Temporal characteristics of neural activity related to target detection during visual search

www.elsevier.com/locate/ynimg NeuroImage 33 (2006) 296 – 306

Temporal characteristics of neural activity related to target detection during visual search Tomoe Hayakawa,a,b,⁎ Norio Fujimaki,a and Toshihide Imaruoka c a

Biophysical ICT Group, National Institute of Information and Communications Technology, 588-2, Iwaoka, Iwaoka-cho, Nishi-ku, Kobe 651-2492, Japan Department of Psychology, Teikyo University, 359 Otsuka, Hachioji-shi, Tokyo 192-0395, Japan c College of Informatics and Human Communication, Kanazawa Institute of Technology, 7-1 Ohgigaoka, Nonoichi, Ishikawa 921-8501, Japan b

Received 19 March 2005; revised 30 January 2006; accepted 21 June 2006 Available online 21 August 2006

A previous MEG study on neural activities during the orientation singleton search showed that both efficient and inefficient searches shared a common neural network and the search efficiency was determined by a neural process executed in the temporal and parietal areas. The target segmentation stage, however, remains to be elucidated. In the present study, MEG and fMRI experiments were conducted, and moment-magnitudes of equivalent current dipoles were estimated with an fMRI-constrained MEG multi-dipole method to obtain differences between target-present and -absent conditions in each brain region for the whole time course. The dipole moments around the calcarine sulcus (CaS) and posterior fusiform gyrus (pFuG) increased at latencies around 70–350 ms. Activity around the CaS consisted of a prominent and a subsequent smaller but still obvious peak (117, 215 ms); the first peak showed no difference between conditions, while the second peak was significantly larger in the target-present condition. Activity around the pFuG had a prominent peak (125 ms) and subsequent small activity (237 ms), whereas the target's presence or not had no influence on either activity. The activity of the right intraparietal sulcus (IPS) was significantly larger than that for the left IPS at latencies around 196 ms irrespective of the target's presence or not. The activity of the other brain regions such as the posterior superior temporal sulcus, cingulate sulcus and central sulcus showed no difference between target conditions. The results demonstrate that neural activities of multiple regions had different temporal characteristics, and the later activity around the CaS was related to the target segregation from its surroundings during the orientation contrast search. © 2006 Elsevier Inc. All rights reserved.

Introduction The visual search is a fundamental task that requires segregation of a target from its surroundings. It is known that certain visual ⁎ Corresponding author. Biological ICT Group, National Institute of Information and Communications Technology 588-2, Iwaoka, Iwaoka-cho, Nishi-ku, Kobe 651-2492, Japan. Fax: +81 78 969 2279. E-mail address: [email protected] (T. Hayakawa). Available online on Direct (www.sciencedirect.com). 1053-8119/$ - see front matter © 2006 Elsevier Inc. All rights reserved. doi:10.1016/j.neuroimage.2006.06.034

elements pop out: that is, they are immediately detected without any effort, while others take considerable time. In terms of behavioral data, a target is detected efficiently or inefficiently; the reaction time for detecting a target is independent of the number of stimulus items in the efficient search, but prolonged as the number of items increases in the inefficient search (Nothdurft, 1991; Treisman and Gelade, 1980; Treisman and Souther, 1985; Wolfe et al., 1989). Models that include parallel processing and serial attentional processing have been proposed to explain the mechanisms of the efficient and inefficient searches. However, evidence running counter to this explanation has also been found; although a conjunction search has been considered to require a serial attentional shift on every item for feature integration and was expected to occur in a serial processing way, the dependence of reaction time on the item number showed that the search was not necessarily inefficient (Duncan and Humphreys, 1989; Nakayama and Silverman, 1986). Based on these findings, a combination of a parallel and serial mechanism was proposed (Grossberg et al., 1994; Wolfe et al., 1989, 2003). For example, the “guided search” model assumes a single search system comprised of two stages: an early parallel feature processing stage and a late serial processing stage. Multiple kinds of feature information are pooled with different weights in the late stage, and the sum of information guides attention for target detection to make the search efficient. In an elementary task of detecting an orientation contrast, a line element that differs in its orientation angles from surrounding lines pops out (Nothdurft, 1991, 2000; Treisman and Gormican, 1988). Many studies have been devoted to revealing the neural basis underlying such an orientation contrast search by recording physiological data from cats and macaque monkeys (Kastner et al., 1997; Knierim and van Essen, 1992; Zipser et al., 1996). Neural activities elicited by a line element presented in the receptive field were compared for two conditions: the line had a different orientation from its surroundings and the line had a homogenous orientation to its surroundings. These studies showed that neural activities in the primary visual cortex were more activated in the former condition than in the latter condition. This suggested that the primary visual cortex has a neural basis for target segregation during

T. Hayakawa et al. / NeuroImage 33 (2006) 296–306

an efficient singleton search, and also that this could be considered the neural basis of pop-out. The event-related potentials (ERPs) have been recorded from human subjects, who executed texture segregation by detecting orientation contrasts for the constituting line segments. They had a negative component over the posterior pole at peak latencies around 161–225 ms, which was considered to be associated with figureground segmentation (Bach and Meigen, 1992, 1997). In this respect, two prominent negative components at mean latencies of 150 and 230 ms were also reported for the orientation contrast (Caputo and Casco, 1999). The latter component was considered to be associated with figure-ground segmentation because the latency and amplitude were sensitive to the discriminability of the figure. The ERPs recorded from both humans and monkeys also showed that the difference in potentials related segmentation appears at peak latencies around 200 ms. It was demonstrated that the equivalent current dipole for this response could be localized in the primary visual cortex (Lamme et al., 1992). In a light of a different situation of target detection during a visual search, the ‘N2pc’ component occurring at latencies around 200–300 ms was found to be closely connected with attentional shifts to a particular feature of a conjunction target that required feature integration (Luck and Hillyard, 1994; Woodman and Luck, 2003). The neural sources of N2pc were revealed by magnetoencephalography (MEG) to be in the parietal region at latencies of 180–200 ms and the occipito-temporal region at latencies of 220–240 ms (Hopf et al., 2000). Interestingly, this component was also detected when subjects discriminated the orientation contrasts without feature integration during the visual search; the amplitude increased in the target-present trials as the item number in a search array increased (Schubö et al., 2004). Although these studies did not focus on the target segmentation for an orientation singleton search, they suggested that activities related to the figure or target segregation from its background appeared around the occipital cortices at late latency ranges. We previously conducted an MEG experiment to investigate the sequence of neural activities during the orientation singleton search, i.e., a visual search in which the arrays consisted of line elements (Hayakawa et al., 2003). Five magnetic components (M1–5) were detected in both efficient and inefficient searches at latencies of 100– 350 ms. The results suggested that both processes shared a common neural system; after early feature processing around the calcarine sulcus (CaS) in M1 and posterior fusiform gyrus (pFuG) in M2, visual information might be modulated by activities of the intraparietal sulcus (IPS) and the posterior superior temporal sulcus (pSTS) in M3, after which the information is processed again in the areas around the CaS of the same hemisphere as the early processing in M4. Although the sources were not successfully estimated, the P300m-like component (M5) appeared after these activities. There was a difference between the magnetic responses of the efficient search and inefficient search at latencies around 200 ms (M3), whereas no such difference was observed in the early visual responses. Although the M4 component in the occipital area may be related to the target segregation as suggested by the related reports mentioned before, there was no significant difference between the M4 amplitudes in the target-present trials and those in the targetabsent trials. However, it should be remarked that our previous analysis based on the magnetic field strength could be limited by overlaps of multiple neural activities in the measured field. Several human studies with functional magnetic resonance imaging (fMRI) have been devoted to understanding the brain regions related to the target segregation. They strongly suggested

297

that the function was executed in the occipital cortices (Ferber et al., 2005; Kastner et al., 2000; Skiera et al., 2000; Schira et al., 2004). However, the temporal resolution was limited by the hemodynamic delay to show the sequence of search processes including target segregation. Therefore, in the present study, in order to clarify how activity related to target segregation could appear in the neural network involving occipital, parieto-temporal cortices during an orientation contrast efficient visual search, we conducted MEG and fMRI measurements, which are complementary brain imaging techniques having high temporal and spatial resolution, respectively, and performed the MEG multi-dipole analysis, which uses the coordinates of the activation detected in the fMRI experiment for placing the dipoles (Fujimaki et al., 2002). Materials and methods Subjects Six normal volunteers (four males and two females, 27–46 years old, right-handed) participated in the study. All subjects had normal or corrected-to-normal visual acuity and a normal visual field. They were informed of the purpose and procedure of the experiments and gave their written consent before the experiment. This study was approved by the Ethical Committee for Human and Animal Research of the National Institute of Information and Communications Technology. Stimuli and task Fig. 1 illustrates the stimulus arrays, which consisted of line elements. The array for the target-present trials included an oblique (45°) bar as the target among vertical bars as distractors. In the target-absent trials, all the items were vertical bars. The reaction time for this oblique bar among vertical bars combination was previously reported to be independent of the number of items (Treisman and Gormican, 1988). An array of 6 × 3 bars, which extended 8° vertically × 4° horizontally, was presented in the left or right visual field around a fixation point. The stimuli were projected on a screen located 50-cm away from the subject. Each bar was 0.7° long and 0.1° wide. The bars were separated by 1.6° from each other with a jitter of 0.4° randomly in both horizontal and vertical directions. The target appeared at a random location in

Fig. 1. Example of stimulus arrays consisting of 6 × 3 vertical and oblique bars for the target-present (A) and target-absent (B) conditions in the left visual field stimulation. The target appeared at random locations in the middle band (the shadowed part itself was not visible to participants) of the array with a 2.5° eccentricity from the fixation point in the target-present trials.

298

T. Hayakawa et al. / NeuroImage 33 (2006) 296–306

the middle band, which is shaded in Fig. 1. The luminance was 83 cd/m2 for the bar and 1 cd/m2 for the background. In the MEG experiment, target-present trials and target-absent trials were executed in a random order with a probability of 50% in the two sessions where stimuli were presented in the left or right visual field, and the stimulus array was presented for 50 ms, with intervals varied randomly in a range of 1000 to 1400 ms. In the fMRI experiment, four tasks, “Efficient search (an oblique bar among vertical bars)”, “Inefficient search (a vertical bar among oblique bars)”, “Rest” and “Fixation”, were executed during the left and right visual field stimulation. During a scan (4 s), the stimuli (1000 ms) and blank (1000 ms) were sequentially presented twice in the search trial. The target-present and -absent trials occurred with a probability of 70% and 30%, respectively, for the search condition. Between the search and fixation condition, a nullevent (rest) was provided to avoid spill-out of activation from the search condition to the fixation condition due to hemodynamic delay. Since the present analysis required MEG and fMRI data for the same condition (“Efficient search”), the stimuli for “Inefficient search” in fMRI were presented but the data for the condition were not used for the analysis. During the task periods, subjects were required to press either of two buttons with the index and middle finger of their right hand, respectively, depending on a judgment of whether a target was present or not in the search array. MEG recordings A 148-channel whole-head MEG system (Magnes® 2500WH, Biomagnetic Technologies Inc.) was used for the recordings. Magnetic signals were digitized at a sampling rate of 678 Hz. To remove epoch data that contained eye movements and blinks, the threshold peak-to-peak value for artifact rejection was determined to be 3 to 5 pT, depending on the subjects. Data of one hundred epochs for each of the target-present and target-absent conditions were selectively averaged and filtered with a pass band of 1 to 40 Hz to remove slow shifts and high-frequency noises. The epoch duration was 980 ms with a pre-trigger period of 250 ms. fMRI data acquisition and analysis fMRI data were acquired with a 1.5 T magnetic resonance imaging system (Vision®, Siemens). Functional images with thirtytwo axial slices were taken in each scan using echo-planar imaging (TE: 55.2 ms, TR: 4.0 s, flip angle: 90°). Each slice consisted of 128 × 128 pixels with a size of 4 × 4 mm and a thickness of 4 mm with no gap. All four tasks (“Efficient search”, “Inefficient search”, “Rest” and “Fixation”) were repeated four times per session. Thus, sixteen blocks were executed, where ten scans were taken in one block, and the visual stimuli were given twice in each scan period (4 s). The fMRI data for only “Efficient search” and “Fixation” were analyzed and contrasted so they could be used in the MEG multi-dipole analysis described later. A whole-head T1-structural image, whose voxel size was 1 × 1 × 1 mm, was taken of each subject in order to co-register functional images and to make a realistically shaped brain model for the multi-dipole analysis. After standard preprocessing (motion correction, co-registration to the T1-structural image and spatial smoothing with a full-width half-maximum value of 8 mm), the statistical analysis of the block design was performed using SPM99 (http://www.fil.ion.ucl.ac.uk/ spm; Wellcome Department of Cognitive Neurology, London UK)

using time constants of twice the fundamental periods of time course and typical hemodynamic response delay (6 s). To detect the difference between “Efficient search” and “Fixation” in each subject, the statistical threshold corrected for multiple comparisons and the extent threshold were set to be Pc < 0.05 and k = 8 or 12 depending on the subjects. fMRI-constrained MEG multi-dipole analysis We analyzed the MEG data with a multi-dipole estimation (Fujimaki et al., 2002). For this analysis, MEG current dipoles were placed in the areas of significant activation detected by fMRI. These activation volumes were then divided into sub-volumes whose radial and tangential lengths were less than 2 cm, and a single dipole was placed at the center of each sub-volume. When no activation was found around the CaS, pFuG, pSTS, IPS and anterior cingulate gyrus (aCgG), whose activation was suggested by the single dipole analysis in the previous MEG study (Hayakawa et al., 2003), dipoles were added manually to these brain regions while taking account of the anatomical structures. This was intended to complement fMRIinvisible dipoles, that is, MEG sources for which fMRI failed to show activation. The magnitudes and directions of these dipole moments were fitted to the measured magnetic fields with the software (ASA®, A.N.T. software BV) using a model with a realistic brain shape. The significant dipoles were selected using a Rayleigh distribution function to which the moment of the noise during the pre-trigger period obeyed and a significant level (Pc < 0.01) with a Bonferroni correction for multiple comparisons of sampling points in latency and number of dipoles. Significant dipoles were grouped if they were located within 2 cm and had common significant latencies. Then, if the grouped dipoles located within 4 cm showed a high temporal cross-correlation coefficient of larger than 80%, they too were grouped. This two-step grouping was executed to combine dipoles that were inseparable due to crosstalk phenomena during fittings. The moment-magnitudes normalized by the significance values (Pc = 0.01) determined for each subject were used for comparison between conditions. A paired t test was carried out for the normalized moment-magnitudes and peak latencies to test the difference between target-present and target-absent trials. To rule out apparent activities due to crosstalk phenomena, we calculated the contribution ratio, i.e., 1 minus the ratio of the sum of the squared difference between the measured and reconstructed magnetic fields to the sum of squared measured fields for each group dipole, where the sum was taken for all sensors for which the reconstructed fields were larger than half the peak reconstructed fields. Grand-average contribution ratios were also calculated in time bins every 20 ms to assess the tendency of the contribution in each brain region. They were expressed as contribution ratios for a 20-ms bin. Results Brain activations detected by fMRI As shown in Fig. 2, multiple cortical regions such as the parietal, temporal and the left occipital visual cortices were involved in target detection during the visual search task, when visual stimuli were presented in the right visual field. In all subjects (Table 1), significant activations were detected around the CaS and pFuG of the hemisphere contralateral to the

T. Hayakawa et al. / NeuroImage 33 (2006) 296–306

299

Fig. 2. fMRI activation in the right visual field stimulation (Subject 1) (p < 0.05 corrected). The activations were divided into the sub-volumes (less than 2 cm), whose coordinates were used for locating dipoles.

636.0 ± 32.4 ms for the target-absent trial with a mean hit rate of more than 95%. MEG waveforms had multiple components before 300 ms and late components such as those related to the button press (Fig. 3), and the contour map of the major component had a similar spatial distribution for all six subjects. The normalized moment-magnitudes of the grouped dipoles located around the CaS in each data showed that this area was continually activated at latencies from 70 ms to 350 ms after stimulus onset (Fig. 4A). The grand-average was obtained by averaging the normalized moment-magnitudes across six subjects, the target-present and target-absent trials and the left and right halffield stimulations. The activity was more than one, that is, it was significantly larger than the noise value (pc < 0.01), at latencies from 84 ms to 257 ms and from 294 ms to 307 ms (n = 24). Since the grand-average curve in the former latency range was composed of two prominent peaks, i.e., a large one and a second small one, the peaks for each data were classified into two time windows: 80– 180 ms for the first peak and 180–260 ms for the second peak. In

stimulus presentation. The activation was minimal around the CaS in some cases on account of the small items in an array and the blank during a scan. Activation around the IPS, superior parietal lobule (SPL) and pSTS in both hemispheres and around the aCgG was observed in most subjects, whereas the pSTS tended to be more activated in the right hemisphere than in the left hemisphere. The areas around the left central sulcus (CS), which were related to a button press with the right index or middle finger, showed intensive activity for all subjects. The right precentral gyrus (preCG) was activated in four subjects. Activation was also obtained around the frontal eye field, posterior cingulate gyrus and angular gyrus in a few subjects.

MEG results obtained with the multiple dipole estimation During the MEG experiment, mean reaction time and standard deviation were 553.5 ± 33.9 ms for the target-present trial and Table 1 Brain areas for the fMRI activation in each subject Subject 1 2 3 4 5 6

Visual field

CaS Contralateral

PFuG Contralateral

IPS/SPL L

R

L R L R L R L R L R L R

• • • • • • • • • • • •

• • • • • • • •

• • • •

• • • •

• • •

• • • • •

• • •

pSTS L • •

aCgG

CS L

preCG R

•

• • • • • • • • • • • •

• •

R • • • •

• •

• • • • • • • •

• • • • • •

Dipole locations were fixed in these areas. When no activation was found around the CaS, pFuG, pSTS, IPS and aCgG, whose activation was suggested by the single-dipole analysis in the previous MEG study (Hayakawa et al., 2003), dipoles were manually added to these brain regions while taking account of the anatomical structures.

300

T. Hayakawa et al. / NeuroImage 33 (2006) 296–306

Fig. 3. Representative MEG waveforms averaged across 100 epochs that were recorded by 148 sensors in the target-present trial from a subject. Multiple components were detected at latencies from 75 ms to 300 ms, and P300-like magnetic fields were found in this subject. After these components, some activities related to the button press with the right finger were distributed around the left CS.

each time window, the peak latency and peak value of the normalized moment-magnitude were detected for each of the target conditions and stimulus sides for each subject. For the first time window, the normalized moment-magnitude had the largest peak in

all data. Mean and standard error of the peak latency were 117.8 ± 1.7 ms, and those of the peak moment-magnitude were 3.3 ± 0.2 (n = 24) for the first time window and 215.8 ± 4.4 ms and 1.8 ± 0.1 (n = 23) for the second time window. Mean and standard error for the maximum value of the contribution ratios were 46.1 ± 4.6% for the first time window and 36.6 ± 4.3% for the second one, whereas the maximum values among subjects were 82% and 64% and the contribution ratios for the 20-ms bin were 31.6% and 18.8%, respectively. The moment-magnitude of the first peak was significantly larger than that of the second one (paired t test, p < 0.01). The second peak appeared 98.3 ± 4.1 ms after the first peak appeared (Fig. 4B). The effect of the target' presence was investigated by direct comparison of the target conditions. For the second peak, the moment-magnitude was significantly larger for the target-present trial (2.0 ± 0.2) than for the target-absent trial (1.6 ± 0.1) (paired t test, p < 0.02), while there was no significant difference for the first large peak, where the moment-magnitude was 3.4 ± 0.3 for target-present and 3.2 ± 0.3 for target-absent (Fig. 7A). There were no significant differences in peak latencies between target conditions for either peak. The normalized moment-magnitudes of the grouped dipoles located around the pFuG showed that this area was also activated at latencies from 75 ms to 350 ms (Fig. 5). The grand-average moment-magnitude showed significant activation at latencies of 90 to 180 ms with one prominent peak (n = 24). Although it was slightly lower than the significance level, the activity around the pFuG continued until the latencies around 300 ms. Thus, in addition to the first time window of 80–180 ms, the second time window, 200–300 ms, was used for the following analysis. The first time window showed the largest moment-magnitude peak in all data. The mean and standard error of peak latency and peak moment-magnitude were 125.6 ± 1.2 ms and 3.0 ± 0.3 (n = 24),

Fig. 4. (A) Normalized moment-magnitudes of dipoles located around the CaS (n = 24). Black lines indicate the moment-magnitude elicited by targetpresent and target-absent trials for the left and right visual field stimulation in each data, while the red line shows the grand-average of the normalized moment-magnitudes averaged across all data sets. The gray scale represents the contribution ratio for the 20-ms bin. (B) Peak latency for the first and second peak in each data set. After the first peak, the second peak appeared with a mean delay of 98.3 ms.

T. Hayakawa et al. / NeuroImage 33 (2006) 296–306

Fig. 5. Normalized moment-magnitudes of dipoles located around the pFuG (n = 24). Black lines indicate the moment-magnitude elicited by targetpresent and target-absent trials for the left and right visual field stimulation, and the red line shows the grand-average of the normalized momentmagnitudes averaged across all data sets. The gray scale represents the contribution ratio for the 20-ms bin.

respectively. For the second time window, 18 of 24 cases showed a small single peak. Their peak latency and moment-magnitude were 237.1 ± 5.7 ms and 1.5 ± 0.1 (n = 18). The second peak appeared 112.7 ± 5.2 ms after the first peak. The maximum contribution ratios of the first and second time windows were respectively 33.9 ± 3.6% and 26.8 ± 2.4%, whereas the contribution ratios for the 20-ms bin were 23.6% and 14.5%. For both peaks, however, there were no significant differences in the peak moment-magnitude between target-present and target-absent conditions (Fig. 7A): target-present 3.2 ± 0.5, target-absent 2.8 ± 0.3 for the first peak and target-present 1.5 ± 0.2, target-absent 1.4 ± 0.1 for the second peak. The peak moment-magnitude for the pFuG in the first time window appeared just after the first peak of the CaS in 20 of 24 data sets, and there was a significant difference between the first peak latency of pFuG and the first peak latency of CaS (paired t test, p < 0.01). The mean and standard error of difference in peak latencies were 10.3 ± 1.1 ms, where we neglected four cases in which peak latency was shorter for the pFuG than for the CaS. The left and right IPS were continually activated at latencies around 75 ms to 400 ms showing multiple peaks (Fig. 6A). The grand-average moment-magnitude showed significant activity at latencies of 105–160 ms, 206–240 ms and 277–294 ms for the left IPS (n = 24) and 103–230 ms for the right IPS (n = 24). For the following analysis, these activations were separated into three time windows (90–160 ms, 160–260 ms and 260– 300 ms). In the first time window, a peak over the noise value was detected in almost all data, and their peak latency and moment-magnitude were 119.1 ± 3.1 ms and 2.2 ± 0.2 for the left IPS (n = 17) and 125.8 ± 3.4 ms and 2.1 ± 0.1 for the right IPS (n = 17). However, the peak contributions were 18.7 ± 3.0% and

301

21.0 ± 3.1% for the left and right IPS, whereas the CaS and FuG showed higher peak contribution ratios in similar latency ranges (46.1% and 33.9%, respectively). The contribution ratio for the 20-ms bin was as low as 11% for both IPS. In the second time window, the right IPS had a peak in 21 out of 24 data with peak latency of 196.9 ± 3.9 ms and moment-magnitude of 2.0 ± 0.2, whereas the left IPS had a small peak in 20 out of 24 data with peak latency of 212.7 ± 4.8 ms and momentmagnitude of 1.5 ± 0.1. The activity for the right IPS was significantly larger than for the left IPS (paired t test, p < 0.01). However, the direct comparison between the left and right visual field stimulation and between the target-present and target-absent trials showed no significant differences. Thus, irrespective of stimulus conditions, the moment-magnitudes were larger for the right IPS than for the left IPS (Figs. 7A and B). The contribution ratio for the right IPS, 32.7 ± 3.3%, was also higher than those for the left IPS, 27.8 ± 3.5%. These contributions for the second time window were better than that for the first time window. In the third time window, peak latency and moment-magnitude were 283.7 ± 2.6 ms, 1.4 ± 0.1 for the left IPS (n = 13) and 281.7 ± 3.1 ms, 1.5 ± 0.1 for the right IPS (n = 14). The peak contribution ratio was 21.0 ± 2.6% for the left IPS and 24.1 ± 3.2% for the right IPS, and the contribution for the 20-ms bin was only 13% for both IPS during this period. The left and right pSTS were also activated at latencies around 75 ms to 400 ms after stimulus onset; however, the grand-average moment-magnitude showed significant activity only at latencies of 108–156 ms for the left pSTS and 128– 162 ms for the right pSTS (Fig. 6B). Although the grandaverage did not exceed the noise value, significant activations were detected during the subsequent latencies in some cases. Thus, in addition to the time window at 90–160 ms, the second time window at 160–300 ms was used for looking into target effects and laterality. For the first time window, peak latency and moment-magnitude were 123.9 ± 3.0 ms and 2.2 ± 0.2 for the left pSTS (n = 14) and 134.9 ± 3.7 ms and 1.7 ± 0.1 for the right pSTS (n = 19), and their maximum contribution ratios were 30.5 ± 3.5% and 36.4 ± 4.1% respectively. The contribution ratio for the 20-ms bin was 13% for both pSTS. Although we did not acquire sufficient data for the second time window, peak latency and moment-magnitude were 218.5 ± 6.8 ms, 1.5 ± 0.1 for the left pSTS (n = 13) and 217.9 ± 6.1 ms, 1.6 ± 0.2 for the right pSTS (n = 13). The peak contribution ratios for the left and right pSTS were 31.0 ± 3.5% and 35.7 ± 4.4% in this time window. Although there was a tendency that activation was larger in the right pSTS than in the left pSTS (paired t test, p < 0.03), no significant difference existed between target conditions (Figs. 7A and B). The peak moment-magnitude for the left and right pSTS occurred mainly after the peak of the right IPS (196 ms), and it overlapped active latencies of the second peak around the occipital cortices. For the other brain regions, such as the aCgG and CS, the grand-average moment-magnitude showed no significant activation. For the aCgG (Fig. 8A), the normalized moment-magnitudes were classified into the two time windows: 80–180 ms for the first peak and 180–260 ms for the second peak. The momentmagnitude was significantly large in 15 of 24 data for the first time window and in 14 of 24 data for the second time window. Their peak latency and moment-magnitude were 118.3 ± 4.5 ms, 1.6 ± 0.2 and 215.1 ± 5.2 ms, 1.5 ± 0.1. The maximum contribution ratio for the second time window (24.4 ± 3.5%) was higher than that for the first time window (16.5 ± 4.3%). However, no significant difference between target conditions was indicated

302

T. Hayakawa et al. / NeuroImage 33 (2006) 296–306

Fig. 6. Normalized moment-magnitudes of dipoles located around the IPS (A) and pSTS (B) in both hemispheres (n = 48). Black lines indicate the momentmagnitudes elicited by target-present and target-absent trials for the left and right visual field stimulation. The red and white line show the grand-average of the normalized moment-magnitudes for the right and left hemispheres, respectively. The gray scale represents the contribution ratio for the 20-ms bin, for the right and left IPS or pSTS. The upper bar graphs are for the right IPS and pSTS, and the lower ones are for the left IPS and pSTS.

for either time window. Although the group dipole around the left CS (Fig. 8B) was expected to activate in relation to the button press, the grand-average moment-magnitude showed no significant activation. This may be due to the variation in reaction time for each epoch. Only the contribution ratio for the 20-ms bin showed high value during the late latency range before button pressing. The other fMRI-constrained dipoles, such as those located in the frontal eye field and posterior cingulate gyrus, showed no significant activities. Discussion The major object of the present work is to clarify the activity of neural network specific to target segregation during singleton (orientation contrast) efficient search processing. In the present analysis, the coordinates of activations detected by fMRI were used to place the dipoles, while MEG data were used to estimate the temporal changes of dipole moments. The analysis enabled us to detect multiple neural activities in comparison with the traditional single-dipole analysis. The brain-imaging studies on the visual search using fMRI and positron emission tomography (PET) showed that the frontal eye field, ventral prefrontal cortices, aCgG, posterior parietal cortices, posterior temporal cortices and the visual cortices were activated in relation to the search tasks (Corbetta et al., 1995; Donner et al., 2002, 2003; Leonards et al., 2000; Nobre et al., 2002, 2003). Several reports revealed that the cerebral networks involved in efficient and inefficient searches overlap each other, except that the frontal eye field and superior frontal region were

involved only in the inefficient search (Leonards et al., 2000), and search efficiency affected the activation in the parietal areas along the IPS (Nobre et al., 2003). In the present study, of the efficient orientation contrast singleton search task, we also detected significant activations by fMRI around the CaS, pFuG, pSTS, IPS, and CS, which corresponded with those in the previous studies except for activation around the CS, which was related to the button press. Although we could place several dipoles around the CaS, no large activation around the CaS was induced in the fMRI data of the present study. Since the visual cortex (V1) has shown a smaller response in salient stimuli than in non-salient stimuli during the orientation contrast search (Hopf et al., 2004), the small activation of the CaS in our study might be attributable to the efficient orientation singleton search (i.e., salient search condition). Furthermore, a major transient activity with short duration in comparison with 2-s repetition time expected to occur in the CaS is considered to have only a small fMRI signal due to the slow hemodynamics. These could be the reasons for the small activation around the CaS in this study. The main results of the present MEG experiment demonstrated that the moment-magnitude of the CaS has two prominent peaks. Activation around the CaS was increased in the early period and again in the late period, the first peak at latencies around 117 ms and the second peak at latencies around 215 ms. Our previous study also showed activity twice in the occipital areas; sources were detected in the CaS and the occipital cortices (CaS, cuneus and lingual gyrus) in similar latency ranges (Hayakawa et al., 2003). Although some differences existed between the present and previous studies, i.e., half-field stimuli and multi-dipole estimation

T. Hayakawa et al. / NeuroImage 33 (2006) 296–306

Fig. 7. Mean of peak moment-magnitudes of dipoles located around the CaS, pFuG, IPS and pSTS. Shaded bars indicate the activity in the target-present condition, and white bars indicate the activity in the target-absent condition. For the IPS and pSTS, cross-hatches denote activity in the left hemisphere and hatches denote activity in the right hemisphere during the 160–260 ms time window. Error bars indicate standard errors. Asterisks indicate significant differences (**p < 0.01, *p < 0.02).

(present) vs. full-field stimuli and single-dipole estimation (previous), the significant latencies suggested that the present first and second peak correspond to the previous first and fourth component. It was interesting that the moment-magnitudes of the present second peak of the CaS showed a significant difference between the target-present and target-absent trials and the previous second one (fourth component) had a tendency to show the same difference, whereas both the present and previous first activity of the CaS showed no significant difference between the target conditions. In order to detect a target, the stimulus feature processing and target segregation were sequentially required. Since the activation related to the feature detection is expected to occur in the early latency range and activation related to the decision as to the target's presence, i.e., the P300 component, is expected to occur later than the two processes (Hayakawa et al.,

303

2003; Luck and Hillyard, 1990), it is most likely that the present first peak at latencies around 117 ms reflects feature processing while the second peak at latencies around 215 ms reflects the target being segregated from its surroundings in the CaS. Although the second activities around the CaS showed only a tendency rather than a significant difference in the previous report, this may be due to a limitation of previous analysis' method for detecting activity because magnetic fields produced by multiple areas overlapped at late latencies. The present results showing a significant difference in the second peak suggest that the multi-dipole analysis overcomes the limitation. In relation to the feature analysis and target segregation, neural activities in V1 of macaque monkeys were reported to show not only orientation selectivity and coding (Das, 1996; Celebrini et al., 1993) but also modulation from surroundings, where neural activities for a flashed bar presented in a classical receptive field were larger when the surrounding bars had an orthogonal orientation than when they had the same orientation (Knierim and van Essen, 1992). This effect in V1 could be regarded as the neural basis of perceptual pop-out and segregation of texture borders. Furthermore, two distinct responses were found for the figure-ground segmentation in V1 of macaque monkeys: response to orientation texture in the early period and modulated response to segmenting figure from surroundings in the late period (Lamme et al., 2000; Roelfsema et al., 2004; Super et al., 2001; Zipser et al., 1996). Another report showed that the former response occurred at latencies around 80 ms while the latter manifested after 90 ms from the first one (Lamme et al., 1999). Since the electrode positions were not the same, ERPs in awake monkeys and humans showed that the difference between segmentation and uniform stimuli was found at latencies around 200 ms: 240 ms for monkeys and 160 ms for humans (Lamme et al., 1992). In the human subjects, the ERPs recorded at the occipital scalp electrodes indicated that neural mechanism of texture segmentation existed in the primary visual cortex at latencies of 161 to 225 ms (Bach and Meigen, 1992, 1997), and of 150 to 230 ms (Caputo and Casco, 1999). Although the latencies may vary depending on the stimuli, our present results are consistent with these previous reports if we assume that activity around the CaS (V1) in the early (117 ms) and late (215 ms) periods is related to the feature processing for the stimulus items and the target segregation from their surroundings, respectively. The moment-magnitude for the pFuG had one prominent peak at a latency of 125 ms. After this peak, it had a small peak at various latencies around 237 ms in 18 of 24 data, whereas the grand-average moment-magnitude had no second peak. Unlike the CaS, the first prominent and second small peaks had no significant relation to the target's presence. Neural activities of macaque ventral V4, which is considered to correspond to the human ventral lateral–occipital cortex and lingual gyrus/FuG complex (Gallant et al., 2000; Tootell and Hadjikhani, 2001), were associated with the pop-out of newly selected stimuli (Motter, 1994). Moreover, the largely suppressive V4 neuron was thought to contribute to figureground segmentation (Desimone et al., 1993). However, no difference was found for target presence effect in our FuG activities. In this respect, the N2pc component, which was distributed around the occipito-temporal cortices during late latency ranges of 220–240 ms, was reported to depend on the target presence; it was supposed to focus attention to eliminate interference from the non-target stimuli in the conjunction search (Hopf et al., 2004). However, in the present study, we used the

304

T. Hayakawa et al. / NeuroImage 33 (2006) 296–306

Fig. 8. Normalized moment-magnitudes of dipoles located around the aCgG (A) and CS (B). Black lines reflect the moment-magnitudes elicited by target-present and target-absent trials for the left and right visual field stimulation (n = 24), and the red line shows the grand-average of the normalized moment-magnitudes averaged across all data sets. The gray scale represents the contribution ratio for the 20-ms bin.

efficient singleton search, which scarcely required any focal attention to integrate features. Our results may also shed some light on another aspect of neural activity around the IPS and pSTS. Although the momentmagnitudes increased for both regions in the early latency range (90–160 ms), the contributions of the estimated dipoles to the measured magnetic field were small during this period. This suggests that the apparent moment may be caused by crosstalk from the neighboring large activations, such as those in the occipital areas. At latencies around 196 ms, the contribution of the IPS increased, suggesting that the estimation of momentmagnitudes was more reliable. However, the moment-magnitudes were larger for the right IPS than for the left IPS at late latencies. The increase in the right IPS activity occurred between the early and late activity of the occipital visual cortices (117 ms and 215 ms), and it was independent of the stimulus conditions, such as the target-present or -absent and the left or right hemi-field stimulation. These results suggest that the right IPS played the dominant role in our search task; that is, the right IPS was recruited to maintain spatial attention to the visual display, irrespective of whether target-present or -absent through the efficient search. In the previous studies using fMRI and PET, the right IPS was reported to be significantly activated during discrimination of spatial properties, such as orientation and size, whereas no significant activation was found in the left IPS (Faillenot et al., 1999). Furthermore, it was suggested that the right IPS on a dorsal network acted to control visuospatial attention and gravitating to salient stimuli (Corbetta et al., 2002; Corbetta and Shulman, 2002). Moreover, the right IPS participates in large-scale attentional shifts while the left IPS was involved in object identification during the visual conjunction search (Muller et al., 2003). The present result suggests the dominant role of the right IPS, which is consistent

with these previous results, and reveals the sequence of the neural activity of the occipital cortices and parietal cortex. On the other hand, the neural activity around the pSTS was small and sustainable over the time course, and the effect of target's presence did not affect the activity during our efficient search. It is considered that the pSTS might be hard to synchronize to exogenous stimuli. The role of the STS has been demonstrated in single neuron recordings and brain imaging, for example, the STS in monkeys was employed to extract a perceptual dominant target and familiar objects in a natural scene (Sheinberg and Logothetis, 1997, 2001) and the STS in human was activated in relation to the detection of relevant stimuli (Corbetta et al., 1991). These previous results suggested that the STS was not employed for automatic target detection. In these respects, our singleton efficient search required automatic item detection but not purposeful target segregation. Since the network for the search processing in the brain shared common regions (Hayakawa et al., 2003; Nobre et al., 2003), the pSTS might be involved in the task, yet not play a major role. This may be why the pSTS showed only small activation in the present study. Although the moment-magnitude was not large, the activation around the pSTS occurred after the early activity of the occipital visual cortices and activity of the right IPS and overlapped with the late activities of the occipital visual cortices. The event-related fMRI demonstrated that the posterior segment of the STS, which was considered to contribute to target detection, was activated following the activity around the IPS (Pollmann et al., 2003). Furthermore, it was interesting that the occipito-temporal activity (220–240 ms) was found after the parietal activity (180–200 ms), where the neural sources in these areas were considered to compose the magnetic counterpart of N2pc (Hopf et al., 2000). Since the latencies and

T. Hayakawa et al. / NeuroImage 33 (2006) 296–306

sequence of activities for N2pc were analogous to those of the IPS and pSTS in the present study, N2pc might reflect the neural activities around the IPS, pSTS and occipital cortices detected by our multi-dipole analysis. In addition, it is notable that the present analysis enabled us to separate the neural activity of the parietal and temporal cortices from the late activity of the CaS, which is related to the target segmentation. Since there are anatomical feedforward and feedback connections between the occipital visual cortices and the STS and IPS (Cavada and Goldman-Rakic, 1989; Felleman and Van Essen, 1991; Rockland and Van Hoesen, 1994), it is possible to assume that functional networks exist among these regions. The neural activity of figure-ground segmentation was also influenced by the contextual and attentional modulation (Hupe et al., 1998; Spratling and Johnson, 2004). These results suggested that V1 activity related to the segmentation was not independent of the other brain regions and their functions. The different temporal characteristics of these regions in the present experiments suggest the following information flow; after early feature processing of exogenous visual stimuli in the occipital visual cortices (117 ms for the CaS and 125 ms for the pFuG), the information might be used in the right IPS to shift spatial attention (196 ms) and in the pSTS (218 ms); then, some of the attentional and/or modulated information is fed back to the occipital visual cortices of the same hemisphere as that of the early processing (215 ms for the CaS and 237 ms for the pFuG), where it is successively used for target segregation from the surroundings. This sequence for target detection during a visual search basically matches the model assuming a single-search process such as a “guided search” (Wolfe et al., 2003). Thus, the present study provides information on the physical brain regions and temporal information for the model. Since activities of multiple brain regions overlapped in time during target detection, it should be emphasized that target segregation in the visual search was not executed in a completely cascaded way. The present study demonstrated that the temporal characteristics of neural activity would be helpful in constructing realistic visual search models for the brain. Acknowledgments We would like to thank Dr. Makoto Kato for his variable suggestions, and we are indebted to Dr. Satoru Miyauchi and Dr. Ryouzi Suzuki for supporting this study. References Bach, M., Meigen, T., 1992. Electrophysiological correlates of texture segregation in the human visual evoked potential. Vision Res. 32, 417–424. Bach, M., Meigen, T., 1997. Similar electrophysiological correlates of texture segregation induced by luminance, orientation, motion and stereo. Vision Res. 37, 1409–1414. Caputo, G., Casco, C., 1999. A visual evoked potential correlate of global figure-ground segmentation. Vision Res. 39, 1597–1610. Cavada, C., Goldman-Rakic, P.S., 1989. Posterior parietal cortex in rhesus monkey: I. Parcellation of areas based on distinctive limbic and sensory corticocortical connections. J. Comp. Neurol. 287, 393–421. Celebrini, S., Thorpe, S., Trotter, Y., Imbert, M., 1993. Dynamics of orientation coding in area V1 of the awake primate. Vis. Neurosci. 10, 811–825. Corbetta, M., Shulman, G.L., 2002. Control of goal-directed and stimulusdriven attention in the brain. Nat. Rev., Neurosci. 3, 201–215.

305

Corbetta, M., Miezin, F.M., Dobmeyer, S., Shulman, G.L., Petersen, S.E., 1991. Selective and divided attention during visual discriminations of shape, color, and speed: functional anatomy by positron emission tomography. J. Neurosci. 11, 2383–2402. Corbetta, M., Shulman, G.L., Miezin, F.M., Petersen, S.E., 1995. Superior parietal cortex activation during spatial attention shifts and visual feature conjunction. Science 270, 802–805. Corbetta, M., Kincade, J.M., Shulman, G.L., 2002. Neural systems for visual orienting and their relationships to spatial working memory. J. Cogn. Neurosci. 14, 508–523. Das, A., 1996. Orientation in visual cortex: a simple mechanism emerges. Neuron 16, 477–480. Duncan, J., Humphreys, G.W., 1989. Visual search and stimulus similarity. Psychol. Rev. 96, 433–458. Desimone, R., Moran, J., Schein, S.J., Mishkin, M., 1993. A role for the corpus callosum in visual area V4 of the macaque. Vis. Neurosci. 10, 159–171. Donner, T.H., Kettermann, A., Diesch, E., Ostendorf, F., Villringer, A., Brandt, S.A., 2002. Visual feature and conjunction searches of equal difficulty engage only partially overlapping frontoparietal networks. NeuroImage 15, 16–25. Donner, T.H., Kettermann, A., Diesch, E., Villringer, A., Brandt, S.A., 2003. Parietal activation during visual search in the absence of multiple distractors. NeuroReport 14, 2257–2261. Faillenot, I., Decety, J., Jeannerod, M., 1999. Human brain activity related to the perception of spatial features of objects. NeuroImage 10, 114–124. Felleman, D.J., Van Essen, D.C., 1991. Distributed hierarchical processing in the primate cerebral cortex. Cereb. Cortex 1, 1–47. Ferber, S., Humphrey, G.K., Vilis, T., 2005. Segregation and persistence of form in the lateral occipital complex. Neuropsychologia 43, 41–51. Fujimaki, N., Hayakawa, T., Nielsen, M., Knosche, T.R., Miyauchi, S., 2002. An fMRI-constrained MEG source analysis with procedures for dividing and grouping activation. NeuroImage 17, 324–343. Gallant, J.L., Shoup, R.E., Mazer, J.A., 2000. A human extrastriate area functionally homologous to macaque V4. Neuron 27, 227–235. Grossberg, S., Mingolla, E., Ross, W.D., 1994. A neural theory of attentive visual search: interactions of boundary, surface, spatial, and object representations. Psychol. Rev. 101, 470–489. Hayakawa, T., Miyauchi, S., Fujimaki, N., Kato, M., Yagi, A., 2003. Information flow related to visual search assessed using magnetoencephalography. Brain Res. Cogn. Brain Res. 15, 285–295. Hopf, J.M., Luck, S.J., Girelli, M., Hagner, T., Mangun, G.R., Scheich, H., Heinze, H.J., 2000. Neural sources of focused attention in visual search. Cereb. Cortex 10, 1233–1241. Hopf, J.M., Noesselt, T., Tempelmann, C., Braun, J., Schoenfeld, M.A., Heinze, H.J., 2004. Popout modulates focal attention in the primary visual cortex. NeuroImage 22, 574–582. Hupe, J.M., James, A.C., Payne, B.R., Lomber, S.G., Girard, P., Bullier, J., 1998. Cortical feedback improves discrimination between figure and background by V1, V2 and V3 neurons. Nature 394, 784–787. Kastner, S., Nothdurft, H.C., Pigarev, I.N., 1997. Neuronal correlates of pop-out in cat striate cortex. Vision Res. 37, 371–376. Kastner, S., De Weerd, P., Ungerleider, L.G., 2000. Texture segregation in the human visual cortex: a functional MRI study. J. Neurophysiol. 83, 2453–2457. Knierim, J.J., van Essen, D.C., 1992. Neuronal responses to static texture patterns in area V1 of the alert macaque monkey. J. Neurophysiol. 67, 961–980. Lamme, V.A., Van Dijk, B.W., Spekreijse, H., 1992. Texture segregation is processed by primary visual cortex in man and monkey. Evidence from VEP experiments. Vision Res. 32, 797–807. Lamme, V.A., Rodriguez-Rodriguez, V., Spekreijse, H., 1999. Separate processing dynamics for texture elements, boundaries and surfaces in primary visual cortex of the macaque monkey. Cereb. Cortex 9, 406–413.

306

T. Hayakawa et al. / NeuroImage 33 (2006) 296–306

Lamme, V.A., Supèr, H., Landman, R., Roelfsema, P.R., Spekreijse, H., 2000. The role of primary visual cortex (V1) in visual awareness. Vision Res. 40, 1507–1521. Leonards, U., Sunaert, S., Van Hecke, P., Orban, G.A., 2000. Attention mechanisms in visual search—An fMRI study. J. Cogn. Neurosci. 12, 61–75. Luck, S.J., Hillyard, S.A., 1990. Electrophysiological evidence for parallel and serial processing during visual search. Percept. Psychophys. 48, 603–617. Luck, S.J., Hillyard, S.A., 1994. Spatial filtering during visual search: evidence from human electrophysiology. J. Exp. Psychol. Hum. Percept. Perform. 20, 1000–1014. Motter, B.C., 1994. Neural correlates of feature selective memory and pop-out in extrastriate area V4. J. Neurosci. 14, 2190–2199. Muller, N.G., Donner, T.H., Bartelt, O.A., Brandt, S.A., Villringer, A., Kleinschmidt, A., 2003. The functional neuroanatomy of visual conjunction search: a parametric fMRI study. NeuroImage 20, 1578–1590. Nakayama, K., Silverman, G.H., 1986. Serial and parallel processing of visual feature conjunctions. Nature 320, 264–265. Nobre, A.C., Sebestyen, G.N., Gitelman, D.R., Frith, C.D., Mesulam, M.M., 2002. Filtering of distractors during visual search studied by positron emission tomography. NeuroImage 16, 968–976. Nobre, A.C., Coull, J.T., Walsh, V., Frith, C.D., 2003. Brain activations during visual search: contributions of search efficiency versus feature binding. NeuroImage 18, 91–103. Nothdurft, H.C., 1991. Texture segmentation and pop-out from orientation contrast. Vision Res. 31, 1073–1078. Nothdurft, H., 2000. Salience from feature contrast: additivity across dimensions. Vision Res. 40, 1183–1201. Pollmann, S., Weidner, R., Humphreys, G.W., Olivers, C.N., Muller, K., Lohmann, G., Wiggins, C.J., Watson, D.G., 2003. Separating distractor rejection and target detection in posterior parietal cortex— An event-related fMRI study of visual marking. NeuroImage 18, 310–323. Rockland, K.S., Van Hoesen, G.W., 1994. Direct temporal–occipital feedback connections to striate cortex (V1) in the macaque monkey. Cereb. Cortex 4, 300–313. Roelfsema, P.R., Lamme, V.A., Spekreijse, H., 2004. Synchrony and covariation of firing rates in the primary visual cortex during contour grouping. Nat. Neurosci. 7, 982–991.

Schubö, A., Schroger, E., Meinecke, C., 2004. Texture segmentation and visual search for pop-out targets. An ERP study. Brain Res. Cogn. Brain Res. 21, 317–334. Sheinberg, D.L., Logothetis, N.K., 1997. The role of temporal cortical areas in perceptual organization. Proc. Natl. Acad. Sci. U.S.A. 94, 3408–3413. Sheinberg, D.L., Logothetis, N.K., 2001. Noticing familiar objects in real world scenes: the role of temporal cortical neurons in natural vision. J. Neurosci. 21, 1340–1350. Spratling, M.W., Johnson, M.H., 2004. A feedback model of visual attention. J. Cogn. Neurosci. 16, 219–237. Super, H., Spekreijse, H., Lamme, V.A., 2001. Two distinct modes of sensory processing observed in monkey primary visual cortex (V1). Nat. Neurosci. 4, 304–310. Schira, M.M., Fahle, M., Donner, T.H., Kraft, A., Brandt, S.A., 2004. Differential contribution of early visual areas to the perceptual process of contour processing. J. Neurophysiol. 91, 1716–1721. Skiera, G., Petersen, D., Skalej, M., Fahle, M., 2000. Correlates of figureground segregation in fMRI. Vision Res. 40, 2047–2056. Tootell, R.B., Hadjikhani, N., 2001. Where is ‘dorsal V4’ in human visual cortex? Retinotopic, topographic and functional evidence. Cereb. Cortex 11, 298–311. Treisman, A.M., Gelade, G., 1980. A feature-integration theory of attention. Cognit. Psychol. 12, 97–136. Treisman, A., Gormican, S., 1988. Feature analysis in early vision: evidence from search asymmetries. Psychol. Rev. 95, 15–48. Treisman, A., Souther, J., 1985. Search asymmetry: a diagnostic for preattentive processing of separable features. J. Exp. Psychol. Gen. 114, 285–310. Wolfe, J.M., Cave, K.R., Franzel, S.L., 1989. Guided search: an alternative to the feature integration model for visual search. J. Exp. Psychol. Hum Percept. Perform. 15, 419–433. Wolfe, J.M., Butcher, S.J., Lee, C., Hyle, M., 2003. Changing your mind: on the contributions of top–down and bottom–up guidance in visual search for feature singletons. J. Exp. Psychol. Hum. Percept. Perform. 29, 483–502. Zipser, K., Lamme, V.A., Schiller, P.H., 1996. Contextual modulation in primary visual cortex. J. Neurosci. 16, 7376–7389. Woodman, G.F., Luck, S.J., 2003. Serial deployment of attention during visual search. J. Exp. Psychol. Hum. Percept. Perform. 29, 121–138.