CoChat

An implementation of integrated information theory in resting-state fMRI

Open access

The DOI matches its Crossref record; 72 of the 73 references it cites were found in a registry or an index.

Abstract

Integrated Information Theory was developed to explain and quantify consciousness, arguing that conscious systems consist of elements that are integrated through their causal properties. This study presents an implementation of Integrated Information Theory 3.0, the latest version of this framework, to functional MRI data. Data were acquired from 17 healthy subjects who underwent sedation with propofol, a short-acting anaesthetic. Using the PyPhi software package, we systematically analyze how Φ max, a measure of integrated information, is modulated by the sedative in different resting-state networks. We compare Φ max to other proposed measures of conscious level, including the previous version of integrated information, Granger causality, and correlation-based functional connectivity. Our results indicate that Φ max presents a variety of sedative-induced behaviours for different networks. Notably, changes to Φ max closely reflect changes to subjects’ conscious level in the frontoparietal and dorsal attention networks, which are responsible for higher-order cognitive functions. In conclusion, our findings present important insight into different measures of conscious level that will be useful in future implementations to functional MRI and other forms of neuroimaging.

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 · 10.1038/s42003-023-05063-y

Cites

72 of 72 references verified0 citation errors · 0 not checkable

Statistics

Figures

Fig. 1 Summary of acquisition, signal processing, and metric extraction. a fMRI data were acquired from 17 healthy subjects and propofol was administered to obtain measurements for four conditions of awareness: Awake: no propofol administered, Mild: Ramsey Scale Level 3 (limited responsiveness), Deep: Ramsey Scale Level 5 (no responsiveness), Recovery: administration of anesthetic terminated (see Methods for more details). We then applied a parcellation scheme and obtained 11 RSNs consisting of five regions, each with their own time-series. The time-series of each region was binarized with respect to its own signal mean, which allowed each network to take on 32 (or 25) possible states in any given time point. Note that we concatenated (joined) the time-series from individual subjects to obtain longer signals for each RSN and condition. b The metric derived from IIT 3.0 is μ[Φmax]: Φmax is computed for a certain state using the earth mover’s distance (EMD) between the cause-and-effect conceptual space of the system S and its partitioned counterparts M. We computed Φmax for each state and obtained a weighted average over the time-series. The three additional metrics included in our analysis are: Φ*, integrated information from the decoding perspective, which is computed as the difference between the mutual information I of the system’s past and present states (S^(t−τ),S^t) and the sum over the mutual information of the partitioned parts (Mk^(t−τ),Mk^t); CD, causal density, the mean Granger causality F between every possible pair of regions in the network; and μ[ρ], the mean Pearson correlation coefficient ρ between every possible pair of regions in the network.
Fig. 1 Summary of acquisition, signal processing, and metric extraction. a fMRI data were acquired from 17 healthy subjects and propofol was administered to obtain measurements for four conditions of awareness: Awake: no propofol administered, Mild: Ramsey Scale Level 3 (limited responsiveness), Deep: Ramsey Scale Level 5 (no responsiveness), Recovery: administration of anesthetic terminated (see Methods for more details). We then applied a parcellation scheme and obtained 11 RSNs consisting of five regions, each with their own time-series. The time-series of each region was binarized with respect to its own signal mean, which allowed each network to take on 32 (or 25) possible states in any given time point. Note that we concatenated (joined) the time-series from individual subjects to obtain longer signals for each RSN and condition. b The metric derived from IIT 3.0 is μ[Φmax]: Φmax is computed for a certain state using the earth mover’s distance (EMD) between the cause-and-effect conceptual space of the system S and its partitioned counterparts M. We computed Φmax for each state and obtained a weighted average over the time-series. The three additional metrics included in our analysis are: Φ*, integrated information from the decoding perspective, which is computed as the difference between the mutual information I of the system’s past and present states (S^(t−τ),S^t) and the sum over the mutual information of the partitioned parts (Mk^(t−τ),Mk^t); CD, causal density, the mean Granger causality F between every possible pair of regions in the network; and μ[ρ], the mean Pearson correlation coefficient ρ between every possible pair of regions in the network.CC BY · PubMed Central
Fig. 2 Results from spatial and temporal control procedures. For each of the four conditions, we generated 100 spatial control networks by randomly grouping five regions from different RSNs. Each RSN’s time-series was then randomly permuted 50 times in the temporal control procedure, leading to 550 randomized time-series for 11 RSNs in each condition. Note that these permutations and calculations were performed for time-series with all 17 subjects concatenated (see Methods for more details). We computed μ[Φmax] (a), Φ* (b), CD (c), and μ[ρ] (d) for the original RSNs (shown with the gray points) and both control permutations (shown with the violin plots). The horizontal lines corresponding to each distribution represent their mean. In the case of μ[ρ] (d), the overlapped data points indicate that the original RSNs’ values did not change when time points were permuted. For source data, see Supplementary Data 2. For statistical information, see Supplementary Note 6, Supplementary Table 3.
Fig. 2 Results from spatial and temporal control procedures. For each of the four conditions, we generated 100 spatial control networks by randomly grouping five regions from different RSNs. Each RSN’s time-series was then randomly permuted 50 times in the temporal control procedure, leading to 550 randomized time-series for 11 RSNs in each condition. Note that these permutations and calculations were performed for time-series with all 17 subjects concatenated (see Methods for more details). We computed μ[Φmax] (a), Φ* (b), CD (c), and μ[ρ] (d) for the original RSNs (shown with the gray points) and both control permutations (shown with the violin plots). The horizontal lines corresponding to each distribution represent their mean. In the case of μ[ρ] (d), the overlapped data points indicate that the original RSNs’ values did not change when time points were permuted. For source data, see Supplementary Data 2. For statistical information, see Supplementary Note 6, Supplementary Table 3.CC BY · PubMed Central
Fig. 3 Modulation of μ[Φmax] by propofol in individual RSNs. For each network, the mean integrated information across the subject population is compared for the four conditions of awareness. Bar heights represent the mean of the sample of μ[Φmax] values, which was obtained by concatenating 16 subjects with a different subject left out for each measurement (N = 17), with individual data points given for each measurement. Error bars represent the population standard deviation, which was obtained by multiplying the standard error of the sample by √N. All statistically significant differences found between the conditions, such as awake vs. deep sedation, are indicated by the horizontal lines above the bars (*p < α^BH, where α^BH is the significance threshold obtained with the Benjamini–Hochberg correction for multiple comparison, which varies when comparing different conditions. For all values derived from significance testing, see Supplementary Note 6; Supplementary Tables 4–7). For source data, see Supplementary Data 3.
Fig. 3 Modulation of μ[Φmax] by propofol in individual RSNs. For each network, the mean integrated information across the subject population is compared for the four conditions of awareness. Bar heights represent the mean of the sample of μ[Φmax] values, which was obtained by concatenating 16 subjects with a different subject left out for each measurement (N = 17), with individual data points given for each measurement. Error bars represent the population standard deviation, which was obtained by multiplying the standard error of the sample by √N. All statistically significant differences found between the conditions, such as awake vs. deep sedation, are indicated by the horizontal lines above the bars (*p < α^BH, where α^BH is the significance threshold obtained with the Benjamini–Hochberg correction for multiple comparison, which varies when comparing different conditions. For all values derived from significance testing, see Supplementary Note 6; Supplementary Tables 4–7). For source data, see Supplementary Data 3.CC BY · PubMed Central
Fig. 4 Modulation of reference metrics by propofol in individual RSNs. The networks, presented across rows, are analyzed using the reference metrics Φ*, CD, μ[ρ], which were assigned to columns. As before, each network is analyzed over the four conditions of awareness. To focus on each network’s behavior under anesthesia, the values were normalized with respect to each network’s awake value (set to equal 1). Bar heights represent the normalized means of CD, Φ*, and μ[ρ] values obtained by concatenating 16 subjects with a different subject left out for each measurement (i.e., these measures were applied to the same set of time-series used to compute μ[Φmax] in Fig. 3). Error bars represent the population standard deviation, which was obtained by multiplying the standard error of the sample by √N. For absolute measurements and statistical tests, see Supplementary Note 3; Supplementary Figs. 2–4, which includes an equivalent version of Fig. 3 for each measure. For source data, see Supplementary Data 3.
Fig. 4 Modulation of reference metrics by propofol in individual RSNs. The networks, presented across rows, are analyzed using the reference metrics Φ*, CD, μ[ρ], which were assigned to columns. As before, each network is analyzed over the four conditions of awareness. To focus on each network’s behavior under anesthesia, the values were normalized with respect to each network’s awake value (set to equal 1). Bar heights represent the normalized means of CD, Φ*, and μ[ρ] values obtained by concatenating 16 subjects with a different subject left out for each measurement (i.e., these measures were applied to the same set of time-series used to compute μ[Φmax] in Fig. 3). Error bars represent the population standard deviation, which was obtained by multiplying the standard error of the sample by √N. For absolute measurements and statistical tests, see Supplementary Note 3; Supplementary Figs. 2–4, which includes an equivalent version of Fig. 3 for each measure. For source data, see Supplementary Data 3.CC BY · PubMed Central
Fig. 5 Magnitude (M) vs. degree of modulation pattern that reflects awareness level (S). The overall magnitude (M) generated for a particular RSN is plotted against the degree to which a network reflects the four conditions of awareness (S) for each metric of interest: μ[Φmax] (a), Φ* (b), CD (c), and μ[ρ] (d). Vertical error bars represent the standard deviation of each network’s M, while horizontal error bars represent the standard error of each network’s S value. Within each subplot, the dashed vertical lines correspond to the median value of S obtained out of all networks. We use this median to separate RSNs whose modulation patterns strongly correspond to changes in awareness level. Networks are labeled as: FPN frontoparietal, DMN Default mode, RS Retrosplenial, DAN Dorsal attention, VAN Ventral attention, CO Cing. Opercular, CP Cing. Parietal, SMH Sensorimotor hand, SMM Sensorimotor mouth, VIS Visual, AUD Auditory. For source data, see Supplementary Data 4.
Fig. 5 Magnitude (M) vs. degree of modulation pattern that reflects awareness level (S). The overall magnitude (M) generated for a particular RSN is plotted against the degree to which a network reflects the four conditions of awareness (S) for each metric of interest: μ[Φmax] (a), Φ* (b), CD (c), and μ[ρ] (d). Vertical error bars represent the standard deviation of each network’s M, while horizontal error bars represent the standard error of each network’s S value. Within each subplot, the dashed vertical lines correspond to the median value of S obtained out of all networks. We use this median to separate RSNs whose modulation patterns strongly correspond to changes in awareness level. Networks are labeled as: FPN frontoparietal, DMN Default mode, RS Retrosplenial, DAN Dorsal attention, VAN Ventral attention, CO Cing. Opercular, CP Cing. Parietal, SMH Sensorimotor hand, SMM Sensorimotor mouth, VIS Visual, AUD Auditory. For source data, see Supplementary Data 4.CC BY · PubMed Central
Fig. 6 Obtaining five representative regions for an RSN. Starting with 40 regions in the DMN, a k-means algorithm was used to obtain five clusters, each associated with a centroid. The regions originally associated with each RSN were narrowed down to five clusters, which yielded the five regions used in our analysis. For the representative regions included in each of the 11 RSNs, see Supplementary Note 1; Supplementary Fig. 1.
Fig. 6 Obtaining five representative regions for an RSN. Starting with 40 regions in the DMN, a k-means algorithm was used to obtain five clusters, each associated with a centroid. The regions originally associated with each RSN were narrowed down to five clusters, which yielded the five regions used in our analysis. For the representative regions included in each of the 11 RSNs, see Supplementary Note 1; Supplementary Fig. 1.CC BY · PubMed Central
Fig. 7 Summary of procedure for computing μ[Φmax]. Each binarized time-series was used to construct a transition probability matrix (TPM) with dimensions 32 × 32, which was constructed by counting the number of times any one state transitioned to any other. Calculations of Φ are performed for a subsystem, which is defined by the state of the system and a subset of two or more nodes within it. For each state appearing in the time-series, the maximally integrated conceptual information Φmax was obtained from the subset of nodes that maximizes Φ. Finally, we count the occurrences of each state and compute the weighted average of Φmax over the time-series, which we refer to as μ[Φmax].
Fig. 7 Summary of procedure for computing μ[Φmax]. Each binarized time-series was used to construct a transition probability matrix (TPM) with dimensions 32 × 32, which was constructed by counting the number of times any one state transitioned to any other. Calculations of Φ are performed for a subsystem, which is defined by the state of the system and a subset of two or more nodes within it. For each state appearing in the time-series, the maximally integrated conceptual information Φmax was obtained from the subset of nodes that maximizes Φ. Finally, we count the occurrences of each state and compute the weighted average of Φmax over the time-series, which we refer to as μ[Φmax].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

Show all 72 references

Cited by

See all 23

Metadata

Authors
Nemirovsky, I. E., Popiel, N. J. M., Rudas, J., Caius, M., Naci, L., Schiff, N. D., Owen, A. M., & Soddu, A.
Affiliations
  • Western University (CA)
  • University of Cambridge (GB)
  • Universidad Nacional de Colombia (CO)
  • Weill Cornell Medicine (US)
Corresponding author
Nemirovsky, Idan E.
Venue
Communications Biology
Published
2023-07-05
Volume
6
Issue
1
Pages
692-692
ISSN
2399-3642 (ISSN-L)
OpenAlex
W4383263660
Cited by
15
References
73
Citations by year
2026: 2 · 2025: 7 · 2024: 4 · 2023: 2
FWCI
2.24
Type
journal article
Language
en
Open access
Licence
Locations
  • Communications Biology (publishedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
Funders
Gouvernement du Canada | Natural Sciences and Engineering Research Council of Canada
Topics
  • Life Sciences › Neuroscience › Cognitive Neuroscience › Functional Brain Connectivity Studies
  • Social Sciences › Arts and Humanities › Philosophy › Philosophy and Theoretical Science
  • Life Sciences › Neuroscience › Cognitive Neuroscience › Neural dynamics and brain function
Keywords
integrated information theory, consciousness, resting-state networks, propofol sedation, Granger causality, functional connectivity, frontoparietal network, dorsal attention network
MeSH terms
Consciousness, Humans, Hypnotics and Sedatives, Information Theory, Magnetic Resonance Imaging, Propofol

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

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: An implementation of integrated information theory in resting-state fMRI. https://app.cochat.ai/papers/10.1038/s42003-023-05063-y
The paper
Nemirovsky, I. E., Popiel, N. J. M., Rudas, J., Caius, M., Naci, L., Schiff, N. D., Owen, A. M., & Soddu, A. (2023). An implementation of integrated information theory in resting-state fMRI. Communications Biology. https://doi.org/10.1038/s42003-023-05063-y

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.