Dissociating the Neural Mechanisms of Visual Attention in Change Detection Using Functional MRI

Scott A. Huettel, Gu¨ven Gu¨zeldere, and Gregory McCarthy Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/13/7/1006/1759185/089892901753165908.pdf by guest on 18 May 2021

Abstract & We investigated using functional magnetic resonance ventral visual areas was temporally associated with the imaging (fMRI) the neural processes associated with perform- duration of visual search. As such, our results support a ance of a change-detection task. In this task, two versions of distinction between brain regions whose activation is modu- the same picture are presented in alternation, separated by a lated by attentional demands of the visual task (extrastriate brief mask interval. Even when the two pictures greatly differ cortex) and those that are not affected by it (primary visual (e.g., as when a building is in different locations), subjects cortex). A second network of areas including central sulcus, report that identification of the change is difficult and often insular, and inferior frontal cortical areas, along with the take 30 or more seconds to identify the change. This thalamus and basal ganglia, showed phasic activation tied to phenomenon of ‘‘change blindness’’ provides a powerful and the execution of responses. Finally, parietal and frontal regions novel paradigm for segregating components of visual attention showed systematic deactivations during task performance, using fMRI that can otherwise be confounded in short-duration consistent with previous reports that these regions may be tasks. By using a response-contingent event-related analysis associated with nontask semantic processing. We conclude technique, we successfully dissociated brain regions associated that detection of change, when transient visual cues are not with different processing components of a visual change- present, requires activation of extrastriate visual regions and detection task. Activation in the calcarine cortex was associated frontal regions responsible for eye movements. These results with task onset, but did not vary with the duration of visual suggest that studies of change blindness can inform under- search. In contrast, the pattern of activation in dorsal and standing of more general attentional processing. &

INTRODUCTION One challenge considered by recent functional neuro- Visual attention allows organisms to allocate processing imaging studies is the dissociation of different compo- resources to selected locations or objects in the visual nents of visual attention, such as cue-directed attention field. Lesion studies in humans and nonhuman primates versus target responses (Hopfinger, Buonocore, & Man- have suggested that visual attention depends upon a gun, 2000) or orienting to locations versus detection of distributed network of brain regions, including the stimuli in unattended locations (Corbetta, Kincade, Ol- posterior parietal cortex (Posner, Walker, Friedrich, & linger, McAvoy, & Shulman, 2000). In the present study, Rafal, 1984; Mesulam, 1981), frontal cortex (Paus, 1996), our goal was to dissociate, in a single experimental task, and cingulate cortex, as well as the superior colliculus brain regions associated with attentional processing and thalamus (Posner & Petersen, 1990). In conjunction, from those associated with other components of the functional neuroimaging studies have demonstrated the task, such as nonattentive perceptual processing or participation of the intraparietal sulcus (IPS), posterior response execution. We employed a change detection superior frontal gyrus (SFG), and precentral gyrus in paradigm known as a ‘‘flicker’’ task (Rensink, O’Regan, visual attention tasks (Courtney, Petit, Maisog, Unger- & Clark, 1997; Rensink, 2000b) while testing subjects leider, & Haxby, 1998; Nobre et al., 1997; Corbetta, using functional magnetic resonance imaging (fMRI). On Miezen, Shulman, & Petersen, 1993; Corbetta, Shulman, each trial, two photographs were presented in alterna- Miezin, & Petersen, 1995). This network may also in- tion, separated by a short-duration mask. The images clude visual cortical regions, depending upon the stimuli differed in one aspect, such as the presence/absence, and task (e.g., Corbetta, Miezen, Dobmeyer, Shulman, & color, or position of a single object. The subject’s task Petersen, 1990), in line with the distinction between was to identify the change. Figure 1 provides an example ventral and dorsal visual pathways (Ungerleider & Mis- of an image pair used in the current experiment. The hkin, 1982; Ungerleider & Haxby, 1994). two photographs in the pair differ in that a sign in the upper right above the building is present in the left image but absent in the right image. The short-duration Duke University, Durham, NC mask prevented automatic detection of change that

D 2001 Massachusetts Institute of Technology Journal of Cognitive Neuroscience 13:7, pp. 1006–1018

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/089892901753165908 by guest on 24 September 2021 Figure 1. The flicker task used in the present experiment. Shown in A is an example stimulus pair from the current experiments. These two scenes differ in one aspect, the presence or absence of a sign at upper right. Typical changes across scenes were the presence/absence of an object, the location of an object, or the color of an object. The order of Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/13/7/1006/1759185/089892901753165908.pdf by guest on 18 May 2021 events on any one trial is shown in B. Between each pair of trials was a 2-sec fixation cross on a black screen. For the first 30 sec of each trial, the pictures were presented in alternation, each for 300 msec with a 100-msec grayscale mask between them. During the final 10 sec of each trial, the mask was removed and the pictures were presented for 400 msec.

occurs when low-level motion transients are available, sees the same images and mask flickering at a constant such as when the interval between stimuli is less than rate, with no interruptions from cuing, extended inter- about 80 msec (Pashler, 1988; Phillips, 1974). Behavioral stimulus interval, or feedback. Differences in the pattern evidence suggests that, when this automatic detection of activation over time may therefore be attributed to process is not available, subjects engage in a controlled cognitive processing and not to stimulus presentation. serial search among display elements (Rensink et al., Despite these characteristics and the significant recent 1997). The search process is guided by the semantic interest in behavioral studies of change detection (e.g., content of the image, such that changes are more Rensink, 2000a; Simons & Levin, 1997), functional neu- quickly detected on objects that are named in verbal roimaging studies of change detection in a flicker task descriptions (Rensink et al., 1997). As such, cuing has have not been previously conducted. relatively little effect on search, with facilitation only The subject’s behavioral response in a change-detec- reported for color changes (Aginsky & Tarr, 2000). tion task provides an objective marker for the subjective Finally, detection of change is associated with the locus process of visual search. Taking advantage of this, we of attention rather than of eye position, although the developed an analysis technique that uses response- two move similarly: Even when looking right at the contingent event-related fMRI. Because of the extended change location, subjects fail to detect 40% of changes duration of our experimental task, we can use informa- across blinks (O’Regan, Deubel, Clark, & Rensink, tion about the duration and timing of responses to guide 2000). analyses. Typical short-duration visual search tasks do Change-detection tasks of this kind have two charac- not temporally separate search and response processes teristics that map well onto fMRI analysis techniques. by durations greater than the temporal resolution of the First, change detection is a slow process in comparison fMRI hemodynamic response. Furthermore, there is to other visual searches (hence the name, ‘‘change little variability in the duration of the search process blindness’’). Response times in masked change-detec- (e.g., from 1 to 2 sec). Our analysis uses the response- tion tasks typically range from 5 to 40 sec or more, time variability over trials to identify voxels whose depending on the scene characteristics. In the present activity is associated with visual search. Simply put, the study, mean response time was approximately 23 sec. In duration of voxel activation, if that voxel is associated contrast, most visual search tasks used in cognitive with search, should show sustained activation through- experiments have response times that are roughly one out search, with short-duration activation when the order of magnitude faster. Thus, the extended durations target is found quickly and long-duration activation of change-detection trials match well the temporal when the target is found slowly. In contrast, voxels properties of the fMRI hemodynamic response, which associated with low-level visual processing should only rises and falls over a minimum of 10–15 sec. A second show phasic activation at trial onset, due to the stimulus characteristic is display homogeneity, in that the raw appearance. Voxels associated with response processing visual stimulus has a constant pattern over an extended should show a hemodynamic response time-locked to interval. Throughout each trial of the task, the subject the behavioral response.

Huettel, Gu¨zeldere, and McCarthy 1007

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/089892901753165908 by guest on 24 September 2021 We had two goals for the current study. First, we evaluated whether the extended-duration change-detection task, in conjunction with response-contingent analyses, can be used to dissociate brain regions responsible for different cognitive processes within a complex task. We then investigated how patterns of activation associated with different components vary with processing demands of the task. For example, if

activation in primary visual areas is associated with the Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/13/7/1006/1759185/089892901753165908.pdf by guest on 18 May 2021 duration of visual search, that result would suggest that those areas are susceptible to attentional influences. No duration-related changes in activity would suggest Figure 3. Trial-onset-related activation in calcarine cortex. Shown at that primary visual regions are not sensitive to atten- right are the time courses of activation, in percent signal change over a tional demands, at least from search tasks similar to prestimulus baseline, across the four response-time categories for the that conducted. region of active voxels in calcarine cortex (shown in the slice at left). In this and all subsequent figures, the colormap displaying fMRI activation indicates voxels whose time courses significantly correlated with RESULTS experimental hypotheses, with red indicating a mean correlation across conditions of .63 (p < .001) and yellow indicating a mean correlation Behavioral Testing across conditions of .85 or greater. Activation in this region was For the behavioral analyses reported here, all trials were modulated by trial onset, but did not show a temporal modulation associated with duration of visual search. categorized into one of four response bins according to the subject’s response time on that trial: 0–10, 10–20, 20–30, or 30–40 sec. These response categories were ing from that bin than in any other. Intersubject differ- used in the fMRI analyses reported in the next sections. ences in response time were small, in comparison. The For our experimental task and analysis techniques to be range of mean response times across subjects was 16–22 appropriate for fMRI, response times should be evenly sec (mean: 19 sec, standard deviation: 2 sec). Further- distributed across the trial interval, to ensure maximum more, responses to each stimulus were significantly variability, but should be similar across subjects and correlated across subjects, as revealed by Monte Carlo stimuli. simulation of the subject response data ( p < .00001). The distribution of response-time categories for the So, the flicker task used provides the advantages of high individual stimulus trials is shown in Figure 2. As is response-time variability combined with low intersubject evident from the figure, response time was relatively and low interstimulus variability. uniformly distributed across stimulus trials, with slightly fewer trials in the 20- to 30-sec bin than in the others, so Brain Regions Associated with Trial Onset greater variability was expected in the fMRI data result- Figure 3 presents the activation in the calcarine cortex (see Table 1 for voxel locations and statistical information). For every response bin, there was a transient peak response, followed by a return to a lower level. A two- factor analysis of variance investigated whether there was a difference in calcarine activation as a function of TR (À8 to 38 sec) and response-time category (0–10, 10– 20, 20–30, or 30–40 sec). Although the form of the response was visibly similar across all four response-time categories, the ANOVA revealed significant differences between the four response categories, such that more activation was found for the longer response categories, F(3,69) = 12.4, p < .0001. This difference was observed throughout the trial interval, possibly representing differences in stimulus properties (e.g., number of objects) Figure 2. The histogram of mean response time to stimulus pairs. that contribute both to task difficulty and to greater Shown are the mean response categories for all stimulus pairs used in visual input. As evident from the task design shown at the fMRI testing. A response category rating of 1.0 indicates that every the bottom of Figure 3, the transient response at task subject responded within 0–10 sec for that stimulus pair. Conversely, a onset resulted from the reduced visual input present in response category rating of 4.0 indicates that every subject responded within 30–40 sec for that stimulus pair. As is evident from the figure, the no-mask (i.e., not flickering) and fixation cross the response times across stimuli were roughly uniformly distributed (black background) periods. The only other area active across the range of possible values. to trial onset was the medial frontal gyrus, which was

1008 Journal of Cognitive Neuroscience Volume 13, Number 7

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/089892901753165908 by guest on 24 September 2021 Table 1. Brain Regions Significantly Active Across Subjects, Organized by Analysis Condition and Lobe within the Brain

No. of Subjects MNI Coordinates Mean Size Region L/R x y z L/R Notes

Trial onset Occipito-parietal Lingual gyrus/cuneus 9 14 À68 À8 2734/3244 Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/13/7/1006/1759185/089892901753165908.pdf by guest on 18 May 2021 Frontal Precentral gyrus (L) 8 À44 À83455 Medial frontal gyrus 9 À6 8 52 84/42 A

Visual search Occipito-parietal Fusiform/lingual gyri 10 28 À66 À16 * B Intraparietal sulcus 10 22 À72 34 * B Parahippocampal gyrus 6/7 À22 À34 À8 70/85 Insular Anterior insula 7/8 32 32 À4 137/257 Frontal Precentral gyrus 9/10 À28 À8 54 917/659 C Medial frontal gyrus 9/10 8 6 54 90/215 A Superior frontal gyrus (R) 7 16 66 À10 141 D Precentral sulcus 9 48 2 30 420/458 Inferior frontal gyrus (L) 7 À42 16 26 405

Response Frontal Central sulcus (L) 7 À44 À36 54 658 E Inferior frontal gyrus (R) 4 48 14 26 94 Anterior cingulate gyrus (R) 4 8 10 42 39 Medial frontal gyrus 7 2 10 54 363 F Insular Insula 6/7 44 20 À14 272/506 Temporal Hippocampus 2/5 14 0 À20 65/75 Hippocampus (R) 2 24 À26 À14 53 Subcallosal gyrus (L) 4 À22 8 À14 70 Occipito-parietal Parahippocampal gyrus (L) 2 À18 À38 À14 39 Inferior parietal/cuneus (R) 2 40 À60 42 67

(continued on next page)

Huettel, Gu¨zeldere, and McCarthy 1009

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/089892901753165908 by guest on 24 September 2021 Table 1. (continued)

No. of Subjects MNI Coordinates Mean Size Region L/R x y z L/R Notes Subcortical Subthalamic nucleus 5/4 4 À20 À6 66/170 Caudate 2/4 8 6 0 39/89 Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/13/7/1006/1759185/089892901753165908.pdf by guest on 18 May 2021 Putamen (L) 4 À22 6 0 99 Thalamus 4/6 8 À20 2 39/116

Deactivation Occipito-parietal Angular gyrus 9 À52 À64 16 4654/3383 Precuneus 9 0 À62 16 3043 Inferior parietal lobule 6 32 À40 68 285/1431 Temporal Middle/superior temporal gyrus 9/8 À58 À16 À32 1223/484 Superior temporal gyrus 8 60 À6 4 476/918 G Lateral sulcus (R) 8 60 À32 18 707 G Frontal Frontal pole 4/5 À34 60 À18 155/143 Anterior cingulate 8 À2 54 2 1172 Middle frontal gyrus (L) 8 À22 54 28 369 Central sulcus 6/7 42 À20 44 153/448 Cingulate gyrus 6 2 À22 44 982 Superior frontal gyrus 7/9 À24 28 50 995/850

Notes: A. This region shows significant positive correlation to both trial onset and experimental task (only such overlapping activation); area corresponds to supplementary eye fields. B. Activation region is continuous from the fusiform gyrus in the temporal lobe through the IPS in the parietal lobe, corresponding to extrastriate visual cortical areas. This activation is sufficiently large to make voxel counts not relevant. C. Activation corresponds to frontal eye fields. D. Activation not present in random-effects analysis. E. Includes one subject whose activation was right lateralized (see text). F. Activation is contiguous with SEF (note A), but does not overlap them. G. Contiguous activations in right hemisphere. Indicated are all foci consisting of 35 or more contiguous suprathreshold voxels (t > 3.5; see text for analysis details). This cluster size threshold was chosen to reflect a volume equivalent to four or more uninterpolated voxels (acquisition voxel volume: 3.75 Â 3.75 Â 5 mm; normalized voxel volume: 2 Â 2 Â 2 mm). Shown for each focal activation are the numbers of subjects with significant activation in that region, the coordinates in MNI space, and the size of the activation averaged across subjects. All activations were bilateral, unless explicitly indicated otherwise. No voxel counts are provided for the activations in fusiform/lingual gyri and along the IPS, as they include the entire anatomical regions identified.

also associated with visual search and will be described coordinates for these activations are presented in Table in the next section. 1. Activation in the ventral extrastriate cortex is shown in Figure 4A. Bilateral activation along the posterior extent of the fusiform gyrus (FFG) was found in all subjects; Brain Regions Associated With Visual Search this activation abutted the trial-onset activation found in A network of brain regions, including the extrastriate the calcarine cortex in superior slices. Additional foci of visual cortex, frontal and supplementary eye fields, and activation were found in the bilateral parahippocampal insular/inferior frontal cortex, was systematically acti- gyrus. The parahippocampal activation was not contig- vated for the duration of visual search. Stereotaxic uous with the FFG activation.

1010 Journal of Cognitive Neuroscience Volume 13, Number 7

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/089892901753165908 by guest on 24 September 2021 ever, the subjects’ response times were roughly uniformly distributed within the intervals. This resulted in relatively less temporal smoothing in the 30- to 40-sec bin than in the other bins, allowing for a sharper hemodynamic time course. Similar activation time courses were found for the anterior cingulate (activation shown in Figure 5A; time courses not plotted). Significant activation was also found in the thalamus

and basal ganglia. Figure 5B indicates the loci and time Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/13/7/1006/1759185/089892901753165908.pdf by guest on 18 May 2021 courses of thalamic activation, which was observed bilaterally. As seen in the central sulcus activation, the different response times evoked transient hemodynamic responses with similar latencies. Bilateral caudate activation was also observed (Figure 5C), with response time courses nearly identical to those from the thalami. Visible in Figure 5C are the observed activations in the putamen (larger spatial extent in left hemisphere) and insular cortex (bilateral), whose time courses are not plotted. Cerebellar activation was bilateral, but of larger Figure 4. Visual-search-related activation in extrastriate visual regions. spatial extent in the hemisphere ipsilateral (right) to the At right are the time courses of activation, in percent signal change responding hand. over a prestimulus baseline, for voxels in the FFG (A) and the IPS (B) foci. For both regions, activation rose at trial onset and fell following One subject exhibited the opposite pattern of lateral- the cessation of visual search. ization from the other subjects. In this subject, the cerebellar activation was in the left hemisphere, the Dorsal extrastriate activation is shown in Figure 4B. anterior insula activation was in the left hemisphere Visible at the top of the figure (parietal lobe) is bilateral only, the central sulcus activation was in the right hemi- activation in the IPS. These activation foci extended from the superior parietal to occipital cortex, but did not extend into the calcarine cortex. The time course of activation was similar to that found in the FFG, save that the activation reached a peak slightly later in the IPS (12–14 sec) than in the FFG (8–10 sec). Significant, well-defined activation was found in the precentral gyrus and medial frontal gyrus, both bilaterally (Figure 4B). These locations are consistent with previous stereotaxic reports of the frontal eye fields and supplementary eye fields, respectively (Paus, 1996). Additional foci of activation in the frontal lobe were found across subjects in the inferior frontal gyrus (IFG) and insular cortex (not shown, see Table 1 for locations).

Brain Regions Associated With Response Execution Response execution was associated with activation around the central sulcus (contralateral hemisphere), anterior cingulate, insular cortex, basal ganglia, and cerebellum. Figure 5A indicates the activation around the left central sulcus. Evident from the time course plots is the systematic relation between time of behavioral response and onset of the hemodynamic response. The increased hemodynamic amplitude of the 30- to 40- sec bin is due to the lack of its within-category variability Figure 5. Response-related activation. At right are the time courses of activation, in percent signal change over a prestimulus baseline, for in subjects’ response times; that is, when the mask was voxels in the central sulcus (A), the thalamus (B), and the caudate removed, all subjects reported seeing the changes with- nucleus (C). All regions showed transient hemodynamic responses in the first few seconds. In the other categories, how- temporally consistent with response execution.

Huettel, Gu¨zeldere, and McCarthy 1011

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/089892901753165908 by guest on 24 September 2021 sphere, and the putamen activation was in the right hemisphere. As this subject was left-handed, we assume that the subject turned the response box in the scanner so that he could use his dominant hand to press the response button.

Brain Regions Showing Deactivations During Task Performance Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/13/7/1006/1759185/089892901753165908.pdf by guest on 18 May 2021 One limitation of any long-duration experimental task is that subjects’ cognitive processing is heterogeneous, in that different processes are likely to be invoked during different aspects of the task. In our experimental task, the stimuli remained on screen following the behavioral response, which indicated the conclusion of visual search. Therefore, we considered the possibility that subjects would engage in task-unrelated cognitive processing following the detection of the stimuli. As will be seen below, we identified a network of brain regions that showed systematic deactivations, relative to baseline, during the experimental task, including the angular gyrus, precuneus, middle frontal gyrus (MFG), and SFG. Figure 6A shows regions evidencing deactivations during task performance. Plotted at right is the set of time courses found for the angular gyrus, which showed a bilateral deactivation pattern. Regardless of eventual response time, activation in this region decreased over the first 8 sec of the trial. Then, for trials where the Figure 7. The overall patterns of activation in the current study. subject responded within the first 10 sec, the response Shown are the voxels whose activation time courses were associated began returning to baseline. The trough for the other with task onset, visual search, response execution, or deactivation during task performance. The overlays indicate areas active at a significance threshold of p < .001.

three response categories occurred about 10 sec following trial onset, after which time the hemodynamic response returned to baseline. Interestingly, although the three latter response categories began their return to baseline at the same time point, the rate of return was most rapid for the earlier behavioral responses. Because this hemodynamic response antedated the subjects’ button presses, the differential returns to baseline suggest that stimulus effects, such as complexity or number of objects, might mediate both response time and hemodynamic properties in these regions. Similar time course patterns, with return to baseline faster for earlier behavioral responses, were found for all brain regions discussed in this section. Also visible in Figure 6A are the regions of activation in the precuneus and MFG (left only), whose time courses are not plotted. Bilateral deactivation patterns were found in the SFG, as shown in Figure 6B. Figure 6. Deactivations during task performance. At right are the time courses of activation, in percent signal change over a prestimulus baseline, in the angular gyrus (A) and in the superior frontal gyrus (B). Summary of Activations Observed These regions showed deactivations during the performance of the visual search task, although the time course of the rise to baseline did Figure 7 displays the patterns of activations found for not match the time course of visual search (see text for details). each of our experimental hypotheses. Readily apparent

1012 Journal of Cognitive Neuroscience Volume 13, Number 7

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/089892901753165908 by guest on 24 September 2021 is the complementary nature of the processing compo- 1996). Our results supported the functional association nents: The areas associated with trial onset were adja- between these regions, as their activation patterns cent to those associated with task performance, while across response bins were similar. Gitelman et al. neighboring parietal regions showed task-related deac- (1999) found a similar activation network in an fMRI tivations. study of spatial attention using a variant of the Posner cuing task (Posner, 1980). However, their network included activation in basal ganglia and thalamus, which DISCUSSION were restricted to response execution in the present

We were able to dissociate and identify different brain study. Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/13/7/1006/1759185/089892901753165908.pdf by guest on 18 May 2021 systems associated with attentionally guided search, While our results were consistent with studies inves- response execution, passive perceptual viewing, and tigating neural substrates of spatial attention, they were non–task-related cognitive processing using fMRI in a only partially consistent with research addressing the complex visual search task. Dissociation was accom- related construct of spatial working memory. Our con- plished through a response-contingent analysis techni- cept of working memory as advanced in the cognitive que that utilized the variability in subject response time. neuroscience literature (e.g., Cohen et al., 1997) has two Our technique readily segregated key components that primary components: short-term storage and informa- can be confounded in short-duration visual attention tion manipulation/comparison. For subjects to perform tasks. Here, we evaluate the results of this segregation the change-detection task used in the present experi- for understanding of attentional processing, while con- ment, they must attend to a spatial location, remember sidering extensions of this experimental design to other objects over a short interval (500 msec), evaluate topics. whether something changed, and select a new location based on their memory of what has been previously searched. Thus, the change-detection task required both Attention and Visual Search short-term storage and comparison of presented and The primary regions found to be associated with visual remembered displays. It differed from many working search were dorsal and ventral extrastriate visual cortical memory tasks in the absence of information rehearsal, areas, along with the frontal and supplementary eye because the continuous flicker cycle eliminated the need fields. In contrast, calcarine cortical areas were associ- to hold information in memory for extended durations. ated with task onset, but not visual search. Regions Our posterior regions of activation were similar to associated with response execution included central those implicated in spatial working memory tasks sulcus, insular, and inferior frontal cortical areas, thala- (Belger et al., 1998; Smith & Jonides, 1997; Ungerleider, mus, and basal ganglia. 1995), in that significant parietal cortex activation to A central question in the visual attention literature is visual search was largely restricted to the regions sur- the identification of brain regions whose activation rounding the IPS. Given that the flicker task shares appears to be modulated by visuospatial attention. Our spatial comparison processes with many spatial working results suggest a distinction between medial calcarine memory tasks, our results were consistent with the idea cortex, which likely includes primary visual areas, and that parietal cortex activation in spatial working memory the surrounding visual cortex, such that the former is tasks may be associated with the process of attending to not affected by attentional demands of the task while spatial locations and evaluating their contents. activation in the latter varies with task demands. This Spatial working memory studies have consistently distinction is consistent with most earlier work, which shown dorsolateral prefrontal cortex (dlPFC) activation has suggested that extrastriate but not striate visual in the anatomical region of the MFG (D’Esposito et al., cortex is modulated by attention (Heinze et al., 1994; 1995; Cohen et al., 1994; McCarthy et al., 1994). In the Moran & Desimone, 1985). However, a number of present study, our primary activation foci to visual recent studies have demonstrated attentional effects in search in prefrontal cortex were not in the MFG, but primary visual cortex (Gandhi, Heeger, & Boynton, 1999; instead in the IFG near the MFG border. One possible Somers, Dale, Seiffert, & Tootell, 1999), perhaps due to reason for this difference may reside in the diversity of reentrant feedback from extrastriate areas (Martı´nez cognitive processes associated with working memory; in et al., 1999). More established attentional effects exist in different experiments, ‘‘working memory’’ labels con- parietal and frontal regions, which have been associated trolled selection (as in the present study), maintenance with spatial shifts of attention and attentionally guided of information over time, or response selection. Previous eye movements. Eye movements and spatial attention studies that have found dlPFC activation typically require shifts have been shown to elicit similar activation pat- extended memory maintenance (e.g., delayed match to terns when independently performed (Corbetta et al., sample or n-back tasks; Braver et al., 1997; Cohen et al., 1998). In both cases, parietal cortex activation is tightly 1997; Courtney, Ungerleider, Keil, & Haxby, 1997; Hax- coupled with activity in the precentral cortex, identified by, Ungerleider, Horwitz, Rapoport, & Grady, 1995) or in the frontal eye fields (Courtney et al., 1998; Paus, inhibition of a prepotent response (e.g., oddball tasks;

Huettel, Gu¨zeldere, and McCarthy 1013

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/089892901753165908 by guest on 24 September 2021 McCarthy, Luby, Gore, & Goldman-Rakic, 1996, 1997). In adjacent gyri. Because the regions of deactivation were contrast, tasks involving shifting of attention rapidly consistent across both PET and fMRI and are present across locations have not reliably elicited dlPFC activa- whether or not the PET data are globally normalized, tion (Nobre et al., 1997; Corbetta et al., 1995; Corbetta Votaw and colleagues concluded that deactivations do et al., 1993). Furthermore, response execution, in not result from data analysis artifacts. Instead, they may itself, did not activate dlPFC, as seen in Table 1. Given reflect areas inhibited during performance of the naming these previous results in conjunction with our find- task or preferentially activated by the size-judgment task. ings, the role of dlPFC in spatial working memory The regions identified in the present study as less

tasks may be limited to maintenance processing or to active during visual search may underlie baseline seman- Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/13/7/1006/1759185/089892901753165908.pdf by guest on 18 May 2021 response selection, rather than to spatial search. tic processing that is interrupted when the subject begins the perceptual task. Deactivation of a similar neural system has been found during visual processing Deactivations during Task Performance compared to passive visual stimulation (Shulman et al., An interesting and unexpected finding was the presence 1997). Importantly, in the present study, the return to of large, well-defined deactivations during task perform- signal baseline does not occur following task comple- ance. These were largely restricted to the angular gyrus tion, but 8–12 sec after task onset. We suggest that this and posterior cingulate in the posterior cortex (both latency delay resulted from the initial organization of bilateral), and to the left MFG and bilateral SFG in the scene information during the first few seconds of the frontal cortex. These deactivations were not temporally task, at which time the subjects set up a general layout of associated with subject response time, in that the trough the image and a search strategy. This delay has not been latency was similar for all response categories. However, previously reported in blocked designs (e.g., Binder the return to baseline was slower for the longer re- et al., 1999), which confound time of task completion sponse categories. with the onset of a rest period. Our results thus argue Deactivation patterns similar to that found in the for multiple dissociable neural systems, as proposed by present study have recently been reported by two Binder et al. (1999), including a perceptual processing groups using very different tasks. Binder et al. (1999) system active throughout task performance and a con- compared tone detection, semantic comparison, and ceptual/reflective processing system active following ini- phonetic comparison tasks with a no-stimulus rest peri- tial task processing. od in a blocked design. In the comparison rest–tones (e.g., areas more active during the rest period), they Change Detection and Change Blindness found activation in the angular gyrus, posterior cingulate, and dorsal prefrontal cortex, primarily in the left The experimental task we used takes advantage of the hemisphere. The stereotaxic coordinates of the activa- phenomenon of change blindness to investigate visual tion center of mass in these areas reported by Binder attention. Although change blindness provides a fertile et al. fall within the activations observed in the research paradigm for investigation of the neural sub- present study, although our activations were generally strates of perceptual and cognitive processing, other bilateral. Furthermore, they reported deactivation in important topics can be addressed using the experimen- the parahippocampal gyrus, which exhibited task- tal techniques of the present study. related activation in our experiment. With that excep- The central phenomenal aspect of change-detection tion, our event-related design identified a similar set of tasks is the shift in the display from a hidden change brain regions as their earlier blocked design work. before detection to a conspicuous change afterward. Binder et al. reported that similar regions are active in During the postexperiment debriefing, our subjects a semantic–tone comparison, concluding that these reported that the change ‘‘pops out at them’’ following regions underlie conceptual, not perceptual function- its detection, so that ‘‘it is visible no matter where you ing, due to their relative isolation from primary sensory look.’’ The comparison between effortful search during cortices (Felleman & Van Essen, 1991; Mesulam, 1985). predetection and effortless noticing during postdetec- Votaw et al. (1999) described a similar pattern of tion is striking and impressive, as one of the hallmarks of deactivations in a comparison between a confrontational change blindness is that even very large changes (e.g., naming task, which requires subjects to verbally identify the reflection of a large building in water) can take tens the name of a picture, and a control size-judgment task. of seconds for detection. Detection is phenomenally In both PET and fMRI modalities, they demonstrated similar to unmasked flicker; in both cases, the change that the naming task, relative to the size-judgment task, becomes readily apparent without controlled attention showed a decrease in activation in inferior parietal areas to its location. Although an understanding of the neural (nearly identical to our region labeled as ‘‘angular processing underlying this phenomenal shift would be gyrus’’), the precuneus, and MFG and SFG (left lateral- of considerable interest, our results suggested a poten- ized, as in the present study). Increases in activation tial confound to be avoided: the presence of task- were found in ventral visual areas, notably the FFG and unrelated conceptual processing. As seen in Figure 7,

1014 Journal of Cognitive Neuroscience Volume 13, Number 7

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/089892901753165908 by guest on 24 September 2021 multiple brain regions showed increased activation fol- 40 sec of alternating presentations of the two images in lowing change detection, although this activation can be the pair (see Figure 1). For the first 30 sec of the trial, attributed to conceptual processing rather than to phe- the two images were presented for 300 msec, separated nomenal changes in the display. Careful attention to by the 100-msec mask. The mask was removed during control conditions will be necessary to deconfound the last 10 sec of the trial, during which time the stimuli conceptual activation from perceptual changes. alternated every 400 msec. This allowed all subjects to A second aspect of change blindness that lends itself find every stimulus change, even if previously unde- to functional neuroimaging investigation is the phenom- tected. Each run lasted 7 min 12 sec.

enon of ‘‘sensing’’ of change (Fernandez-Duque & The subjects made behavioral responses on a fiber Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/13/7/1006/1759185/089892901753165908.pdf by guest on 18 May 2021 Thornton, 2000; Rensink, 1998). Some subjects report optic response box. First, each subject pressed a button the feeling that something is changing in the display with the left hand when he or she felt that there was before they can explicitly identify the specific object that something changing on the trial but could not identify it. is changing. Catch trials and conservative response-time This feeling of covert recognition is described as ‘‘sens- criteria validate the accuracy of the subjects’ sensing ing’’ of change and is reported by a minority of subjects reports. In the current experiment, we provided subjects (Rensink, 1998). Second, each subject pressed a button with the option for making sensing responses, but it was with the right hand when the specific change could be rarely used by the subjects in our sample. Previous identified. Four trials were randomly selected to be reports from behavioral studies suggest that about 35% catch trials with no change present; no subject identified of individuals report sensing change (Rensink, 1998). a change occurring on those trials. Only two of the Although our results could not address sensing directly, subjects in our group reported sensing of change on investigation of sensing of change may provide a techni- more than 20% of the trials, so no additional analyses of que for examining implicit perception. the sensing responses were conducted. All analyses reported in this manuscript investigate the identification of change (right-button responses). Subjects were al- METHODS lowed to move their eyes freely during the experimental trials, with the one instruction that they were to keep Subjects their eyes on the display at all times. Ten young adults participated in the study, which was approved by the Institutional Review Board of the Duke fMRI Imaging Parameters University Medical Center. fMRI scans were conducted on a 1.5-T GE NVi SIGNA scanner with 41 mT/m gradients for fast echo-planar Stimuli imaging. Image transfer and reconstruction was con- On each trial, two versions of the same image were ducted using a GE Advanced Development Workstation. presented in alternation separated by a short-duration A vacuum-pack system restricted head motion without grayscale mask. Images were photographs of complex compromising patient comfort. Axial slices, chosen par- scenes acquired from publicly available image libraries. allel to the line connecting the anterior and posterior The grayscale masks were individually generated for commissures, were selected in each subject following each image pair to match for mean luminance level. initial sagittal structural imaging (2-D SPGR; nine slices Each image or mask subtended about 158 by 108 of around midline). These slices encompassed the entire visual angle, presented against a black background. The cerebral cortex (22–24 slices, 5 mm thick, no skip). High- two images within a pair differed in either the presence/ resolution spin-echo structural images were acquired for absence of a single object, the position of an object, or each slice (in-plane resolution: 0.94 mm2). Functional the color of an object or part of an object. The exper- images were acquired at the same slice locations using imental stimulus set consisted of 120 image pairs and gradient-echo echo-planar imaging (TR: 2000 msec, TE: their associated masks. 40 msec, flip angle: 908, in-plane resolution: 3.75 mm2).

Experimental Design fMRI Preprocessing and Analysis Each subject participated in 10–12 experimental runs Initial preprocessing involved correction for head mo- (mean: 11.1 runs), with each run consisting of 10 tion and temporal order of slice acquisition within a TR, stimulus trials. At the beginning of each run, the subjects using SPM 99 software (Wellcome Department of Cog- viewed a single practice image pair for 12 sec to ensure nitive Neurology, London, UK). Following those correc- baseline equivalence across all trials; the same pair of tions, each subject’s brain was normalized into a images was used at the start of all 12 runs. A fixation common stereotaxic space (MNI 305). No additional cross was next presented at the center of the screen for spatial smoothing was performed. All subsequent analy- 2 sec, followed by the first trial. Each trial consisted of ses used custom MATLAB scripts written by the authors.

Huettel, Gu¨zeldere, and McCarthy 1015

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/089892901753165908 by guest on 24 September 2021 To analyze the fMRI data, we used response-contin- voxel, one for each of the hypotheses tested. The gent event-related analyses. For each trial, we identified number of subjects who had significant activation at an epoch of 24 fMRI time points that included four TRs the locations indicated by the group data is shown in before trial onset and the 20 TRs of the trial itself. Each Table 1. We then used the individual subject t values as epoch was then categorized into one of four bins dependent measures of effect size for each subject, and depending upon when the subject made the change- conducted a t test on those t values. We set the identification response: 0–10, 10–20, 20–30, or 30–40 significance threshold (different from a mean of zero) sec. All trials in each bin were then averaged together, to p < .001. We found that this random-effects analysis

resulting in a set of four response-time–based activation provided similar results to the fixed-effects analysis Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/13/7/1006/1759185/089892901753165908.pdf by guest on 18 May 2021 estimates. These activation estimates were in turn aver- reported above. Only one area, the SFG activation to aged across subjects. We then statistically compared the visual search, was not active in the random-effects activation at each voxel in the brain to a priori exper- analysis at that threshold. Given this correspondence, imental hypotheses, which were generated by convolu- we conclude that our results are generalizable to the tion of an empirically generated fMRI hemodynamic population from which our subject sample was drawn. response with predictions of neural activity. Search- This analysis technique allows segregation of activa- related voxels were identified as those with activity tion components in a manner not possible with short- correlated to an extended waveform, where activation duration tasks. It is difficult, during short-duration tasks, levels rose at task onset to a maximum value (reached at to distinguish hemodynamic responses evoked during a 8 sec) and returned to baseline following the subject’s task (e.g., visual search) from those evoked at its con- response. Response-related voxels were identified as clusion (e.g., motor response). However, task-related those with transient activity beginning at the onset of and response-related activation patterns should change the response bin (i.e., a hemodynamic response with a in different ways as response time increases. Our ap- rise and fall over 16 sec, time-locked to bin onset). proach uses the known variability in subject response Onset-related voxels had transient activity at the start time over the long response interval to identify different of the task, regardless of eventual response bin. More- patterns in voxel activation. over, voxels were classified as having a deactivation pattern if their time courses followed an initial decrease, Acknowledgments followed by a return to baseline after the conclusion of search. The authors thank Jeff Singerman and Jeff Wu for assistance in For each hypothesis tested, we measured the corre- fMRI data analysis, Elizabeth Redcay for assistance in behavioral data analysis, and Josh Wills for creation of the experimental lation of the experimentally observed activation at each stimuli used. We also thank Ron Rensink for comments and voxel with the predicted pattern of activation. The suggestions during the course of this project. This research threshold for voxel significance was set at a mean was supported by NIMH grants MH-05286 and MH-12541 and correlation of r = .63 (t = 3.5, p < .001, for each the Department of Veterans Affairs. response bin) across the four response bins. To control Reprint requests should be sent to Scott A. Huettel, Brain for Type I error, we adopted a minimum cluster-size Imaging and Analysis Center, Duke University Medical Center, threshold, following the calculations reported by For- Box 3918, Durham, NC 27710, or via e-mail: scott.huettel@ man et al. (1995) and by Xiong, Gao, Lancaster, and Fox duke.edu. (1995). Given the approximate number of voxels in our The data reported in this experiment have been deposited in uninterpolated brain volume (17,000) and our uncor- the National fMRI Data Center (http://www.fmridc.org). The rected significance threshold (p < .001), a cluster size of accession number is 2-2001-111T9. three or more voxels controls for Type I error at an overall value of .01. However, we adopted a minimum REFERENCES cluster size of four voxels because of suggestions of Aginsky, V., & Tarr, N. (2000). How are different properties of a disagreements between empirical and theoretical data scene encoded in visual memory? Visual Cognition, 7, 147– at smaller cluster sizes (Xiong et al., 1995). To localize 162. activations, the centroid of each cluster was identified Belger, A., Puce, A., Krystal, J. H., Gore, J. C., Goldman-Rakic, within MNI stereotaxic space (Montreal Neurological P., & McCarthy, G. (1998). Dissociation of mnemonic and Institute), as reported in Table 1. All activation time perceptual processes during spatial and nonspatial working memory using fMRI. Human Brain Mapping, 6, 14–32. courses in Figures 3–6 are averaged across all active Binder, J. R., Frost, J. A., Hammeke, T. A., Bellgowan, P. S. F., voxels within the anatomical region discussed. Rao, S. M., & Cox, R. W. (1999). Conceptual processing We additionally conducted a random-effects analysis during the conscious resting state: A functional MRI study. (e.g., Postle & D’Esposito, 1999) to account for the Journal of Cognitive Neuroscience, 11, 80–93. possibility that significant results in the group-averaged Braver, T. S., Cohen, J. D., Nystrom, L. E., Jonides, J., Smith, E. E., & Noll, D. C. (1997). A parametric study of prefrontal data could be driven by large effects in a few subjects. cortex involvement in human working memory. Neuro- We repeated our hypothesis tests for each subject image, 5, 49–62. individually, resulting in a set of four t values at each Cohen, J. D., Forman, S. D., Braver, T. S., Casey, B. J., Servan-

1016 Journal of Cognitive Neuroscience Volume 13, Number 7

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/089892901753165908 by guest on 24 September 2021 Schreiber, D., & Noll, D. C. (1994). Activation of prefrontal The neural mechanisms of top-down attentional control. cortex in a non-spatial working memory task with functional Nature Neuroscience, 3, 284–291. MRI. Human Brain Mapping, 1, 293–304. Martı´nez, A., Anllo-Vent, L., Sereno, M. I., Frank, L. R., Buxton, Cohen, J. D., Perlstein, W. M., Braver, T. S., Nystrom, L. E., Noll, R. B., Dubowitz, D. J., Wong, E. C., Hinrichs, H., Heinze, H. J., D. C., Jonides, J., & Smith, E. E. (1997). Temporal dynamics & Hillyard, S. A. (1999). Involvement of striate and extra- of brain activation during a working memory task. Nature, striate visual cortical areas in spatial attention. Nature 386, 604–608. Neuroscience, 2, 364–369. Corbetta, M., Akbudak, E., Contouro, T. E., Snyder, A. Z., Ol- McCarthy, G., Blamire, A. M., Puce, A., Nobre, A. C., Bloch, G., linger, J. M., Drury, H. A., Linenweber, M. R., Petersen, S. E., Hyder, F., Goldman-Rakic, P., & Shulman, R. G. (1994). Raichel, M. E., Van Essen, D. C., & Shulman, G. L. (1998). A Functional magnetic resonance imaging of human prefrontal

common network of functional areas for attention and eye cortex activation during a spatial working memory task. Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/13/7/1006/1759185/089892901753165908.pdf by guest on 18 May 2021 movements. Neuron, 21, 761–773. Proceedings of the National Academy of Sciences, U.S.A., 91, Corbetta, M., Kincade, J. M., Ollinger, J. M., McAvoy, M. P., & 8690–8694. Shulman, G. L. (2000). Voluntary orienting is dissociated McCarthy, G., Luby, M., Gore, J., & Goldman-Rakic, P. (1997). from target detection in human posterior parietal cortex. Infrequent events transiently activate human prefrontal and Nature Neuroscience, 3, 292–297. parietal cortex as measured by functional MRI. Journal of Corbetta, M., Miezin, F. M., Dobmeyer, S., Shulman, G. L., & Neurophysiology, 77, 1630–1634. Petersen, S. E. (1990). Attentional modulation of neural McCarthy, G., Luby, M., Gore, J. C., & Goldman-Rakic, P. processing of shape, color, and velocity in humans. Science, (1996). Functional magnetic resonance imaging in a visual 248, 1556–1559. oddball task. Neuroimage, 3, S548. Corbetta, M., Miezin, F. M., Shulman, G. L., & Petersen, S. E. Mesulam, M. (1981). A cortical network for directed attention (1993). A PET study of visuospatial attention. Journal of and unilateral neglect. Annals of Neurology, 10, 309–325. Neuroscience, 13, 1202–1226. Mesulam, M. (1985). Patterns in behavioral neuroanatomy: Corbetta, M., Shulman, G. L., Miezin, F. M., & Petersen, S. E. Association areas, the limbic system, and hemispheric spe- (1995). Superior parietal cortex activation during spatial at- cialization. In M. Mesulam (Ed.), Principles of behavioral tention shifts and visual feature conjunction. Science, 802– neurology. Philadelphia: F. A. Davis. 805. Moran, J., & Desimone, R. (1985). Selective attention gates Courtney, S. M., Petit, L., Maisog, J. M., Ungerleider, L. G., & visual processing in the extrastriate cortex. Science, 229, Haxby, J. V. (1998). An area specialized for spatial working 782–784. memory in human frontal cortex. Science, 279, 1347–1351. Nobre, A. C., Sebestyen, G. N., Gitelman, D., Mesulam, M., Courtney, S. M., Ungerleider, L. G., Keil, K., & Haxby, J. V. Frackowiak, R. S. J., & Frith, C. D. (1997). Functional locali- (1997). Transient and sustained activity in a distributed zation of the system for visuospatial attention using positron neural system for human working memory. Nature, 386, emission tomography. Brain, 120, 515–533. 608–611. O’Regan, J. K., Deubel, H., Clark, J. J., & Rensink, R. A. (2000). D’Esposito, M., Detre, J. A., Alsop, D. C., Shin, R. K., Atlas, S., & Picture changes during blinks: Looking without seeing and Grossman, M. (1995). The neural basis of the central ex- seeing without looking. Visual Cognition, 7, 191–211. ecutive system of working memory. Nature, 378, 279–281. Pashler, H. (1988). Familiarity and visual change detection. Felleman, D., & Van Essen, D. (1991). Distributed hierarchical Perception and Psychophysics, 16, 283–290. processing in the primate cerebral cortex. Cerebral Cortex, Paus, T. (1996). Location and function of the human frontal eye 1, 1–47. field: A selective review. Neuropsychologia, 34, 475–483. Fernandez-Duque, D., & Thornton, I. M. (2000). Change de- Phillips, W. A. (1974). On the distinction between sensory tection without awareness: Do explicit reports underesti- storage and short-term visual memory. Perception and mate the representation of change in the visual system? Psychophysics, 16, 283–290. Visual Cognition, 7, 323–344. Posner, M. I. (1980). Orienting of attention. Quarterly Journal Forman, S. D., Cohen, J. D., Fitzgerald, M., Eddy, W. F., Mintun, of Experimental Psychology, 32, 3–25. M. A., & Noll, D. C. (1995). Improved assessment of signifi- Posner, M. I., & Petersen, S. E. (1990). The attention system cant activation in functional magnetic resonance imaging of the human brain. Annual Review of Neuroscience, 13, (fMRI): Use of a cluster-size threshold. Magnetic Resonance 25–42. in Medicine, 33, 636–647. Posner, M. I., Walker, J. A., Friedrich, F. J., & Rafal, R. D. (1984). Gandhi, S. P., Heeger, D. J., & Boynton, G. M. (1999). Spatial Effects of parietal injury on covert orienting of attention. attention affects brain activity in human primary visual cor- Journal of Neuroscience, 4, 1863–1874. tex. Proceedings of the National Academy of Sciences, Postle, B. R., & D’Esposito, M. (1999). Dissociation of human U.S.A., 96, 3314–3319. caudate nucleus activity in spatial and nonspatial working Gitelman, D. R., Nobre, A. C., Parrish, T. B., Labar, K. S., Kim, Y., memory: An event-related fMRI study. Cognitive Brain Meyer, J. R., & Mesulam, M. M. (1999). A large-scale distrib- Research, 8, 107–115. uted network for covert spatial attention: Further anatomical Rensink, R. A. (1998). Mindsight: Visual sensing without seeing. delineation based on stringent behavioural and cognitive Investigative Ophthalmology and Visual Science, 38, S707. controls. Brain, 122, 1093–1106. Rensink, R. A. (2000a). The dynamic representation of scenes. Haxby, J. V., Ungerleider, L. G., Horwitz, B., Rapoport, S. I., & Visual Cognition, 7, 17–42. Grady, C. L. (1995). Hemispheric differences in neural sys- Rensink, R. A. (2000b). Seeing, sensing, and scrutinizing. Vision tems for face working memory: A PET-rCBF study. Human Research, 40, 1469–1487. Brain Mapping, 3, 68–82. Rensink, R., O’Regan, J., & Clark, J. (1997). To see or not to see: Heinze, H., Mangun, G., Burchert, W., Hinrichs, H., Scholz, M., The need for attention to perceive changes in scenes. Munte, T., Gos, A., Scherg, M., Johannes, S., Hundeshagen, Psychological Science, 8, 368–373. H., Gazzaniga, M., & Hillyard, S. (1994). Combined spatial Shulman, G., Fiez, J., Corbetta, M., Buckner, R., Meizin, F., and temporal imaging of brain activity during visual selective Raichle, M., & Petersen, S. (1997). Common blood flow attention in humans. Nature, 372, 543–546. changes across visual tasks: II. Decreases in cerebral cortex. Hopfinger, J. B., Buonocore, M. H., & Mangun, G. R. (2000). Journal of Cognitive Neuroscience, 9, 648–663.

Huettel, Gu¨zeldere, and McCarthy 1017

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/089892901753165908 by guest on 24 September 2021 Simons, D. J., & Levin, D. T. (1997). Change blindness. Trends Ungerleider, L. G., & Mishkin, M. (1982). Two cortical visual in Cognitive Sciences, 1, 261–267. systems. In D. J. Ingle, M. A. Goodale, & R. J. W. Mansfield Smith, E. E., & Jonides, J. (1997). Working memory: A view (Eds.), Analysis of visual behavior (pp. 549–586). Cam- from neuroimaging. Cognitive Psychology, 33, 5–42. bridge: MIT Press. Somers, D. C., Dale, A. M., Seiffert, A. E., & Tootell, R. B. H. Votaw, J. R., Faber, T. L., Popp, C. A., Henry, T. R., Trudeau, J. (1999). Functional MRI reveals spatially specific attentional D., Woodard, J. L., Mao, H., Hoffman, J. M., & Song, A. W. modulation in human primary visual cortex. Proceedings of (1999). A confrontational naming task produces congruent the National Academy of Sciences, U.S.A., 96, 1663–1668. increases and decreases in PET and fMRI. Neuroimage, 10, Ungerleider, L. G. (1995). Functional brain imaging studies of 347–356. cortical mechanisms for memory. Science, 270, 769–775. Xiong, J., Gao, J-H., Lancaster, J. L., & Fox, P. T. (1995).

Ungerleider, L. G., & Haxby, J. V. (1994). ‘‘What’’ and ‘‘where’’ Clustered pixels analysis for functional MRI activation Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/13/7/1006/1759185/089892901753165908.pdf by guest on 18 May 2021 in the human brain. Current Opinion in Neurobiology, 4, studies of the human brain. Human Brain Mapping, 3, 157–165. 287–301.

1018 Journal of Cognitive Neuroscience Volume 13, Number 7

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/089892901753165908 by guest on 24 September 2021