* 67.2 (13)* 61.3 (11)* 65.3 (12)* 51.6 (9) 59.8 (14) 
iIPS (R) 
 1 vs. 4I 66.3 (10)* 65.1 (12)* 66.9 (11)* 69.7 (12)* 61.2 (13) 59.4 (13) 
 4I vs. 4D 56.0 (7) 65.3 (14)* 61.0 (10)* 66.9 (11)* 54.4 (7) 59.0 (12) 
sIPS/SPL (L) 
 1 vs. 4I 67.6 (10)* 67.4 (10)* 64.5 (8)* 63.7 (10)* 63.2 (10)* 61.8 (11)* 
 4I vs. 4D 62.9 (9)* 68.7 (11)* 64.0 (10)* 66.5 (12)* 54.8 (9) 67.0 (10)* 
sIPS/SPL (R) 
 1 vs. 4I 66.0 (13)* 66.7 (13)* 63.4 (9)* 65.1 (10)* 63.6 (11)* 62.7 (12) 
 4I vs. 4D 61.3 (11) 67.0 (14)* 62.3 (10) 66.6 (12)* 56.2 (8) 63.6 (11)* 
LOC (L) 
 1 vs. 4I 60.9 (10) 62.4 (10)* 60.1 (9)* 62.8 (8)* 63.5 (9)* 62.2 (11)* 
 4I vs. 4D 60.0 (12) 64.8 (14) 61.6 (10)* 67.6 (11)* 55.4 (9) 60.6 (13) 
LOC (R) 
 1 vs. 4I 62.7 (11)* 60.1 (12) 66.0 (7)* 66.7 (6)* 67.9 (11)* 67.5 (12)* 
 4I vs. 4D 55.6 (9) 61.2 (12) 57.2 (13) 58.0 (11) 54.2 (4) 68.8 (11)* 

Note: Standard deviations are denoted in the parentheses. Decoding accuracies for each region are displayed separately for the “number of objects” comparison (1 object vs. 4 identical objects [1 vs. 4I]) and the “number of object features” comparison (4 identical objects vs. 4 different objects [4I vs. 4D]). All statistical results are Bonferroni-corrected for the number of regions and VSTM stages tested.

*P < 0.050, #P < 0.082 (marginally significant).

Control Analysis

We conducted the same MVPA using data from 2 control regions (left and right auditory cortices) that should not be involved in visual individuation or identification. The purpose of this control analysis was to ensure that the decoding differences described earlier reflect the processes of interest—object individuation and identification—rather than any unanticipated artifact of our data, task design, or analyses. As expected, classifiers could not reliably discriminate between the number of objects in the display (1 object vs. 4 identical objects) and the number of object features (4 identical objects vs. 4 different objects) in the left or right auditory cortices at any VSTM stage, ts18 = 2.64, Ps > 0.855 (corrected). These control results confirm that the observed differences between the display types within the experimental ROIs reflect processes specifically associated with individuating or identifying multiple objects.

General Discussion

The purpose of the current study was twofold: We wanted to test whether object individuation and identification are distinct processes in the brain and whether the brain regions that support these operations vary across different processing stages of VSTM. Using the same logic and a similar experimental approach to that previously employed by Xu (2009), we compared changes in BOLD activity under conditions that differed in either the number of objects (1 object vs. 4 identical objects) or the number of object features (4 identical objects vs. 4 different objects). We used a slow event-related protocol that allowed us to separate activity reflected by encoding, maintenance, and retrieval VSTM stages. Across both univariate analyses and MVPA, we found evidence for both distinct and overlapping neural substrates for individuation and identification. The overlap between these 2 operations was most apparent in the MVPA results, where 9 ROIs showed evidence of individuation and identification at a single or multiple VSTM stages. These results suggest that individuation and identification have distributed and overlapping neural substrates (see also Naughtin et al. 2014), and these operations are not restricted to a few process-specific brain areas, as has previously been suggested (Xu 2009; Xu and Chun 2009).

The involvement of most brain regions in individuation and identification was variable across VSTM stages, particularly at retrieval. The timing of the auditory memory response, however, could account for the absence of individuation- or identification-related activity at the retrieval stage. In other words, activity associated with the auditory memory response could have obscured effects present in the visual memory response. By contrast, the observed modulations associated with differences in number of objects or object features in the display could not simply reflect an artifact of our data, task design, or analysis, as these same comparisons yielded no significant decoding in 2 auditory control regions.

One could argue that data from our main analyses do not rule out the potential confound of general, unspecified task difficulty. In fact, such difficulty effects are also a possible issue for the prior studies by Xu and colleagues, and likely any other study in cognitive neuroscience where performance differs between conditions (e.g., the standard parametric manipulation). Similarly, it is equally problematic to interpret imaging findings when there is no behavioral difference between conditions. Here, we find complementary differences in behavior and the brain. In an attempt to rule out the potential confound of task difficulty, we equated reaction times between our 2 key comparisons, and the results were largely comparable with those from our main decoding analysis. We still found activity in a distributed set of brain regions that was specifically associated with individuation or identification, and 7 of the original 9 regions showed evidence of both processes (ACC, SMFC, right PMC, bilateral iIPS, left sIPS/SPL, and left LOC). Thus, any nonspecific effect of difficulty does not appear to account for the overlap between individuation and identification that we observe.

The large number of common brain regions associated with individuation and identification, either at the same stage or different stages of VSTM, contrasts with theoretical accounts and fMRI results put forward by Xu and colleagues (Xu and Chun 2006, 2007, 2009; Xu 2009; Jeong and Xu 2013). For example, Xu (2009) used univariate analyses and found evidence for individuation tapping only the iIPS and identification only the sIPS and LOC. Within these 3 posterior regions, however, we found the greatest overlap between individuation and identification, particularly during encoding and maintenance stages of VSTM. These discrepant findings resonate with seminal MVPA work by Haxby et al. (2001), who found that decoding techniques, unlike univariate analyses, reveal overlapping stimulus representations in the ventral temporal cortex, even though this cortical area was previously thought to contain subregions that only responded to a single stimulus category (i.e., faces and man-made objects such as houses, scissors, and chairs). Thus, it appears that information pertaining to individuation and identification is reflected by more subtle changes in BOLD activity, which we were able to detect using MVPA.

There are several other reasons why Xu and colleagues might not have observed an overlap between individuation and identification. First, in the present study, participants were required to individuate and identify “each item” within the sample display. By contrast, Xu (2009) had participants judge a test shape based upon its identity only, and not its location. The active involvement of each process in completing the task could have enhanced neural responses associated with individuation and identification. While we cannot say whether the same overlap between object individuation and identification would be observed if we used a pure identification task (e.g., Xu 2009), our findings do suggest that these processes are reflected in common neural substrates in tasks that require participants to individuate and identify all objects in the display.

Second, we used a longer memory retention period. Xu (2009) had participants maintain sample shapes in memory for only 1 s, and thus, activity associated with each VSTM stage was collapsed and finer differences at any given stage could have been temporally smeared. Finally, we analyzed activity in a wider set of ROIs. This broader approach ensured a more thorough exploration of individuation and identification processes in the brain and revealed that signals associated with each of these processes can be detected in a far more distributed neural network.

Results from our recent fMRI study on “temporal” individuation (Naughtin et al. 2014) are consistent with the current findings of involvement of both lower-level perceptual regions and higher-level executive regions in spatial individuation and identification processes. In the previous study, we used a repetition blindness paradigm in which participants had to detect the presence of a scene repetition embedded within a rapid stream of distractors. As participants are typically poorer at individuating an item when it is preceded by one with the same identity (Kanwisher 1987; Park and Kanwisher 1994), we hypothesized that it would be more demanding to successfully individuate 2 temporally distinct scenes when they had the same identity (repeated), relative to different identities (nonrepeated). Using MVPA, we found that activity in the same set of lower- and higher-level regions could discriminate between correctly identified repeated and nonrepeated scenes. Thus, even though our previous investigation (Naughtin et al. 2014) and the current study employed paradigms that differed in both task demands and stimulus properties, findings from both point to a common distributed network for identification and individuation in the spatial and temporal domains.

We are not the first to propose a distributed neural network as the underlying neural basis for perceptual or cognitive abilities. For example, in their MVPA study, Vickery et al. (2011) found that reward processes are reflected across many brain regions, yet a smaller, more focal set of regions emerged in the univariate analysis. In addition, Duncan (2010, 2013) has proposed that a large range of cognitive tasks are underpinned by a distributed set of frontal and parietal regions, which he refers to as the “multiple demands” system. These findings underscore the importance of exploring the role of distributed brain regions in any given perceptual or cognitive process.

Conclusion

The present evidence challenges an earlier view that object individuation and identification are subserved by a relatively small set of distinct brain regions (as suggested by Xu 2009; Xu and Chun 2009). Both univariate analyses and MVPA suggested that individuation and identification are instead reflected across a larger group of brain regions and that these processes have overlapping neural substrates. We propose that individuation and identification might operate within a distributed neural network in which lower- and higher-level regions communicate. Our findings further indicate that earlier work on the neural bases of object individuation and identification may have been restricted by the number of regions tested and the sensitivity of the data analysis techniques employed. At the very least, our data illustrate conditions in which the neural object file theory (Xu and Chun 2009) cannot account for how object individuation and identification are represented in the brain, that is, when the observer must track both the location and identity of objects. Furthermore, we found involvement of these distributed regions varied across different processing stages of VSTM, in a substantial proportion of regions. These results provide new insights into the nature of the neural substrates that give rise to individuation and identification—2 operations that are crucial for a stimulus to reach conscious awareness and be consolidated in VSTM.

Funding

This work was supported by an Australian Research Council (ARC) Discovery grant (DP110102925) and the ARC-SRI Science of Learning Research Centre (SR120300015) awarded to P.E.D. and J.B.M. In addition, P.E.D. was supported by an ARC Future Fellowship (FT120100033) and J.B.M. by an ARC Australian Laureate Fellowship (FL110100103) and the ARC Centre of Excellence for Integrative Brain Function (ARC Centre Grant CE140100007).

Notes

Conflict of Interest: None declared.

References

Baddeley
A
.
1992
.
Working memory
.
Science
 .
255
:
556
559
.
Brainard
DH
.
1997
.
The psychophysics toolbox
.
Spat Vis
 .
10
:
433
436
.
Carp
J
Park
J
Polk
TA
Park
DC
.
2011
.
Age differences in neural distinctiveness revealed by multi-voxel pattern analysis
.
Neuroimage
 .
56
:
736
743
.
Chang
CC
Lin
CJ
.
2011
.
LIBSVM: a library for support vector machines
.
ACM Transact Intell Syst Technol
 .
2
:
1
27
.
Chun
MM
.
1997
.
Types and tokens in visual processing: a double dissociation between the attentional blink and repetition blindness
.
J Exp Psychol Hum Percept Perform
 .
23
:
738
755
.
Cohen
JD
Perlstein
WM
Braver
TS
Nystrom
LE
Noll
DC
Jonides
J
Smith
EE
.
1997
.
Temporal dynamics of brain activation during a working memory task
.
Nature
 .
389
:
604
608
.
Courtney
SM
Ungerleider
LG
Keil
K
Haxby
JV
.
1997
.
Transient and sustained activity in a distributed neural system for human working memory
.
Nature
 .
386
:
608
611
.
Cowan
N
.
2001
.
The magical number 4 in short-term memory: a reconsideration of mental storage capacity
.
Behav Brain Sci
 .
24
:
87
114
.
Duncan
J
.
2010
.
The multiple-demand (MD) system of the primate brain: mental programs for intelligent behaviour
.
Trends Cogn Sci
 .
14
:
172
179
.
Duncan
J
.
2013
.
The structure of cognition: attentional episodes in mind and brain
.
Neuron
 .
80
:
35
50
.
Dux
PE
Ivanoff
J
Asplund
CL
Marois
R
.
2006
.
Isolation of a central bottleneck of information processing with time-resolved fMRI
.
Neuron
 .
52
:
1109
1120
.
Dux
PE
Tombu
MN
Harrison
S
Rogers
BP
Tong
F
Marois
R
.
2009
.
Training improves multitasking performance by increasing the speed of information processing in human prefrontal cortex
.
Neuron
 .
63
:
127
138
.
Emrich
SM
Riggall
A
LaRocque
J
Postle
B
.
2013
.
Distributed patterns of activity in sensory cortex reflect the precision of multiple items maintained in visual short-term memory
.
J Neurosci
 .
33
:
6516
6523
.
Esterman
M
Chiu
Y-C
Tamber-Rosenau
BJ
Yantis
S
.
2009
.
Decoding cognitive control in human parietal cortex
.
Proc Natl Acad Sci USA
 .
106
:
17974
17979
.
Gallivan
JP
McLean
DA
Valyear
KF
Pettypiece
CE
Culham
JC
.
2011
.
Decoding action intentions from preparatory brain activity in human parieto-frontal networks
.
J Neurosci
 .
31
:
9599
9610
.
Harrison
SA
Tong
F
.
2009
.
Decoding reveals the contents of visual working memory in early visual areas
.
Nature
 .
458
:
632
635
.
Haxby
JV
Gobbini
MI
Furey
ML
Ishai
A
Schouten
JL
Pietrini
P
.
2001
.
Distributed and overlapping representations of faces and objects in ventral temporal cortex
.
Science
 .
293
:
2425
.
Haynes
JD
Rees
G
.
2006
.
Decoding mental states from brain activity in humans
.
Nat Rev Neurosci
 .
7
:
523
534
.
Heekeren
HR
Marrett
S
Bandettini
PA
Ungerleider
LG
.
2004
.
A general mechanism for perceptual decision-making in the human brain
.
Nature
 .
431
:
859
862
.
Hyde
KL
Peretz
I
Zatorre
RJ
.
2008
.
Evidence for the role of the right auditory cortex in fine pitch resolution
.
Neuropsychologia
 .
46
:
632
639
.
Jeong
SK
Xu
Y
.
2013
.
Neural representation of targets and distractors during object individuation and identification
.
J Cogn Neurosci
 .
25
:
117
126
.
Kahneman
D
Treisman
A
Gibbs
BJ
.
1992
.
The reviewing of object files: object-specific integration of information
.
Cogn Psychol
 .
24
:
175
219
.
Kamitani
Y
Tong
F
.
2005
.
Decoding the visual and subjective contents of the human brain
.
Nat Neurosci
 .
8
:
679
685
.
Kanwisher
NG
.
1987
.
Repetition blindness: type recognition without token individuation
.
Cognition
 .
27
:
117
143
.
Naughtin
CK
Tamber-Rosenau
BJ
Dux
PE
.
2014
.
The neural basis of temporal individuation and its capacity limits in the human brain
.
J Neurophysiol
 .
111
:
499
512
.
Oosterhof
NN
Tipper
SP
Downing
PE
.
2012
.
Viewpoint (in)dependence of action representations: an MVPA study
.
J Cogn Neurosci
 .
24
:
975
989
.
Park
J
Kanwisher
N
.
1994
.
Determinants of repetition blindness
.
J Exp Psychol Hum Percept Perform
 .
20
:
500
519
.
Pelli
DG
.
1997
.
The videotoolbox software for visual psychophysics: transforming numbers into movies
.
Spat Vis
 .
10
:
437
442
.
Pylyshyn
Z
.
1989
.
The role of location indexes in spatial perception: a sketch of the FINST spatial-index model
.
Cognition
 .
32
:
65
97
.
Rademacher
J
Morosan
P
Schormann
T
Schleicher
A
Werner
C
Freund
HJ
Zilles
K
.
2001
.
Probabilistic mapping and volume measurement of human primary auditory cortex
.
NeuroImage
 .
13
:
669
683
.
Serences
JT
Ester
EF
Vogel
EK
Awh
E
.
2009
.
Stimulus-specific delay activity in human primary visual cortex
.
Psychol Sci
 .
20
:
207
214
.
Smith
EE
Jonides
J
.
1998
.
Neuroimaging analyses of human working memory
.
Proc Natl Acad Sci USA
 .
95
:
12061
12068
.
Spiridon
M
Kanwisher
N
.
2002
.
How distributed is visual category information in human occipito-temporal cortex? An fMRI study
.
Neuron
 .
35
:
1157
1165
.
Szametitat
AJ
Schugbert
T
Muller
K
von Cramon
DY
.
2002
.
Localization of executive functions in dual-task performance with fMRI
.
J Cogn Neurosci
 .
14
:
1184
1199
.
Talairach
G
Tourmoux
P
.
1988
.
Co-planar Stereotaxic Atlas of the Human Brain
 .
New York
:
Thieme
.
Tamber-Rosenau
BJ
Esterman
M
Chiu
Y-C
Yantis
S
.
2011
.
Cortical mechanisms of cognitive control for shifting attention in vision and working memory
.
J Cogn Neurosci
 .
23
:
2905
2919
.
Todd
JJ
Marois
R
.
2004
.
Capacity limit of visual short-term memory in human posterior parietal cortex
.
Nature
 .
428
:
751
754
.
Todd
MT
Nystrom
LE
Cohen
JD
.
2013
.
Confounds in multivariate pattern analysis: theory and rule representation case study
.
NeuroImage
 .
77
:
157
165
.
Tombu
MN
Asplund
CL
Dux
PE
Godwin
D
Martin
JW
Marois
R
.
2011
.
A Unified attentional bottleneck in the human brain
.
Proc Natl Acad Sci USA
 .
108
:
13426
13431
.
Vickery
TJ
Chun
MM
Lee
D
.
2011
.
Ubiquity and specificity of reinforcement signals throughout the human brain
.
Neuron
 .
72
:
166
177
.
Xu
Y
.
2009
.
Distinctive neural mechanisms supporting visual object individuation and identification
.
J Cogn Neurosci
 .
21
:
511
518
.
Xu
Y
.
2010
.
Neural representation of targets and distractors during object individuation and identification
.
J Neurosci
 .
30
:
14020
14028
.
Xu
Y
.
2008
.
Representing connected and disconnected shapes in human inferior intraparietal sulcus
.
NeuroImage
 .
40
:
1849
1856
.
Xu
Y
.
2007
.
The role of the superior intraparietal sulcus in supporting visual short-term memory for multifeature objects
.
J Cogn Neurosci
 .
27
:
11676
11686
.
Xu
Y
Chun
MM
.
2006
.
Dissociable neural mechanisms supporting visual short-term memory for objects
.
Nature
 .
440
:
91
95
.
Xu
Y
Chun
MM
.
2009
.
Selecting and perceiving multiple visual objects
.
Trends Cogn Sci
 .
13
:
167
174
.
Xu
Y
Chun
MM
.
2007
.
Visual grouping in human parietal cortex
.
Proc Natl Acad Sci USA
 .
104
:
18766
18771
.