CoChat

A predictive coding account of bistable perception - a model-based fMRI study

Open access

The DOI matches its Crossref record; 42 of the 43 references it cites were found in a registry or an index.

Abstract

In bistable vision, subjective perception wavers between two interpretations of a constant ambiguous stimulus. This dissociation between conscious perception and sensory stimulation has motivated various empirical studies on the neural correlates of bistable perception, but the neurocomputational mechanism behind endogenous perceptual transitions has remained elusive. Here, we recurred to a generic Bayesian framework of predictive coding and devised a model that casts endogenous perceptual transitions as a consequence of prediction errors emerging from residual evidence for the suppressed percept. Data simulations revealed close similarities between the model's predictions and key temporal characteristics of perceptual bistability, indicating that the model was able to reproduce bistable perception. Fitting the predictive coding model to behavioural data from an fMRI-experiment on bistable perception, we found a correlation across participants between the model parameter encoding perceptual stabilization and the behaviourally measured frequency of perceptual transitions, corroborating that the model successfully accounted for participants' perception. Formal model comparison with established models of bistable perception based on mutual inhibition and adaptation, noise or a combination of adaptation and noise was used for the validation of the predictive coding model against the established models. Most importantly, model-based analyses of the fMRI data revealed that prediction error time-courses derived from the predictive coding model correlated with neural signal time-courses in bilateral inferior frontal gyri and anterior insulae. Voxel-wise model selection indicated a superiority of the predictive coding model over conventional analysis approaches in explaining neural activity in these frontal areas, suggesting that frontal cortex encodes prediction errors that mediate endogenous perceptual transitions in bistable perception. Taken together, our current work provides a theoretical framework that allows for the analysis of behavioural and neural data using a predictive coding perspective on bistable perception. In this, our approach posits a crucial role of prediction error signalling for the resolution of perceptual ambiguities.

Overview

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.1371/journal.pcbi.1005536

Cites

42 of 42 references verified0 citation errors · 0 not checkable

Statistics

Figures

Fig 1 Modelling procedures. A. In the modelling approach illustrated here, we capture the temporal dynamics of bistable perception by changes in a continuously updated stability prior, which is combined with a bimodal likelihood representing the sensory input (see ‘feedback’ arrow). Under ambiguous viewing conditions, the likelihood contains equivalent evidence for both perceptual interpretations of the bistable stimulus. The mean of the prior ‘perceptual stability’ is defined by μstability, which corresponds to the preceding perceptual decision y(t − 1) (here centered around ‘1’ for counter-clockwise rotation of the Lissajous figure). The impact of the prior on the bimodal likelihood is determined by its precision (the inverse of variance) πstability. If a new perceptual decision was adopted at the preceding overlapping configuration of the Lissajous figure, this precision is set to πinit. Otherwise, πstability is repeatedly updated by a prediction error signal. This signal results from residual evidence for the alternative explanation of the bistable stimulus and is given by the difference between P(θ > 0.5) and the current perceptual decision y(t) (see ‘feedforward’ arrow). In this example, the prediction error signal stems from remaining evidence for clockwise rotation (centered around ‘0’), as the current perceptual decision represents counter-clockwise rotation (y(t) = 1) of the stimulus. Overtime, the stability prior is weakened, which is accompanied by an increasing probability for a novel transition in perception. B. Here, we depict the temporal evolution of the stability prior (left panel) and the corresponding posterior (right panel) at three successive overlapping configurations of the Lissajous figure (dark to light blue). As the precision of the stabilizing prior is gradually reduced, the posterior relaxes to equivalent probability for both perceptual interpretations of the stimulus. This is accompanied by escalating prediction error signals and increased likelihood for a perceptual transition. C. Furthermore, our approach accounts for additional sensory evidence, which is realized by a stereodisparity signal and used to disambiguate the Lissajous figure in the ‘replay’ condition. To this end, we introduce a ‘stereodisparity’ distribution (characterized by mean μstereo and precision πstereo), which serves as a weight on the bimodal likelihood. In the ambiguous condition (left panel), μstereo is centered around 0.5 and is thus uninformative with regard to the two perceptual interpretations of the stimulus. In the replay condition (right panel), μstereo is centered around ‘0’ or ‘1’ (depending on the direction of stereodisparity). The strength of the bias in the direction of either percept introduced by the stereodisparity signal scales with the precision πstereo.
Fig 1 Modelling procedures. A. In the modelling approach illustrated here, we capture the temporal dynamics of bistable perception by changes in a continuously updated stability prior, which is combined with a bimodal likelihood representing the sensory input (see ‘feedback’ arrow). Under ambiguous viewing conditions, the likelihood contains equivalent evidence for both perceptual interpretations of the bistable stimulus. The mean of the prior ‘perceptual stability’ is defined by μstability, which corresponds to the preceding perceptual decision y(t − 1) (here centered around ‘1’ for counter-clockwise rotation of the Lissajous figure). The impact of the prior on the bimodal likelihood is determined by its precision (the inverse of variance) πstability. If a new perceptual decision was adopted at the preceding overlapping configuration of the Lissajous figure, this precision is set to πinit. Otherwise, πstability is repeatedly updated by a prediction error signal. This signal results from residual evidence for the alternative explanation of the bistable stimulus and is given by the difference between P(θ > 0.5) and the current perceptual decision y(t) (see ‘feedforward’ arrow). In this example, the prediction error signal stems from remaining evidence for clockwise rotation (centered around ‘0’), as the current perceptual decision represents counter-clockwise rotation (y(t) = 1) of the stimulus. Overtime, the stability prior is weakened, which is accompanied by an increasing probability for a novel transition in perception. B. Here, we depict the temporal evolution of the stability prior (left panel) and the corresponding posterior (right panel) at three successive overlapping configurations of the Lissajous figure (dark to light blue). As the precision of the stabilizing prior is gradually reduced, the posterior relaxes to equivalent probability for both perceptual interpretations of the stimulus. This is accompanied by escalating prediction error signals and increased likelihood for a perceptual transition. C. Furthermore, our approach accounts for additional sensory evidence, which is realized by a stereodisparity signal and used to disambiguate the Lissajous figure in the ‘replay’ condition. To this end, we introduce a ‘stereodisparity’ distribution (characterized by mean μstereo and precision πstereo), which serves as a weight on the bimodal likelihood. In the ambiguous condition (left panel), μstereo is centered around 0.5 and is thus uninformative with regard to the two perceptual interpretations of the stimulus. In the replay condition (right panel), μstereo is centered around ‘0’ or ‘1’ (depending on the direction of stereodisparity). The strength of the bias in the direction of either percept introduced by the stereodisparity signal scales with the precision πstereo.CC BY · PubMed Central
Fig 2 Simulating perceptual decisions during ambiguous stimulation. Data were simulated using πinit of 3.5 at a sampling rate of 0.33 Hz for a total of 6*105 seconds. The distribution of phase durations followed a sharp rise and slow fall resembling a gamma-distribution. The insets A-C show simulated perceptual time-courses (grey dotted lines) next to updated model quantities (black solid lines). A: Prediction errors increase during a dominance phase and are reduced by perceptual transitions. B: Bistable perception can be conceived as resulting from subsequent sampling from a bimodal probability distribution [19], the weight of which is expressed by P(θ > 0.5). This weight is close to 0 or 1 at the beginning of a dominance phase (low transition probability) and gradually relaxes to 0.5 (high transition probability). C: The variance (inverse precision) of the prior distribution ‘perceptual stability’ increases as a consequence of prediction errors and is set to 1/πinit after a transition in perception.
Fig 2 Simulating perceptual decisions during ambiguous stimulation. Data were simulated using πinit of 3.5 at a sampling rate of 0.33 Hz for a total of 6*105 seconds. The distribution of phase durations followed a sharp rise and slow fall resembling a gamma-distribution. The insets A-C show simulated perceptual time-courses (grey dotted lines) next to updated model quantities (black solid lines). A: Prediction errors increase during a dominance phase and are reduced by perceptual transitions. B: Bistable perception can be conceived as resulting from subsequent sampling from a bimodal probability distribution [19], the weight of which is expressed by P(θ > 0.5). This weight is close to 0 or 1 at the beginning of a dominance phase (low transition probability) and gradually relaxes to 0.5 (high transition probability). C: The variance (inverse precision) of the prior distribution ‘perceptual stability’ increases as a consequence of prediction errors and is set to 1/πinit after a transition in perception.CC BY · PubMed Central
Fig 3 Posterior model parameters. A: The geometric mean (i.e. the arithmetic mean in log-space) of posterior πinit and πstereo, averaged across runs and participants, and standard error of the mean. B: Mean prediction errors averaged across runs and participants for ambiguous and replay blocks and standard error of the mean. Prediction errors were significantly decreased during replay stimulation (two-sample t-test, p < 10−6, t19 = 7.69). C: Average transition probabilities correlated significantly with average πinit for individual participants (ρ = −0.88, p < 10−7, Pearson correlation), providing a sanity check for model fit. D: Transition probabilities from run 3 were predictive of posterior πinit averaged over run 1 and 2. The significant Pearson correlation between the two independent measures (ρ = −0.76, p < 10−4) illustrates the predictive power of the model.
Fig 3 Posterior model parameters. A: The geometric mean (i.e. the arithmetic mean in log-space) of posterior πinit and πstereo, averaged across runs and participants, and standard error of the mean. B: Mean prediction errors averaged across runs and participants for ambiguous and replay blocks and standard error of the mean. Prediction errors were significantly decreased during replay stimulation (two-sample t-test, p < 10−6, t19 = 7.69). C: Average transition probabilities correlated significantly with average πinit for individual participants (ρ = −0.88, p < 10−7, Pearson correlation), providing a sanity check for model fit. D: Transition probabilities from run 3 were predictive of posterior πinit averaged over run 1 and 2. The significant Pearson correlation between the two independent measures (ρ = −0.76, p < 10−4) illustrates the predictive power of the model.CC BY · PubMed Central
Fig 4 Model-based fMRI results from standard GLM (A, B) and PPM (C) analyses. GLMs are displayed using FWE correction at p < 0.05. For PPM results, we show voxels above an exceedance probability of 99% with a cluster size n > 10. A: 2-level contrast for ‘Transition vs. baseline’ showing activations left pre- and postcentral gyrus, right inferior frontal gyrus, right inferior parietal lobulus and right middle frontal gyrus with qualitatively equivalent results for all models. B: ‘PE vs. baseline’ (‘PE model’) yielded activations in bilateral insulae and inferior frontal gyri. We found no whole-brain correctable voxels for ‘Ambiguity vs. baseline’ (‘Block model’) nor for ‘Ambiguous vs. replay transitions’ (‘Conventional model’). C: Group exceedance probability maps with right insula, right inferior frontal gyrus, right posterior-medial frontal gyrus as well as left precentral gyrus showed strongest evidence for the ‘PE model’ as compared to the control models.
Fig 4 Model-based fMRI results from standard GLM (A, B) and PPM (C) analyses. GLMs are displayed using FWE correction at p < 0.05. For PPM results, we show voxels above an exceedance probability of 99% with a cluster size n > 10. A: 2-level contrast for ‘Transition vs. baseline’ showing activations left pre- and postcentral gyrus, right inferior frontal gyrus, right inferior parietal lobulus and right middle frontal gyrus with qualitatively equivalent results for all models. B: ‘PE vs. baseline’ (‘PE model’) yielded activations in bilateral insulae and inferior frontal gyri. We found no whole-brain correctable voxels for ‘Ambiguity vs. baseline’ (‘Block model’) nor for ‘Ambiguous vs. replay transitions’ (‘Conventional model’). C: Group exceedance probability maps with right insula, right inferior frontal gyrus, right posterior-medial frontal gyrus as well as left precentral gyrus showed strongest evidence for the ‘PE model’ as compared to the control models.CC BY · PubMed Central
Fig 5 Average eigenvariate time-courses. For visualization, we extracted eigenvariate time-courses from right insula, left insula, right IFG and left IFG (A–D), aligned all phase durations to the timepoint of the upcoming perceptual transition and averaged within and across observers. In analogy to modelled prediction error trajectories, mean eigenvariate time-courses (± standard error of the mean) showed a gradual increase towards a transition in perception.
Fig 5 Average eigenvariate time-courses. For visualization, we extracted eigenvariate time-courses from right insula, left insula, right IFG and left IFG (A–D), aligned all phase durations to the timepoint of the upcoming perceptual transition and averaged within and across observers. In analogy to modelled prediction error trajectories, mean eigenvariate time-courses (± standard error of the mean) showed a gradual increase towards a transition in perception.CC BY · PubMed Central

Verification

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

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?

Self-check

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

The self-check needs the paper’s full text.

Citations

Cites

Cited by

See all 120

Metadata

Authors
Weilnhammer, V., Stuke, H., Hesselmann, G., Sterzer, P., & Schmack, K.
Affiliations
  • Charité – Universitätsmedizin Berlin (DE)
  • Humboldt-Universität zu Berlin (DE)
Corresponding author
Weilnhammer, Veith
Venue
PLOS Computational Biology
Published
2017-05-15
Volume
13
Issue
5
Pages
e1005536
ISSN
  • 1553-734X (ISSN-L)
  • 1553-7358
OpenAlex
W2615931912
Cited by
120
References
43
Citations by year
2026: 12 · 2025: 13 · 2024: 12 · 2023: 12 · 2022: 12 · 2021: 15 · 2020: 24 · 2019: 5 · 2018: 10 · 2017: 5
FWCI
6.91
Type
journal article
Language
en
Open access
Locations
  • PLoS Computational Biology (publishedVersion, cc-by)
  • Europe PMC (PubMed Central) (publishedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • RePEc: Research Papers in Economics (submittedVersion)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
  • Figshare (submittedVersion, cc-by)
  • Refubium (Universitätsbibliothek der Freien Universität Berlin) (submittedVersion, cc-by)
  • Universitätsbibliothek der FU Berlin Hochschulschriftenstelle u. Dokumentenserver (cc-by)
Funders
  • e:Med research and funding concept (01ZX1404A)
  • German Research Foundation (HE 6244/1-2, STE 1430/7-1)
  • Charité Clinical Scientist Program
  • Charité Junior Clinical Scientist Program
Topics
  • Life Sciences › Neuroscience › Cognitive Neuroscience › Visual perception and processing mechanisms
  • Life Sciences › Neuroscience › Cognitive Neuroscience › Embodied and Extended Cognition
  • Social Sciences › Arts and Humanities › Philosophy › Philosophy and Theoretical Science
Keywords
bistable perception, predictive coding, prediction error, perceptual transitions, perceptual stabilization, Bayesian model, inferior frontal gyrus, anterior insula, model-based fMRI, model comparison, reciprocal inhibition, adaptation, noise
MeSH terms
Adult, Bayes Theorem, Computer Simulation, Female, Frontal Lobe, Humans, Magnetic Resonance Imaging, Male, Models, Neurological, Task Performance and Analysis, Visual Perception, Computational Biology, Young Adult

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.

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

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.