CoChat
Sign Up
Cited by 528Visual perception and processing mechanismsNeural dynamics and brain functionFace Recognition and Perception

Shared representations for working memory and mental imagery in early visual cortex

Open access
Albers, Anke Marit, Kok, Peter, Toni, Ivan, Dijkerman, H. Chris, de Lange, Floris P.·2013·Current Biology
DOI 10.1016/j.cub.2013.05.065
Open this paper in CoChat — free accountOpen PDF

Abstract

Early visual areas contain specific information about visual items maintained in working memory, suggesting a role for early visual cortex in more complex cognitive functions [1-4]. It is an open question, however, whether these areas also underlie the ability to internally generate images de novo (i.e., mental imagery). Research on mental imagery has to this point focused mostly on whether mental images activate early sensory areas, with mixed results [5-7]. Recent studies suggest that multivariate pattern analysis of neural activity patterns in visual regions can reveal content-specific representations during cognitive processes, even though overall activation levels are low [1-4]. Here, we used this approach [8, 9] to study item-specific activity patterns in early visual areas (V1-V3) when these items are internally generated. We could reliably decode stimulus identity from neural activity patterns in early visual cortex during both working memory and mental imagery. Crucially, these activity patterns resembled those evoked by bottom-up visual stimulation, suggesting that mental images are indeed "perception-like" in nature. These findings suggest that the visual cortex serves as a dynamic "blackboard" [10, 11] that is used during both bottom-up stimulus processing and top-down internal generation of mental content.

AI snapshot

Written from the paper’s full text.

TL;DR

Early visual cortex (V1–V3) contains orientation-specific multivoxel patterns during both visual working memory and mental imagery, and these patterns generalize to those evoked by bottom-up grating perception. This supports a shared, perception-like code in early visual cortex for maintained and internally generated images.

Key findings

  • Working memory (WM) decoding of maintained orientation in V1–V3 was above chance: 54% vs 33.3% chance, t(23) = 5.88, p < 1 × 10⁻²⁵, Cohen’s d = 1.21.
  • Mental imagery (IM) decoding of internally generated orientation was also above chance: 46%, t(23) = 3.09, p = 0.005, Cohen’s d = 0.63; cross-task decoding was reliable for WM/IM (45%, t(23) = 3.88, p < 1 × 10⁻²³, d = 0.78) and IM/WM (45%, t(23) = 4.13, p < 1 × 10⁻²³, d = 0.83).
  • A classifier trained on actual grating perception decoded orientation during WM (46%, t(23) = 4.50, p < 1 × 10⁻²⁴, d = 0.90) and IM (49%, t(23) = 5.92, p < 1 × 10⁻²⁵, d = 1.21), indicating shared bottom-up/top-down activity patterns.
  • Time-resolved decoding showed the target could be decoded from 4 s after delay onset in WM but only from 8 s after delay onset in IM; mean neural activity decreased during the delay and was indistinguishable between WM and IM.

Methods

  • N = 24 participants performed fMRI tasks with retinotopically defined early visual areas V1, V2, and V3 (360 voxels combined). Participants either maintained a grating in WM or mentally rotated a cued grating and held the generated image for a 10 s delay.
  • Multivariate pattern analysis/classifiers were trained on WM, IM, or perceptual localizer data and tested within and across tasks using leave-one-run-out cross-validation, including time-resolved decoding.
  • Behavioral data confirmed successful task performance, with increasing reaction times as a function of the amount of mental transformation (Figure S1).

Limitations

  • The paper states that generalization between bottom-up and top-down patterns was robust but not perfect, possibly because internally generated images are less robust and more variable due to internal fluctuations in attentional state; it also notes possible attention during the perceptual localizer, though eye movements were considered unlikely because of fixation task and brief 200 ms stimuli.
  • The paper states that the exact role and representational content of frontoparietal regions during mental imagery remain to be determined; nonvisual areas did not show reliable stimulus-related information when training on perceptual input (chance-level generalization, all p > 0.05).
  • Evident from the text: decoding accuracies were modest (46–54% vs 33.3% chance) and based on a small set of three orientations; the fMRI decoding design is correlational and cannot establish a causal role for V1–V3, and the mental rotation task may recruit attention/executive processes not fully separated from imagery.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1016/j.cub.2013.05.065

Cites

38 of 38 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

0 of 14 cited sources checked · 0 supported · 0 flagged

  • Findingown data

    Early visual cortex (V1-V3) contains information about maintained content during working memory.

    “We found that early visual cortex (V1-V3) indeed contained information about maintained content [WM: decoding accuracy 54%, chance level 33.3%: t(23) = 5.88, p < 1 3 10 25 ] in the period 8-12 s after onset of maintenance).”
    • Also stated in: § Discussion · hedged
      “We found that activity patterns in early visual areas (V1-V3) could reliably predict which of three oriented gratings was either held in working memory or mentally imagined, even though overall levels of neural activity were low.”
    Quote in the paperSubstantial claim
  • Findingown data

    A classifier trained on actual perception of gratings can discriminate orientation from early visual cortex activity patterns during working memory trials.

    “This ''perceptual'' classifier could also reliably discriminate between activity patterns in early visual cortex evoked by the different orientations during both WM trials [decoding accuracy 46%, t(23) = 4.50, p < 1 3 10 24 , Cohen's d = 0.90] and IM trials [decoding accuracy 49%, t(23) = 5.92 p < 1 3 10 25 , Cohen's d = 1.21].”
    Quote in the paperSubstantial claim
  • Findingown data

    A classifier trained on actual perception of gratings can discriminate orientation from early visual cortex activity patterns during mental imagery trials.

    “This ''perceptual'' classifier could also reliably discriminate between activity patterns in early visual cortex evoked by the different orientations during both WM trials [decoding accuracy 46%, t(23) = 4.50, p < 1 3 10 24 , Cohen's d = 0.90] and IM trials [decoding accuracy 49%, t(23) = 5.92 p < 1 3 10 25 , Cohen's d = 1.21].”
    Quote in the paperSubstantial claim
  • Findingown data

    Early visual cortex (V1-V3) contains information about internally generated images during mental imagery.

    “Indeed, early visual cortex also contained information about internally generated images [IM: decoding accuracy 46%, t( 23 S1; all accuracies > 39%, all p < 0.007).”
    • Also stated in: § Discussion · hedged
      “We found that activity patterns in early visual areas (V1-V3) could reliably predict which of three oriented gratings was either held in working memory or mentally imagined, even though overall levels of neural activity were low.”
    Quote in the paperSubstantial claim
  • Findingown data

    During mental imagery trials, shortly after stimulus presentation, the cued starting grating was predominantly selected by the classifier.

    “Again, shortly after this, the cued (i.e., starting orientation that had to be mentally rotated) grating was predominantly selected by the classifier.”
    Quote in the paperSubstantial claim
  • May combine two claims. 13

  • May combine two claims.

    Findingown data

    Activity patterns for working memory and mental imagery trials were highly similar, as shown by reliable cross-decoding in both directions.

    “Moreover, activity patterns for WM and IM trials were highly similar: when training the multivariate pattern classifier on the delay period during WM and testing on the delay period during IM, we found equally reliable pattern information [WM/IM decoding accuracy 45%, t(23) = 3.88, p < 1 3 10 23 , Cohen's d = 0.78].”
    • Also stated in: § Supplemental Information · plain
      “Training on IM and testing on WM also resulted in reliable classification [IM/WM decoding accuracy 45%, t(23) = 4.13, p < 1 3 10 23 , Cohen's d = 0.83].”
    • Also stated in: § Discussion · plain
      “In conclusion, we observed analogous sensory representations during visual working memory and mental imagery in early visual cortex.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Some regions within the frontoparietal network, notably left parietal cortex and supplementary motor area, showed some evidence of stimulus information when training and testing within the main experiment, but generalization from the perceptual classifier to the main task resulted in chance-level performance.

    “Interestingly, although some of these regions within this network (notably the left parietal cortex and supplementary motor area) showed some evidence of stimulus information when training and testing within the main experiment (Table S2), generalization from the perceptual classifier to the main task resulted in chance-level performance (all p > 0.05).”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingcitation

    Overall neural activity levels appeared low during mental imagery and working memory, even though patterns in early visual cortex carried stimulus information during this period.

    “It may seem surprising that overall neural activity levels appeared low during mental imagery and working memory, even though the patterns in early visual cortex carried stimulus information during this period.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    During working memory, the target could be decoded as early as 4 s after delay-period onset, whereas during mental imagery the target could only be decoded from 8 s after delay-period onset.

    “During WM, the target could be decoded as early as 4 s after delay-period onset, whereas during IM the target could only be decoded from 8 s after delay-period onset.”
    • Also stated in: § Supplemental Information · plain
      “Decoding was significant from 4 to 16 s for WM trials and from 8 to 16 s for IM trials (all p < 0.001).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingreasoning

    Generalization from stimulus-driven activity patterns to mental imagery-induced activity patterns in early visual cortex was robust but not perfect.

    “It should be noted here that generalization was robust but not perfect, which may be due to the fact that internally generated images can lead to less robust and more variable activation patterns than bottom-up visual stimulation, due to internal fluctuations in attentional state.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    The internally generated target grating was not physically presented on that trial but was mentally created by participants and after several seconds became the preferred orientation of the classifier.

    “This target grating was not physically presented on that trial but was mentally created by the participants and after several seconds became the preferred orientation of the classifier.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    During working memory trials, after stimulus presentation, visual cortical activity patterns initially reflected a combination of the two presented gratings, and subsequently the cued grating was predominantly selected by the classifier.

    “During WM trials, the classifier was initially at chance, selecting each option approximately one-third of the time.After stimulus presentation, visual cortical (V1-V3) activity patterns reflected a combination of the two presented gratings.Subsequently, the cued (i.e., to be remembered) grating was predominantly selected by the classifier (Figure 2A).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationown data

    Activity patterns during mental imagery resemble those elicited by physically presented stimuli, suggesting analogous neural codes for internally generated mental images and stimulus representations.

    “Crucially, by training on patterns of activity during physical presentation of gratings, we show that activity patterns during mental imagery resemble those elicited by physically presented stimuli, suggesting analogous neural codes for internally generated mental images and stimulus representations.”
    • Also stated in: § Abstract · hedged
      “Crucially, these activity patterns resembled those evoked by bottomup visual stimulation, suggesting that mental images are indeed ''perception-like'' in nature.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationreasoning

    The delay in mental imagery target decoding likely reflects a combination of factors: selecting the cued grating, retrieving the task cue, and performing mental rotation.

    “This delay likely reflects a combination of factors.During imagery trials, participants had to not only select the cued grating but also retrieve the task cue, which instructed them about the direction and extent of mental rotation.They subsequently had to perform the mental rotation, with each of these steps contributing to the delay in the formation of the internally generated target image.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationown data

    There are three sequential stages of representation in early visual cortex during mental imagery: first, physically presented stimuli are represented; second, one of the presented stimuli is selected for transformation; third, a new representation is formed in early visual cortex.

    “This suggests three sequential stages of representation in early visual cortex during the mental imagery process: first, the physically presented stimuli are represented; second, one of the presented stimuli is selected for transformation; and third, a new representation is formed in early visual cortex.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationcitation

    One reason for low overall activity may be that visual areas exhibit high spontaneous activity during rest, whose functional significance may be similar to mental imagery.

    “One reason for this may be that visual areas also exhibit an overall high level of spontaneous activity during rest [34], the functional significance of which may be quite similar to mental imagery [35].”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites [34] Kenet, 2003, Spontaneously emerging cortical representations of visual attributes, Nature source

      Not checked in the free run.

    • cites [35] Kosslyn, 1995, Topographical representations of mental images in primary visual cortex, Nature source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    The idea that mental imagery plays a functional role in facilitating future perception is supported by a recent study that found that mental imagery biases subsequent perception in a binocular rivalry task, as well as by the correlation between mental imagery performance and representational precision.

    “The idea that mental imagery plays a functional role in facilitating future perception is supported by a recent study that found that mental imagery biases subsequent perception in a binocular rivalry task [35,38], as well as by the correlation between IM performance and representational precision (Figure S2).”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [35] Kosslyn, 1995, Topographical representations of mental images in primary visual cortex, Nature source

      Not checked in the free run.

    • cites [38] Pearson, 2008, The functional impact of mental imagery on conscious perception, Curr. Biol. source

      Not checked in the free run.

  • May combine two claims.

    Interpretationown data

    Mental imagery partly depends on the same mechanisms as visual perception, in line with depictive accounts of mental representations.

    “By showing that there is content-specific overlap of activation patterns during mental imagery and bottom-up visual stimulation in primary visual cortex, we show that mental imagery partly depends on the same mechanisms as visual perception, in line with depictive accounts of mental representations [20].”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [20] Lee, 2012, Disentangling visual imagery and perception of real-world objects, Neuroimage source

      Not checked in the free run.

13 of 13

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

  • Imagery can be decoded from early visual areas.

    “studies using techniques to decode brain activity suggest that imagery can be decoded from early visual areas”

    From Aphantasia and hyperphantasia: exploring imagery vividness extremes (2024) · cites this paper as [12]

  • Activity patterns in nonstimulated parts of early visual cortex contain content-specific information from the surrounding visual context, from objects presented in the periphery, and from visual stimuli solely held in working memory rather than being actively perceived.

    “For example, activity patterns in nonstimulated parts of early visual cortex contain content-specific information from the surrounding visual context [31, 32], from objects presented in the periphery [33], and from visual stimuli solely held in working memory rather than being actively perceived [34, 35].”

    From Decoding sound and imagery content in early visual cortex (2014) · cites this paper as [35]

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

  • The third grating orientation is listed as both 135° and 115°

    Minor

    The Results list the three possible orientations as 15, 75, and 135, but the experimental-design description lists 15, 75, and 115.

1 of 1

Cites

Flagged first, then by citations

Cited by

Sorted by citations

Related

Authors
Albers, A. M., Kok, P., Toni, I., Dijkerman, H. C., & de Lange, F. P.
Affiliations
  • Radboud University Nijmegen (NL)
  • Utrecht University (NL)
ORCID
Venue
Current Biology
Publisher
Elsevier BV
Published
2013-08
Volume
23
Issue
15
Pages
1427-1431
ISSN
0960-9822 (ISSN-L)
DOI
10.1016/j.cub.2013.05.065
OpenAlex
W1969387624
Cited by
528
References
38
Citations by year
2026: 34 · 2025: 47 · 2024: 72 · 2023: 55 · 2022: 38 · 2021: 56 · 2020: 39 · 2019: 37 · 2018: 31 · 2017: 34 · 2016: 28 · 2015: 29 · 2014: 23 · 2013: 4
FWCI
15.30
Type
journal article
Language
en
Open access
Licence
Locations
  • Current Biology (publishedVersion)
  • PubMed (publishedVersion)
  • Radboud Repository (Radboud University) (submittedVersion)
Topics
  • Life Sciences › Neuroscience › Cognitive Neuroscience › Visual perception and processing mechanisms
  • Life Sciences › Neuroscience › Cognitive Neuroscience › Neural dynamics and brain function
  • Life Sciences › Neuroscience › Cognitive Neuroscience › Face Recognition and Perception
Keywords
working memory, mental imagery, early visual cortex, multivariate pattern analysis, shared representations
MeSH terms
Humans, Imagination, Functional Laterality, Magnetic Resonance Imaging, Memory, Short-Term, Photic Stimulation, Reaction Time, Visual Cortex, Visual Perception, Multivariate Analysis, Nontherapeutic Human Experimentation

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-10.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-10.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review