REVIEW 5 major objections 5 minor 10 references
Dynamic EEG-fMRI mapping: Revealing the relationship between brain connectivity and cognitive state
T0 review · 5 major / 5 minor · reviewed 2026-08-12 · deepseek-v4-flash
Pith's one-line read The paper claims that simultaneous EEG-fMRI recordings, analyzed in 40-second sliding windows and clustered by modularity, reveal distinct connectivity states dominated by sensory systems and the default mode network, supporting the…
desk verdict The methods text merges two incompatible experiments and the Results report no statistics, so the central claims aren't verifiable as written. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The load-bearing object is the EEG-fMRI connectivity graph: an $84 \times 84$ Pearson correlation matrix ($30$ EEG electrodes plus $54$ fMRI independent-component networks) computed per EEG frequency band, after convolving the EEG power time series with a standard hemodynamic response function. Dynamic analysis slides a $20$-TR ($40$ s) window across the data to produce $237$ time-varying graphs, from which node-level connection strength, clustering coefficient, and global efficiency are computed. A second correlation matrix over time windows is then clustered by a modularity algorithm, and each resulting module of time windows is averaged into a single representative connectivity state.
What would settle it
Recompute the dynamic connectivity states after replacing the fMRI component time courses with phase-randomized surrogate signals that preserve each time course's autocorrelation; if the same modular states in sensory and default-mode networks emerge from the surrogates, the claimed states are a statistical artifact rather than neural connectivity. A second check: repeat the ICN selection from the same 100 ICA components with an independent rater or an automated classifier and test whether the modular organization and the 30-60 second state separation survive.
Extended reading notes
Core claim
The central claim is that EEG spectral power and fMRI intrinsic-connectivity-network time courses, when convolved with a hemodynamic response function and correlated in a sliding 40-second window, produce brain graphs whose modular structure is stable enough to define distinct connectivity states. Across eyes-open and eyes-closed resting conditions, the detected modules are dominated by sensory (visual, auditory, sensorimotor) systems and the default mode network, with anti-correlations between these broad regions. The paper further claims that these states align with earlier findings that cognitive states can be identified from short-duration data, specifically windows of 30-60 seconds.
Load-bearing premise
The whole analysis depends on treating Pearson correlations between EEG spectral power (after hemodynamic convolution) and fMRI component time courses as genuine neural connectivity, and on the manual selection of 54 of 100 ICA components as an unbiased set of intrinsic connectivity networks; if those correlations or that selection are dominated by artifacts, the reported modular states would not reflect brain organization.
Editorial extensions
If this is right
- Cognitive-state classification can be performed on roughly 40-second multimodal recordings, making EEG-fMRI connectivity analysis feasible for settings where long scans are impractical.
- The modular separation between sensory systems and the default mode network appears in both static and dynamic EEG-fMRI graphs, suggesting a stable cross-modal backbone of brain organization.
- Sliding-window graph metrics fluctuate in a low-frequency range, so dynamic connectivity changes are slow enough to be captured at this temporal resolution.
- Because connectivity states are defined without a task label, the same pipeline could be applied to naturalistic or clinical recordings to discover state changes from data alone.
- Anti-correlations between sensory and default-mode modules are preserved in the negative-correlation graph, indicating that the sign of EEG-fMRI coupling carries information about state.
Reading between the lines
- Editorial inference: the 40-second window choice is justified by prior reports, but the paper does not directly compare window lengths; a natural extension would be to test windows from 10 to 60 seconds on the same data to see whether the 30-60 second claim is a plateau or a sharp optimum.
- Editorial inference: since the pipeline operates on unlabeled eyes-open/eyes-closed data, applying it to task or clinical data with ground-truth state labels would test whether discovered states correspond to behavior, not just to recording condition.
- Editorial inference: the reliance on a 1.5 T scanner and 30 scalp electrodes suggests the approach may be reproducible in lower-cost settings, which would be relevant for EEG-based neurofeedback if the fMRI nodes can be replaced by a template atlas.
- Editorial inference: the paper's evidence for ICNs comes from group-level ICA, so individual-level state detection would require per-subject components; verifying that the modular states appear in single subjects would be a direct test of practical use.
Signed reviews
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The manuscript reports an EEG-fMRI connectivity study that constructs static and dynamic graph-theoretic brain networks from 30 EEG channels and 54 fMRI ICA components across five frequency bands in eyes-open and eyes-closed conditions. It claims to reveal modular organization in sensory and default-mode networks and to show that cognitive states can be identified from 30-60 s windows of data. The methods describe sliding-window correlation analysis, graph metrics, and a modularity-based state-detection procedure, but the results section contains only figure descriptions and no statistical values.
Significance. If the reported findings were supported, the proposed multimodal graph framework could be a useful contribution to EEG-fMRI connectivity research, particularly the idea of using short sliding windows to track cognitive-state-related network reconfiguration. However, the manuscript as written does not substantiate these claims: no test statistics, p-values, effect sizes, or variance measures are reported, and the central dataset is not unambiguously identified. The temporal claim about 30-60 s windows rests on a choice made from the same assumption, making it circular. No code or data are provided to support reproducibility. The significance of the work is therefore currently unverifiable.
major comments (5)
- [II.B, II.E, II.F] The manuscript does not identify which dataset is used for the connectivity analysis. Section II.B describes only a resting-state design with 8.5 minutes of eyes-closed followed by 8.5 minutes of eyes-open, while Section II.E abruptly introduces three 320-second motor-imagery neurofeedback sessions per participant with alternating 20-second rest and motor blocks. Section II.F then refers to a matrix 'EFEFEF' of dimensions 256×84 without defining this matrix or stating whether it comes from the resting sessions or the neurofeedback sessions. This ambiguity is load-bearing because the abstract and conclusions make claims about cognitive and visual states, and the reader cannot determine whether the reported graphs and states were computed from resting data, motor-imagery data, or both.
- [III.A, II.F] No inferential statistics are reported anywhere in the results. Section II.F states that a 5×2 repeated-measures ANOVA and paired t-tests were conducted on the static and dynamic measurements, but Section III reports no F values, p-values, effect sizes, confidence intervals, or measures of variability. Claims such as 'modular organization in sensory systems and default mode regions' and 'structural differences across five frequency bands' are asserted based on visual inspection of Figure 4B. These are central empirical claims of the paper, and their absence of statistical support means the findings cannot be evaluated.
- [II.F, IV] The central temporal claim is circular. The sliding-window length is set to 20 TRs (40 seconds) in Section II.F based on prior research 'indicating that cognitive states can be accurately identified with data from short periods of 30 to 60 seconds.' The conclusion then presents as a finding that 'cognitive states can be effectively identified through short-duration data, specifically within the 30-60 second timeframe.' Because the window length was chosen from the very assumption the paper claims to demonstrate, the result does not provide independent evidence for the 30-60 second claim.
- [II.H, III] The connectivity-state analysis described in Section II.H is not reported in the results. The methods promise detection of distinct connectivity states via modularity of a time-window correlation matrix, but the results section does not state the number of states found, their module compositions, their temporal durations, or any validation against null models. The abstract's assertion of 'distinct connectivity states' is therefore unsupported by any presented quantitative output.
- [II.F] Key computational details are missing, preventing reproducibility. The matrix 'EFEFEF' is never defined; the construction of the EEG spectral power time series is not specified (e.g., how band power is computed, what window is used, and how the 5 kHz EEG data are matched to the 2-second TR timescale); the HRF convolution parameters are not given; and the modularity algorithm used in Section II.H is not named or parameterized. Without these details, even the described pipeline cannot be reimplemented or checked.
minor comments (5)
- [II.A] The section heading 'Participator' contains a typo; it should read 'Participants.'
- [II.F] Before equation (1), the text contains 'RRR' where the correlation matrix R is presumably meant, and the displayed equations are not rendered correctly, making it difficult to read the definitions of positive and negative weighted connections.
- [III.A] The text says the static maps were computed 'for more than 25 participants,' but Section II.A reports exactly 25 participants. This numerical inconsistency should be corrected.
- [II.E, Figure 2] Figure 2 is described as the 'dual-modal neurofeedback metaphor presented during the conference,' but it is never integrated into the analysis, and the text does not explain how the 1D versus 2D display conditions relate to the connectivity results.
- [References] Several references, such as [3], [4], [6], and [8], appear to be about layer-specific fMRI or unrelated topics and are not clearly cited in the relevant parts of the text; the reference list should be aligned with the actual citations.
Circularity Check
The 30-60 s cognitive-state finding restates the window-length premise; the modular-organization result remains independent.
-
fitted input called prediction
[Section II.F (window-length justification); Abstract and Section IV (30-60 s finding)]
"We utilized a window width of 20TRs (40 seconds) based on research indicating that cognitive states can be accurately identified with data from short periods of 30 to 60 seconds. ... Our previous work has demonstrated that shorter time windows reduce the number of statistically significant correlations ... Abstract: 'our findings align with previous literature, reinforcing the notion that cognitive states can be effectively identified through short-duration data, specifically within the 30-60 second timeframe.'"
The sliding-window width (40 s) is set because the authors assume cognitive states are identifiable from 30-60 s data, and the paper's abstract and conclusion then report, as a finding, that cognitive states are effectively identified within the 30-60 s timeframe. The claimed result is the same proposition used to choose the window parameter; no alternative window lengths are tested, so the data cannot confirm the 30-60 s range independent of the design choice. The adjacent self-citation ('Our previous work has demonstrated...') is part of this same justification chain, so the conclusion restates the premise rather than deriving it from the present data.
full rationale
The derived modular-organization claim (sensory and default-mode ICNs, anti-correlations between them) is not circular: it follows from group ICA, HRF-convolved EEG power correlation with ICN time courses, thresholded graph construction, and community detection — standard tools with no parameter fitted to the claimed modularity outcome. The single genuinely circular element is the 30-60 s cognitive-state claim, which is the input assumption behind the 40 s (20 TR) window choice and is then reported as an aligned finding in the Abstract and Conclusion; that specific prediction reduces by construction to the window-selection premise and is additionally supported by an in-text self-citation of the authors' prior work. Other concerns — the undefined 'EFEFEF' matrix, the merged resting-state and motor-imagery neurofeedback paradigms, and the absence of statistics in the Results — are correctness or coherence risks, not circularity, and are therefore excluded from the score. On balance, the central modularity result has independent content, so the paper is only partially circular.
Assumptions & free parameters
free parameters (5)
- Sliding window length =
20 TRs (40 seconds), with 22 TRs mentioned as alternative
- Number of ICA components =
100
- Number of retained ICNs =
54
- PCA component retention =
120
- EEG frequency bands =
delta, theta, alpha, beta, low gamma
assumptions (5)
- domain assumption Pearson correlation between EEG spectral power and fMRI time courses reflects meaningful brain connectivity.
- domain assumption The 54 retained ICs are intrinsic connectivity networks that are stable across participants and conditions.
- domain assumption Sliding-window correlations capture time-varying functional connectivity.
- domain assumption Cognitive states can be identified from 30-60 seconds of data.
- domain assumption Standard graph metrics are meaningful for weighted signed connectivity matrices with negative correlations.
Cite this review
Pith. "Pith review of Dynamic EEG-fMRI mapping: Revealing the relationship between brain connectivity and cognitive state." pith.science (2026). https://pith.science/paper/FO2OCKST
@misc{pith2026241119922,
author = {Pith},
title = {Pith review of: Dynamic EEG-fMRI mapping: Revealing the relationship between brain connectivity and cognitive state},
year = {2026},
howpublished = {\url{https://pith.science/paper/FO2OCKST}},
note = {Machine review of arXiv:2411.19922}
}
read the original abstract
This study investigated the dynamic connectivity patterns between EEG and fMRI modalities, contributing to our understanding of brain network interactions. By employing a comprehensive approach that integrated static and dynamic analyses of EEG-fMRI data, we were able to uncover distinct connectivity states and characterize their temporal fluctuations. The results revealed modular organization within the intrinsic connectivity networks (ICNs) of the brain, highlighting the significant roles of sensory systems and the default mode network. The use of a sliding window technique allowed us to assess how functional connectivity varies over time, further elucidating the transient nature of brain connectivity. Additionally, our findings align with previous literature, reinforcing the notion that cognitive states can be effectively identified through short-duration data, specifically within the 30-60 second timeframe. The established relationships between connectivity strength and cognitive processes, particularly during different visual states, underscore the relevance of our approach for future research into brain dynamics. Overall, this study not only enhances our understanding of the interplay between EEG and fMRI signals but also paves the way for further exploration into the neural correlates of cognitive functions and their implications in clinical settings. Future research should focus on refining these methodologies and exploring their applications in various cognitive and clinical contexts.
Reference graph
Works this paper leans on
-
[5]
Woolrich,M.W.,Jbabdi,S.,&Behrens,T.E.(2009).Bayesiananalysis of neuroimaging data in FSL. NeuroImage, 45(S1): S173-S186. https://doi.org/10.1016/j.neuroimage.2008.10.055
-
[1]
Zotev, V., Mayeli, A., Misaki, M., & Bodurka, J. (2020). Emotion self-regulation training in major depressive disorder using simultaneous real-time fMRI and EEG neurofeedback. NeuroImage: Clinical, 27: 102331.https://doi.org/10.1016/j.nicl.2020.102331
arXiv 2020
- [2]
-
[3]
Deshpande, V., Mehta, R., Dhawan, A., & Singh, K. (2022). Layer-specific BOLD responses reveal distinct functional connectivity patterns in the visual cortex. Journal of Neuroscience, 42(23): 4527–4540.https://doi.org/10.1523/JNEUROSCI.0178-22.2022
-
[4]
Therole ofcorticaldepth in modulating brain responses: Insights from layer-specific fMRI
Koiso, T., Fujimoto, K., &Amano, K.(2022). Therole ofcorticaldepth in modulating brain responses: Insights from layer-specific fMRI. Frontiers in Neuroscience, 16: 893245. https://doi.org/10.3389/fnins.2022.893245
- [6]
-
[7]
Zotev, V., Phillips, R., Misaki, M., & Wong, C. K. (2018). Real-time fMRI neurofeedback training of the mediodorsal and anterior thalamus enhances correlation between thalamic BOLD activity and alpha EEG rhythm. Human Brain Mapping, 39(3): 1024–1042. https://doi.org/10.1002/hbm.23902
-
[8]
Stanley, J. A., Olivito, G., & Walsh, R. (2021). Cortical depth-specific mapping of neural activity using phase regression modeling in fMRI. Frontiers in Systems Neuroscience, 15: 774311. https://doi.org/10.3389/fnsys.2021.774311
Show all 10 references
-
[9]
Zich, C., Debener, S., & Kranczioch, C. (2015). Real-time EEG feedback during simultaneous EEG-fMRI identifies the cortical signature of motor imagery. NeuroImage, 114: 438–447. https://doi.org/10.1016/j.neuroimage.2015.04.020
2015 doi
-
[10]
Y., Feng, Z
Xu, Y. Y., Feng, Z. Q., & Xie, Y. J. (2018). Frontal alpha EEG asymmetry before and after positive psychological interventions for medical students. Frontiers in Psychiatry, 9: 432. https://doi.org/10.3389/fpsyt.2018.00432
2018
Reviewed August 12, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.