An fMRI study on memory discriminability for complex visual scenes

Abstract

The fan effect represents an increase in reaction time for the recognition of an item as a function of the amount of information associated with that item in memory. The present study used fMRI to study the neural correlates of the fan effect for complex visual scenes. We used a test in which landscape pictures were divided vertically into three equal segments. In the high discriminability condition only one segment was presented during encoding, whereas in the low discriminability condition two different segments from the same picture were presented. During a subsequent forced‐choice recognition test, reaction times were significantly faster for the high discriminability condition. Increase in brain activity for the low relative to high discriminability condition was observed in the right prefrontal cortex, several regions of parietal cortex bilaterally, and several late visual processing areas, including the occipito‐temporal regions, precuneus, and cuneus. These results support the hypothesis that a region of the prefrontal cortex is involved in the control of memory interference at retrieval elicited by the amount of related information in memory, and further suggests that this involvement is right‐lateralized for nonverbal material. The high versus low discriminability contrast showed an increase in activity principally in the bilateral medial temporal gyrus, including the enthorinal cortex/hippocampus and in several bilateral prefrontal cortex regions mostly located in BA 10. These activations were associated with a condition, in which the stimuli were more salient in memory and thus could represent the perceptual salience of items in memory. Hum Brain Mapp, 2008. © 2007 Wiley‐Liss, Inc.

INTRODUCTION

The amount of information contextually associated with an item has an impact on the discriminability of that item in memory [Anderson,1974]. This so‐called fan effect takes the form of an increased response time and/or error rate on a memory test for the retrieval of items as a function of the number of studied related items [Radvansky,2005]. In other words, the time to retrieve information about an item increases as more related information is studied about that item. There is evidence to suggest that the interference process responsible for the fan effect is specific to the memory retrieval stage [Anderson and Reder,1999]. Hence, the longer reaction time observed with an increase of the fan suggests that the item is less discriminable or salient, during retrieval tests because of the interference induced by the related information stored in memory.

Few studies to date have examined the neural correlates of the decrease in discriminability as a function of the amount of associated information stored in memory. For instance, Sohn et al. [2003] presented participants during an fMRI study with person–location pairs such that each person was associated with one, two, or three different locations and vice versa. Their region of interest analysis in a subsequent memory test showed that increasing the fan produced greater activity in a region mostly located in the inferior dorsolateral prefrontal cortex bilaterally, but no significant effect in the posterior parietal regions. Since the prefrontal region activity not only responds to the presence of competing information in memory but increases as competition increases, they concluded that this region is crucial for the retrieval of information in the presence of distracting information. However, since their parietal activation was independent of the extent of competition, they suggested that this region is responsible for encoding the stimulus and updating its representation.

In a later study, Sohn et al. [2005] asked participants to memorize sentences in the form of subject–verb–object (e.g., Programmer punches chemist) and subsequently gave them a recall and recognition test. Each subject and verb pair was associated with one, two, or three different objects. Their region of interest analysis showed that the prefrontal region activation increased as a function of the fan for both the recognition and recall tests, whereas the parietal region activation increased as a function of the fan only in the recall test. In a second experiment, they tested if this dissociation between prefrontal and parietal regions depended on the modality of stimulus presentation. The methodology of the second experiment was almost the same as the first, except that in the recognition test the probes from half the blocks were presented visually, whereas the other half was presented aurally (in the first experiment they were only presented visually). Increase in prefrontal activity as a function of the fan was observed for both modalities, suggesting that the involvement of the prefrontal in the retrieval of information in the presence of distracting information is independent of perceptual modality.

These studies, however, did not systematically examine the opposite pattern of brain activation, which is increased activity for the low fan condition. Such a contrast would be particularly interesting to better understand neural correlates of salient, or highly discriminable, memory traces. Sohn et al. [2003] briefly reported increased activity in the putamen and in a region of the ventrolateral prefrontal cortex for the low‐fan compared with the high‐fan condition in an exploratory analysis. In their recent study [Sohn et al.,2005] the left angular gyrus was the only region showing greater activity in the low versus high fan condition.

In a recent fMRI study on memory interference we explored the neural correlates of both interference caused by related information in memory and highly discriminable memory traces [Lepage et al.,2005]. To that end we presented pictures of landscapes that were divided into three equal segments. During encoding, a segment of each picture was presented. In another block following encoding, a second different segment was presented for half the pictures. During a forced‐choice recognition test, participants had to choose between a segment presented during encoding and a new segment that had never been presented. We observed that the recognition of picture segments, for which a second segment had been presented, was associated with an increase in activity in the left precuneus, left angular gyrus, right cingulate sulcus, and right postcentral gyrus. The opposite contrast (only one picture segment presented during encoding) yielded activation, among others, in the anterior and medial prefrontal regions bilaterally and the right parahippocampal gyrus. Unfortunately, no difference in the mean response time was observed between the two conditions. This absence can be attributed to the high rate of omission observed in the study. Since the mean response time was 1,839 ms for the interference trials and 1,801 ms for the control trials and the recognition stimulus disappeared after 2,500 ms, it is possible that subjects were inadvertently biased not to respond beyond that point. This preliminary study was interesting as it allowed us to examine the neural correlates of salience in memory. However, it was limited by the absence of a fan effect in the behavioral results, which rendered it impossible to establish a link between these activations and the fan effect for complex images, and between these activations and better salience in memory.

The aim of the present study was threefold. Its first goal was to extend the results of previous studies on the fan effect by showing that a fan effect can be observed with complex visual scenes. Its second goal was to determine whether the neural correlates of this fan effect would show a similar pattern to those observed in studies using sentences. Finally, it also aimed to observe which brain regions had greater activation in response to items that were more salient in memory. To attain these goals, we adapted a memory test originally used by Chandler [1991] to study memory interference that was similar to the one used by Lepage et al. [2005]. The test was comprised of complex landscape pictures divided into three vertical segments. At encoding, one segment of each picture was presented for half the pictures (high discriminability condition) and two different segments were presented for the other half (low discriminability condition). This encoding task was followed by a forced‐choice recognition task, in which one of the previously seen picture segments was presented with a never presented (third) picture segment. The participants were asked to decide which of the two segments was presented earlier. Since the information contained in two segments of the same picture are very similar, because they are intrinsically related, presenting a second different segment coming from the same image during encoding should be sufficient to reduce the discriminability of the first segment.

At the behavioral level, it was expected that participants would show a fan effect by responding slowly and less accurately in the recognition test for segments of the low discriminability condition (from a picture for which two segments were presented during encoding). At the brain level, we expected the bilateral prefrontal cortex to show an increase in activity for the low discriminability condition. This would support the hypothesis that this region is involved in the control of memory interference at retrieval elicited by the quantity of related information in memory, and further suggest that this interference is independent of the type of material used (e.g., sentences, images). The study was also designed to study neural correlates of priming during the presentation of picture segments at encoding, but the current article will only focus on the recognition results. The encoding findings have been published elsewhere [Blondin and Lepage,2005].

MATERIALS AND METHODS

Participants

Twenty four participants (15 females, mean age 23.5, range 18–31 years) participated in the fMRI study. All participants were right‐handed according to the Edinburgh Inventory [Oldfield,1971] and none had a current or previous history of neurological or psychiatric disorders. The study was approved by the Research Ethics Board of the Montreal Neurological Institute (MNI) and written informed consent was obtained from all participants.

Materials

The stimulus material consisted of 72 color pictures. Approximately half the pictures portrayed nature scenes (e.g., forest, river), whereas the remaining half, termed human scenes, included humans or human‐made objects (e.g., building, boat). Each of the pictures was divided vertically into three segments of equal size for a total of 216 segments. For an illustration of the stimuli used and the procedure of the present experiment see Blondin and Lepage [2005].

Design and Procedure

The experiment consisted of two successive tasks: (i) a memory encoding task, and (ii) a forced‐choice recognition memory test. Both encoding and recognition were scanned. The encoding task was comprised of four types of stimuli: (i) picture segments followed later by related (i.e., taken from the same picture) segments, (ii) picture segments related to previously presented segments (i.e., coming from the same picture as one of the segments already presented), both Types i and ii being part of the low discriminability condition, (iii) picture segments not followed by any related segments (high discriminability condition), and (iv) a low level baseline image. There was a lag of 7–31 stimuli (mean 12.2 on a total of 144, including the baseline images) between the presentation of a first picture segment and the presentation of the second related one. The low level baseline image (Type iv above) was used to introduce jitter in the timing of the hemodynamic response and allow for the extraction of the response for the other conditions [Buckner et al.,1998; Burock et al.,1998]. Participants were instructed to not respond to this image. There were 36 occurrences of each of the four types of stimuli and they were presented in a pseudo‐random order. Picture segments were presented in the middle of the screen for 4,500 ms at a rate of 1 segment/6 s. Participants were instructed to judge whether the picture was a nature or human scene. Participants were instructed to press one button when the picture segment included humans or human‐made objects and another button when the segment included only natural elements. This task was chosen to provide a deep (semantic) processing of each segment and therefore to enhance the memorization of the segments. As mentioned earlier, the results from the encoding task were presented and discussed as part of a separate communication.

Forced‐choice recognition task immediately followed the encoding. This task consisted of the successive presentation of each and every picture segment presented during encoding that was followed by a related picture segment (Type i above) or that was not followed by a related segment (Type iii above) concurrently with a related segment that had never been presented before in the experiment. None of the picture segments presented during encoding that were related to a previously presented segment (Type ii above) was presented during the recognition test. Participants had to indicate which of the segments had been presented during encoding. Recognition trials were presented at a rate of one stimulus every 4 s, with each pair of picture segments appearing for 2,500 ms. As in the encoding task, an abstract image consisting of a visually distorted stimulus and arranged in the same side by side layout as the other picture segments served as a baseline event. Again, participants were instructed to not respond to this stimulus. Thirty‐six occurrences of this baseline event were interspersed in the recognition run. All stimuli in the recognition run were presented in a random order that differed for each participant.

fMRI Data Acquisition and Analyses

Scanning was carried out at the MNI on a 1.5T Siemens Sonata system, using gradient echo EPI sequences. Stimuli were generated by a Pentium class PC Laptop computer running E‐Prime (http://www.pstnet.com) and projected via a LCD projector and mirror system. A four key response mouse connected to the computer recorded participants' responses. The scanner triggered the start of stimulus presentation. A scanning session began by two functional runs (encoding and recognition) comprising 370 and 200 T ₂*‐weighted images, respectively, acquired with blood oxygenation level‐dependent contrast (TR = 2,500 ms, TE = 50 ms, 25 slices, 2 mm × 2 mm × 5 mm voxels) and was followed by a high‐resolution T ₁‐weighted acquisition for anatomical localization. Functional scans were acquired parallel to the anterior–posterior commissural plane.

fMRI images were analyzed with fmristat [Worsley et al.,2002]. The T ₂* images were first realigned to the 336th image (14th min out of 15 min 30 s) for the encoding run and to the 168th image (7th min out of 8 min 25 s) for the recognition run and then spatially smoothed with a 6 mm (fwhm) isotropic Gaussian kernel. The images of each functional run were realigned to images close to the end of the run to be closer to the anatomical run that was taken last. Statistical analysis of the fMRI data was based on a linear model with correlated errors. The design matrix of the linear model was convolved with a hemodynamic response function modeled as a difference of two gamma functions timed to coincide with the acquisition of each slice. Low frequency drifts were removed by including polynomial covariates, up to Degree 3, in the design matrix. The correlation structure was modeled as an autoregressive process of Degree 1. At each voxel, the autocorrelation parameter was estimated from the least squares residuals, using the Yule–Walker equations. The autocorrelation parameter was first regularized by spatial smoothing with a 10‐mm fwhm Gaussian filter, and then used to whiten the data and the design matrix. The linear model was then re‐estimated using least squares on the whitened data to produce estimates of effects and their standard errors. For each participant, the effects and standard deviations from the encoding and recognition runs were normalized to standard space using the MNI template [Cocosco et al.,1997] as a reference.

In a second step, data from all participants were combined into a mixed effect linear model. The mixed effect analysis is in principle a random effect analysis. Like a typical random effect analysis, it takes into account the subject‐by‐subject variance as an extra source of variability (in addition to the scan‐to‐scan residual errors, which formed the fixed effects variance). However, unlike a typical random effect analysis, the mixed effect analysis takes advantage of the fact that both sources of variability are commonly informed by the same anatomical structure. Put simply, once the imaging data are smoothed, the regions in the brain susceptible to induce a large scan‐to‐scan variance are prone to induce a large variance of the residual errors resulting from the sum of the random and fixed effects residual errors [see Fig. 6 of Worsley et al.,2002]. As a consequence, the extra variance taken into account by the random effect analysis can be estimated by the ratio of the random effect variance to the fixed effects variance, as a function of smoothing. It is important to note that this variation of random effect variance throughout the brain is orthogonal to the variation of activations expected by the experimental manipulation of interest. The estimation of the random effect residual errors has the advantage to reduce the residual noise across subjects, which results in an increased sensitivity as reflected by a raise of the “effective” degree of freedoms (DF). This increased sensitivity was recently illustrated by a study that directly compared this mixed‐effects analysis with an analysis based on SPM99. The study found bilateral temporal activations for processing sentences, whereas SPM99 had only found left‐lateralized activations [Taylor and Worsley,2006].

The estimation of the random effect errors are performed by estimating the ratio of the fixed‐ and random‐effects residual errors to the fixed residual errors. This estimation was fitted using restricted maximum likelihood estimates implemented by the maximization (EM) algorithm [Worsley et al.,2002]. The regularizing of this ratio by spatial smoothing was set to obtain an effective DF of 100 DF, which resulted in an ∼8‐mm fwhm Gaussian filter. This number is used as the default number in a compromise between maintaining a high degree of effective DF while still capturing most of the random effect. Statistical maps were thresholded at P < 0.001, uncorrected for multiple comparisons [t(100) = 3.39]. Only regions consisting of at least 10 contiguous voxels above threshold are reported.

RESULTS

Behavioral Results

Both behavioral and fMRI data from two participants were discarded because of a technical problem for the first one and chance level performance in the memory test for the second one. The mean response time at recognition was 1,538 ms (SD 199) for the high discriminability condition and 1,593 ms (SD 194) for the low discriminability condition. These mean response times include only the trials, in which the segments were remembered (hits). This difference in response time reached the significance level [t(21) = 2.71, P = 0.01]. The mean hit rate was 74% for the high discriminability condition and 75% for the low discriminability condition. The difference in mean hit rate did not reach significance [t(21) = 0.204, P = 0.84].

fMRI Results

The fMRI analysis included only the trials, in which the segments were remembered (hits). The first contrast examined the activity elicited by the recognition of the low discriminability trials compared with the high discriminability trials. An increase in activity for low discriminability trials was observed in the left postcentral gyrus (BA 1, 2, 3), left insula (BA 13), right inferior frontal gyrus (BA 45, 46), right cingulate gyrus (BA 24), right superior parietal lobule (BA 19, 37), inferior parietal lobule bilaterally (BA 40), left precuneus (BA 7, 19), cuneus bilaterally (BA 17, 18, 19), left inferior temporal gyrus (BA 19), left medial temporal gyrus (BA 19, 37), lingual gyrus bilaterally (BA 18, 19), right inferior occipital gyrus (BA 18), left middle occipital gyrus (BA 19), right fusiform gyrus (BA 18), cerebellum bilaterally, and the paracentral lobule (BA 5). Figure 1 illustrates the activations overlaid onto an average MRI template. Table I depicts the main foci of activations and their MNI coordinates.

Figure 1.

Event‐related imaging results (P < 0.001, uncorrected) for the contrast showing increased activity for the low discriminability condition relative to with the high discriminability condition, superimposed upon horizontal MR slices from the average brain computed from the structural images of participants. The number on the upper left corner of each slice denotes the distance in millimeter from the anterior commissural point on the z‐axis of the Talairach coordinate system. The three‐dimensional rendering of an MRI brain on the upper left corner of the picture indicates the localization on the z‐axis for each slices presented below. Significant activity can be observed in left postcentral gyrus, left insula, right inferior frontal gyrus, right cingulate gyrus, right superior and inferior parietal lobules, left precuneus, cuneus bilaterally, left inferior temporal gyrus, left medial temporal gyrus, lingual gyrus bilaterally, right inferior occipital gyrus, left middle occipital gyrus, right fusiform gyrus, cerebellum bilaterally, and the paracentral lobule. [Color figure can be viewed in the online issue, which is available at www.interscience.wiley.com.]

Table I.

Brain regions showing significant activation (P < 0.001, uncorrected) for the contrast between low discriminability versus high discriminability correct recognition (only the regions consisting of at least 10 contiguous voxels are reported)

		MNI coordinates				Cluster size
Region	BA	x	y	z	t‐Value
Left hemisphere
Insula	13	−40	−8	24	3.62	10
Postcentral gyrus	1/2	−56	−24	48	5.1	214
Inerior parietal lobule	40	−36	−30	40	3.53	10
Postcentral gyrus	2/3	−38	−32	66	4.12	50
Postcentral gyrus	3	−30	−34	52	4.4	37
Precuneus	7	−16	−54	54	3.66	18
Middle temporal gyrus	19/37	−44	62	−4	5.43	668
Precuneus	7	−22	−62	52	4	18
Inferior temporal gyrus	19	−44	−66	−18	4.13	21
Cerebellum		−22	−66	−26	4.02	10
Precuneus	7	−12	−70	62	3.6	13
Lingual gyrus	18	−4	−82	−14	3.95	12
Precuneus	19	−12	−86	42	3.76	32
Cuneus	17	−22	−92	0	3.7	14
Middle occipital gyrus	19	−22	−96	16	5.12	166
Right hemisphere
Inferior frontal gyrus	45/46	38	32	14	3.99	44
Cingulate gyrus	24	2	−12	40	4.41	19
Superior parietal lobule	19/37	34	−42	38	4.45	89
Inferior parietal lobule	40	44	−50	52	3.92	14
Cerebellum		32	−50	−32	4.26	17
Fusiform gyrus	19	34	−64	−18	4.86	225
Cerebellum		18	−72	−18	3.51	11
Cerebellum		38	−72	−26	3.93	15
Lingual gyrus	19	18	−74	−6	3.76	31
Inferior occipital gyrus	18	30	−86	−4	4.47	432
Fusiform gyrus	18	32	−86	−18	3.76	20
Cuneus	19	6	−86	42	3.79	17
Cuneus	18	12	−98	6	4.71	91
Midline
Cerebellum		0	−46	−26	4.14	33
Cerebellum		0	−68	−26	5.27	27
Lingual gyrus	18	0	−68	−6	3.72	16
Paracentral lobule	5	2	−26	52	3.78	31

BA, Brodmann area.

xyz‐Coordinates of local maxima are listed in MNI coordinates according to the Talairach system [Talairach and Tournoux,1988].

The second contrast compared the activity elicited by the recognition of high discriminability trials versus low discriminability trials. An increase in activity for high discriminability trials was observed in the left superior frontal gyrus (BA 10), right medial frontal gyrus (BA 10), right inferior frontal gyrus (BA 45), medial temporal gyri bilaterally (BA 38, 21), left uncus (BA 20), left enthorinal cortex/hippocampus (BA 28), left inferior temporal gyrus (BA 20), left retrosplenial cortex (BA 30), and right cerebellum. Figure 2 illustrates this pattern of activity on horizontal slices of an MRI template. Table II depicts the main foci of activations and their MNI coordinates.

Figure 2.

Event‐related imaging results (P < 0.001, uncorrected) for the contrast showing increased activity for the high discriminability condition relative to with the low discriminability condition, superimposed upon horizontal MR slices from the average brain computed from the structural images of participants. The number on the upper left corner of each slice denotes the distance in mm from the anterior commissural point on the z‐axis of the Talairach coordinate system. The three‐dimensional rendering of an MRI brain on the upper left corner of the picture indicates the localization on the z‐axis for each slices presented below. Significant activity can be observed in the left superior frontal gyrus, right medial frontal gyrus, right inferior frontal gyrus, medial temporal gyrus bilaterally, left uncus, left enthorinal cortex/hippocampus, left inferior temporal gyrus, left retrosplenial cortex, and right cerebellum. [Color figure can be viewed in the online issue, which is available at www.interscience.wiley.com.]

Table II.

Brain regions showing significant activation (P < 0.001, uncorrected) for the contrast between high discriminability versus low discriminability correct recognition (only the regions consisting of at least10 contiguous voxels are reported)

		MNI coordinates				Cluster size
Region	BA	x	y	z	t‐Value
Left hemisphere
Superior frontal gyrus	10	−6	60	24	4.35	61
Superior frontal gyrus	10	−18	60	16	4.46	24
Middle temporal gyrus	38	−38	6	−36	4.95	112
Uncus	20	−22	−6	−38	4.22	21
Entorhinal cortex/hippocampus	28	−20	−12	−22	5.05	93
Inferior temporal gyrus	20	−36	−12	−36	5.07	29
Retrosplenial cortex	30	−12	−60	16	3.6	17
Right hemisphere
Medial frontal gyrus	10	8	46	2	3.64	25
Medial frontal gyrus	10	4	42	−10	4.5	141
Inferior frontal gyrus	45	60	24	8	4.56	37
Middle temporal gyrus	21	36	8	−40	5.21	47
Cerebellum		20	−64	−32	5.11	20

BA, Brodmann area.

xyz‐Coordinates of local maxima are listed in MNI coordinates according to the Talairach system [Talairach and Tournoux,1988].

DISCUSSION

The present study examined the neural correlates of the fan effect associated with the recognition of segments of complex visual scenes as a function of the number of different segments presented during the encoding phase. As expected, the response time was significantly lower in the high relative to the low discriminability condition. This confirms the presence of a fan effect and suggests that increasing the amount of related information at study induced interference at retrieval, which increased the time necessary to recognize the picture segment. However, no significant difference was observed in the mean hit rate between the low and high discriminability conditions, suggesting that, in this case, the presentation of more related information at study did not impair the ability to recognize a previously presented picture segment alongside a never presented related segment.

Brain Activity for the Low Discriminability Condition Relative to High Discriminability

The increase in activity observed in the prefrontal cortex was limited to the right hemisphere. We initially expected it to be bilateral, as Sohn et al. [2003,2005] reported an increase in activity in prefrontal cortex bilaterally for the high fan condition compared with the low fan condition. They suggested the involvement of a region of the prefrontal cortex in the retrieval of information in the presence of distracting information that is independent of perceptual modality. However, we did find an increase in activity in the right inferior frontal gyrus in the low discriminability condition. Our peak activation (x = 38, y = 32, z = 14) was relatively close to the central voxel of Sohn et al.'s [2003,2005] prefrontal cortex region of interest (x = 44, y = 21, z = 21 for the right region of interest). The present study suggests that there may be a material specific laterality in the prefrontal cortex region involved in the retrieval of information in the presence of distracting information. Such laterality in this region would not be surprising since laterality differences in prefrontal cortex have been a common finding in retrieval studies (left for semantic retrieval and right for episodic retrieval) [Cabeza and Nyberg,2000; Habib et al.,2003].

Several regions of the parietal cortex bilaterally showed an increase in activity in the low discriminabity condition although more foci were observed in the right hemisphere. Sohn et al. [2003,2005] did not find an increase in activity in the parietal cortex as a function of the fan for visually presented sentences and they hypothesized that its role was to encode the stimulus and update its representation. Following that hypothesis, the increase in activity seen in the parietal cortex in the low discriminability condition in the present study has two causes: the encoding of the third picture segment (the foil) and the updating of the representation of the complete picture to include the two segments seen during encoding as well as the foil seen during recognition. In the high discriminability condition this updated representation would only include two picture segments (the one seen during encoding and the foil seen during recognition).

Parietal cortex activations are a common finding in neuroimaging studies on memory retrieval [Wagner et al.,2005]. They have been observed in studies investigating retrieval success [Rugg and Henson,2002], in a study directly comparing retrieval and encoding [McDermott et al.,1999] and even in a study comparing the retrieval of words that elicited a judgement of recollection with words that elicited familiarity [Henson et al.,1999]. In a PET study comparing recall to recognition of word pairs, Cabeza et al. [1997] showed that the inferior parietal regions, particularly in the right hemisphere, were more involved in recognition than in recall. They hypothesized that the parietal cortex may be involved in the perceptual component of recognition. Lateral parietal activations have also been related to the processing of spatial location during retrieval [Moscovitch et al.,1995]. We suggest that the increase in parietal activity observed in the low relative to the high discriminability condition originates in the increase in visual information subjects retrieve in the low discriminability condition. In other words, in the low discriminability conditions, subjects would be retrieving the visual images of both related segments seen in the encoding part upon seeing the forced‐choice recognition cue, whereas in the high discriminability condition they would only be retrieving the visual image of one segment. The absence of differential parietal activation as a function of the fan in the Sohn et al. [2003,2005] studies could be explained by the subjects' use of alternate retrieval strategies. It is possible that their subjects used a more semantic than visual strategy to do the retrieval task, so there would be no differences in the number of items visually retrieved across the fan conditions.

An increase in activity for the low compared with the high discriminability condition was observed in several late visual processing areas, including occipito‐temporal regions. In cognitive tasks the involvement of these regions is usually associated with perceiving and manipulating visuospatial information and the deactivation of these regions is associated with perceptual priming [Cabeza and Nyberg,2000]. A decrease in activity in occipito‐temporal regions, including the fusiform gyrus, is a common finding amongst studies on perceptual priming of objects and visual scenes [e.g., Jessen et al.,2002] and it has even been observed for objects presented from different viewpoints or different exemplars of a same object [e.g., Vuilleumier et al.,2002]. Interestingly, those regions showed a decrease in activity in the encoding stage when the brain activity elicited by the presentation of the second related segment was compared with the brain activity elicited by the first segment [Blondin and Lepage,2005], thus suggesting that a segment of a picture could prime the processing of another one. The present results suggest that the interference generated by presenting a second related segment during encoding resulted in a negative priming effect having as neural correlates an increase in activity during retrieval in the late visual processing areas that show a reduction in activity in perceptual priming. A potential explanation for the decrease in activity seen in priming studies is that the stimulus is processed more efficiently when it is encountered for a second time, compared with the first time, thereby producing a decreased response in brain regions associated with that processing [Henson,2003]. It is therefore not surprising to observe the opposite effect: an increased response in the brain regions associated with processing in the presence of interference.

The precuneus and cuneus also showed an increase in activity in the low discriminability condition. The precuneus was first though to play a role in mental imagery occurring in the context of retrieval [Fletcher et al.,1995], but other studies have found precuneus activation in retrieval tasks where imagery was not a factor [e.g., Krause et al.,1999]. Although its role in memory retrieval is still unclear, Cabeza et al. [2001] showed that its activity tends to be greater when retrieval is successful. Cabeza et al. [2003] speculated that medial parieto‐occipital regions would be involved in orienting attention to internally generated stimuli. This speculation could explain the current result since during the forced‐choice recognition task, the second related segment in the low discriminability condition could be generated internally in addition to the target. Thus, more orienting of attention would be necessary to attain the same hit rate as in the high discriminability condition.

Brain Activity for the High Discriminability Condition Relative to Low Discriminability

Another interesting aspect of the present study was to identify brain regions showing greater activity as a function of the salience of items in memory. Several regions of the medial temporal gyrus bilaterally, including the left enthorinal cortex/hippocampus, exhibited an increase in activity in the high discriminability condition. Using a remember/know paradigm, Eldridge et al. [2000] presented participants with a memory retrieval task in which, in addition to stating if they recognized the words, they had to make a remember response if they could recollect the moment the word was studied or a know response if the word seemed familiar. They reported an increase in activity in the hippocampus only for items that were consciously recollected (remember judgement) and not for items that were recognized based on familiarity (know judgement). This activation was not due to the time spent on the task or to the mental effort required, since the decisions leading to a correct “remember” response were faster than those leading to a correct “know” response. These results suggest that the hippocampus supports the retrieval of episodic memories but not memory judgements based on familiarity. In line with these results we could hypothesize that in the present study memory judgements for items in the high discriminability condition depend on recollection, whereas memory judgements in the low discriminability condition depend more on familiarity. Thus, being able to better discriminate information in memory could be a basis for the use of conscious recollection over the use of familiarity in a memory test.

The high discriminability condition was also associated with an increase in activity for several bilateral prefrontal regions, principally located in BA 10. In two experiments [Burgess et al.,2001; King et al.,2005] Burgess and his colleagues studied the neural correlates of memory retrieval for the spatial context of events, using a virtual reality environment. In these studies, participants controlled a character in a first‐person perspective virtual reality town. In the first study [Burgess et al.,2001] participants received 16 objects from two people in two different places. In the second study [King et al.,2005] participants received 20 objects from a different person in a different location. Direct comparison of the brain activity at retrieval for the two studies showed a significantly greater involvement of the anterior BA 10 in the earlier study. The authors concluded that BA 10 is the prefrontal region most involved in dealing with contextual interference. These results contrast with the results of the present study, in which the increase in BA 10 activity was observed in the high discriminability condition, in which there was no interference from related information in memory. A possible explanation for this discrepancy could lie in the different nature of the interference induced in these two studies as compared with the present study. In the study of Burgess et al. [2001], the interference was at the level of associations between a person, a place, and an object, whereas in the present study it was at the perceptual level of different segments from a visual scene. Since in their first study there was only two different places and people, it is possible that the perceptual memory for these two people and places was much more salient than it was in the second study, in which there were 20 different places and people. In the present study memory in the high discriminability condition was also more salient, as no perceptual interference was induced by the presentation of a second related segment during encoding. The activity in the BA 10 could thus be a function of the perceptual salience of the stimuli in memory. However, another study specifically designed to test this hypothesis would be necessary to state that this is indeed the case.

Results for the contrast of high versus low discriminability observed in the present study contrast with those from the control versus interference contrast observed in Lepage et al. [2005], although the experimental design was quite similar. For instance, right lateralized parahippocampal activation was observed in the earlier study, whereas left lateralized hippocampus activation was observed in the present study. Moreover, the frontal cortex activations observed in the earlier study were included in BA 10, 11, 13, and 47, whereas in the present study there are observed in BA 10 and 45. It is important to note that no significant difference in reaction time or hit rate was observed between the control and interference conditions in the earlier study. Without behavioral evidence, it is difficult to assess if the presentation of the second related segments indeed created interference and thus if the first segments were more salient in memory. Moreover, in the earlier study the second related segments were presented in a second block following the presentation of all the first segments instead of being interspersed in one run. This separation of the first and second segments into different blocks in the design could have reduced or eliminated the interference produced by the second segments.

CONCLUSION

The present fMRI study examined the neural correlates of the fan effect elicited by the recognition of segments of complex visual scenes as a function of the number of different segments presented during the encoding phase. At the behavioral level, the presence of the fan effect was confirmed by significantly lower reaction times for the high discriminability condition (only one segment presented during encoding) than for the low discriminability condition (two different related segments presented during encoding). At the brain level, a region of the right prefrontal cortex, several regions of the parietal cortex, and several late visual processing areas, including occipito‐temporal regions exhibited an increase in activity for the low compared with the high discriminability condition. The inverse contrast (high vs. low discriminability condition) was associated with an increase in activity in bilateral medial temporal gyri, including the left enthorinal cortex/hippocampus and in several bilateral prefrontal cortex regions.

The contribution of the present study is threefold. First, it shows that a behavioral fan effect can be observed with complex visual scenes as study material. Second, it supports the hypothesis that a region of the prefrontal cortex is involved in the control of memory interference at retrieval elicited by the amount of related information in memory, and further suggests that this involvement is right‐lateralized for nonverbal material (at least for complex visual scenes). Third, it suggests that being able to better discriminate information in memory could be a basis for the use of conscious recollection over the use of familiarity in a memory test.