Pith. sign in

REVIEW 4 major objections 5 minor 8 references

Complexity synchronization analysis of neurophysiological data: Theory and methods

T0 review · 4 major / 5 minor · reviewed 2026-08-12 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read The paper claims that a standardized diffusion-entropy method can detect synchronized complexity across EEG, ECG, and respiration during cognitive tasks.

desk verdict Useful automation of MDEA parameter selection with honest limitations, but the central empirical claim about brain-heart-lung synchronization is not yet supported without null-surrogate controls and corrected statistics. read the letter →

arxiv 2411.14602 v1 pith:S6RVAHJM submitted 2024-11-21 q-bio.NC nlin.AO

Ioannis Schizas , Sabrina Sullivan , Scott E. Kerick , Korosh Mahmoodi , J. Cortney Bradford , David L. Boothe , Piotr J. Franaszczuk , Paolo Grigolini
show 1 more author
Bruce J. West
This is my paper · ORCID
classification q-bio.NCnlin.AO
keywords complexitysynchronizationmodifieddiffusionentropyanalysismultifractalitycrucialeventsinversepowerlawEEGECGrespiration
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper tries to establish that complexity synchronization — simultaneous, time-varying changes in the multifractal scaling of different organ signals — is a real and measurable property of brain, heart, and lung time series recorded during cognitive task performance. The vehicle is modified diffusion entropy analysis (MDEA), which converts each signal into a sequence of crucial events (amplitude-threshold crossings) and reads off the slope δ of the diffusion entropy versus log-window plot, a scaling index tied to the inverse power-law exponent of the waiting times between events. The authors add two automatic parameter choices: a Kolmogorov-Smirnov statistic that selects the stripe size and power-law exponent by fitting the event inter-times to an inverse power law, and automatic detection of the linear fit region. Across 54 datasets from a Go-NoGo study, the method finds significant EEG–ECG synchronization in 46 of 54 datasets and significant three-way synchronization among EEG, ECG, and high-pass-filtered respiration in 24 of 54 datasets. A sympathetic reader would care because, if the claim holds, it gives a reproducible, data-driven way to measure coordination among heterogeneous physiological signals without frequency-band decompositions or trial averaging.

What carries the argument

The load-bearing machinery is modified diffusion entropy analysis (MDEA). MDEA coarse-grains a time series into stripes by amplitude, marks crucial events when the signal crosses stripe boundaries, builds a binary event series, and accumulates it into a diffusion trajectory; the Shannon-Wiener entropy of segment endpoints, plotted against log window length, has slope δ, related to the inverse power-law exponent µ of the waiting-time distribution by µ = 1 + 1/δ. The paper's main technical additions are (1) a Kolmogorov-Smirnov statistic that scores how well the crucial-event inter-times fit an IPL complementary cumulative distribution, minimized over stripe size and µ by grid search, and (2) an automated change-point-based selection of the longest linear region of the entropy plot, with a special middle-segment rule for ECG. Complexity synchronization is then defined by correlations over sliding windows of the δ values across signals; the resulting time-varying δ tracks are the objects whose co-variation is tested.

What would settle it

Run the paper's order-8192 2-Hz high-pass filter on a synthetic signal built from a known deterministic oscillator plus a known inverse-power-law component, then apply the KS-stripe-size MDEA pipeline and check whether the recovered δ matches the injected IPL exponent; additionally, phase-shuffle the filtered RESP signal and recompute the EEG–RESP and ECG–RESP correlations, which should vanish if the synchronization reflects genuine coupling.

Watch

Extended reading notes

Core claim

The paper's central claim is that complexity synchronization (CS) — high-order co-variation of the inverse power-law scaling index δ extracted by modified diffusion entropy analysis (MDEA) — is present among brain, heart, and lung organ-network time series during cognitive task performance, and that this can be detected repeatably once two MDEA parameters are chosen automatically. The authors first validate a KS-based estimator of stripe size and of the IPL index µ on synthetic duration-time data with known power-law exponents, showing lower bias and variance as sample size grows. Applied to 54 datasets (27 subjects × 2 stress conditions), the pipeline finds significant EEG–ECG synchronization in 46 of 54 datasets, and significant three-way synchronization among EEG, ECG, and a 2-Hz high-pass-filtered RESP signal in 24 of 54 datasets. A central caveat, stated by the authors, is that the unfiltered RESP signal did not synchronize with EEG and ECG, and meaningful lung synchronization appeared only after the dominant periodic component of respiration was removed, a preprocessing decision whose interpretation they flag as an open question.

Load-bearing premise

The load-bearing premise is that the 2-Hz high-pass filter applied to the respiration signal removes a deterministic oscillatory component while leaving an inverse-power-law complexity component intact; if the filter instead creates or distorts the IPL statistics, the reported EEG–RESP and ECG–RESP synchronization is an artifact of preprocessing.

Editorial extensions

If this is right

  • With automated parameter selection, MDEA/CS can be run on large-scale datasets: the authors estimate stripe sizes across 190,938 moving windows, and provide code and sample data for independent replication.
  • The empirical pattern of correlations, with stronger EEG–ECG than EEG–RESP or ECG–RESP coupling, implies that brain–heart complexity coupling is the most consistent organ-network link during cognitive tasks.
  • For periodic physiological signals such as respiration, CS analysis requires removing the dominant deterministic oscillation; otherwise the scaling index is not informative about complexity synchronization.
  • Tracking δ over time in sliding windows yields a real-time, single-trial measure of organ-network coordination, applicable to continuous recordings without trial averaging.
  • Cross-modal synchronization of scaling indices, if replicated, offers a candidate signature of network-level information transfer distinct from linear correlation, coherence, or phase coupling.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • An explicit test the authors did not run: apply the same order-8192 2-Hz high-pass filter to synthetic periodic-plus-power-law signals to confirm that the filter preserves the true IPL exponent; until then, the RESP results rest on preprocessing.
  • The correlation between EEG and ECG scaling indices could partly be driven by a shared global state such as arousal, movement, or task engagement rather than organ-network information flow; window-shuffled surrogates or partial correlations controlling for a common driver would discriminate.
  • If the CS measure generalizes, it could be applied to other multimodal recordings, such as fMRI with peripheral physiology or hyperscanning between individuals, wherever signals are long enough for reliable entropy-slope estimation; the paper's own claims are limited to the neurophysiological data analyzed.
  • The discussion's 'information force' interpretation goes beyond the data, which establish co-variation of scaling indices but not direction or mechanism of information transfer.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The paper presents an automated pipeline for modified diffusion entropy analysis (MDEA) of neurophysiological time series, with the goal of estimating multifractal scaling indices δ and detecting complexity synchronization (CS) across EEG, ECG, and respiration. Two parameter-selection problems are addressed: automated stripe-size selection via a Kolmogorov–Smirnov (KS) fit to an inverse power law (IPL) complementary cumulative distribution function (Eq. 5), and automated linear-fit-region selection using Matlab's ischange() function. The method is validated on Mittag-Leffler synthetic data and then applied to 54 datasets (27 subjects under low and high time stress) from a Go-NoGo neurofeedback study. The empirical results indicate significant EEG-ECG CS in 46/54 datasets and significant three-way CS with high-pass filtered RESP in 24/54 datasets. The authors provide open-source code and a sample dataset, and they candidly discuss limitations, including the post-hoc decision to filter RESP and unresolved interpretability questions.

Significance. If the central claim is upheld, the paper would make a useful methodological contribution to network physiology by offering a repeatable, automated procedure for measuring complexity synchronization across heterogeneous organ time series. The open-source implementation, Monte Carlo bias/variance evaluation, and honest reporting of limitations are clear strengths. However, the main empirical headline—that brain, heart, and lung complexity 'significantly co-vary'—is only supported in 24 of 54 datasets, and the pipeline's specificity is not established: there are no negative controls or goodness-of-fit rejection criteria, and the post-hoc RESP filter is a potential source of artifact. These issues do not invalidate the method but they do block acceptance as a validated methods paper.

major comments (4)
  1. [Results, after Table 4] The abstract states that 'the complexity of brain, heart, and lung ONTS significantly co-vary over time during cognitive task performance,' but the three-way CS (EEG-ECG-RESP) was significant in only 24 of 54 datasets, with 30 datasets lacking significance. Under a per-dataset threshold of p<.05, about 2.7 false-positive datasets are expected by chance, so the aggregate result is above chance, but the headline overstates the consistency of the effect. The authors should report effect sizes and confidence intervals for the correlations and ideally perform a multilevel analysis with subjects and conditions as random effects, rather than relying on a count of 'significant' datasets.
  2. [Section 2.3.1, following Figure 7] The decision to high-pass filter RESP at 2 Hz with a Kaiser filter of order 8192 was made only after finding that unfiltered RESP did not synchronize with EEG and ECG. The authors themselves raise the interpretability concern ('what does it mean to remove the dominant feature of the RESP time series?'). The paper provides no control analysis to show that this filter preserves an IPL complexity component rather than creating an IPL-like artifact. I request a surrogate validation: apply the identical filter, stripe-size selection, and MDEA pipeline to phase-randomized RESP, to simulated mixtures of periodic and IPL components, and to pure noise, and show that the resulting EEG-RESP and ECG-RESP correlations are not reproduced without genuine coupling.
  3. [Section 2.3.1, Eqs. (3)-(5)] The KS-based stripe-size and μ selection minimizes the KS statistic against an IPL CCDF with two free parameters, and MDEA subsequently estimates δ which is interpreted through the same IPL relation μ = 1 + 1/δ. This couples the measurement to the assumed model: for almost any signal, the grid search will find a stripe size that brings the inter-event intervals close to some IPL curve, unless a goodness-of-fit threshold is enforced. The paper demonstrates positive controls (Mittag-Leffler, Figs. 4-5) but no negative controls. I ask the authors to report the KS statistics and the proportion of windows that would fail an IPL goodness-of-fit test, and to apply the pipeline to non-IPL signals (e.g., AR(1), sinusoids, white noise) to demonstrate that spurious scaling indices and spurious CS are not produced.
  4. [Section 2.3.1 and Results] The use of 30-s sliding windows with 20-s overlap creates strong serial dependence in the δ time series, so the effective sample size for the correlation tests in Tables 3-4 is much smaller than the nominal number of windows. The reported p-values do not appear to account for this dependence and are therefore anti-conservative. The authors should either use non-overlapping windows or correct for autocorrelation, for example with block bootstrap or an effective-degrees-of-freedom adjustment, before claiming statistical significance for the CS correlations.
minor comments (5)
  1. [Section 2.3.1, Eq. (4)] There is a typo in the text before Eq. (4): 'pr oviding' should be 'providing', and the notation is inconsistent because the definition of F_emp uses the indicator 1_{τi ≤ T} while the text describes the probability that 'the inter-arrival time variable τ < T'; please clarify the roles of τ and T.
  2. [References] The reference to Corder and Foreman (2014) lists 'The city' as the publisher location; this placeholder should be replaced with the actual location.
  3. [Figure 4] The caption says 'mean (blue) and variance (red)', but the text in Section 2.3.1 refers to bias and variance; please clarify which curve is bias and which is variance, and add axis labels to the figure.
  4. [Supplementary Materials] The supplementary tables are numbered Table 1-6, which duplicates the numbering of the main-text tables; please label them 'Supplementary Table S1' etc. to avoid confusion.
  5. [Section 2.3.2] The description of the adaptive adjustment of the ischange threshold ('adjusted in small increments') does not specify the increment size or the stopping criterion, which reduces reproducibility; please provide the exact algorithm or pseudocode for this procedure.

Circularity Check

2 steps flagged · score 6.0 of 10

The MDEA pipeline's stripe-size selection enforces the IPL model it later uses to interpret δ, and the RESP filter was chosen after observing the synchronization it then reports; the CS findings are therefore partially constructed by these choices.

  1. self definitional [Section 2.3.1, Eqs. (1)-(5) and Section 2.1]
    "we introduce an automated stripe-size selection method building on the property that CE time-intervals should follow an IPL PDF according to the physical model utilized by Mahmoodi et al. (2023a). ... (μ̂, ŝℓ) = arg min(μ,sℓ) D_{Nsamp}(μ, sℓ) ... If the waiting time distribution PDF of the τs has an inverse power law (IPL) form of ψ(τ) ∝ 1/τµ, the system has temporal complexity scaled with µ which is related to scaling δ by µ = 1 + 1/δ."

    The stripe size is selected by minimizing the KS distance between the empirical inter-event CCDF and a theoretical IPL CCDF, so the event stream fed into MDEA is forced to resemble the assumed IPL model. The resulting entropy-slope δ is then interpreted as a complexity measure through the very same IPL relation μ = 1 + 1/δ. There is no goodness-of-fit threshold: the grid search always returns a minimizing pair (μ, sℓ), so any signal, even one without genuine IPL scaling, yields an IPL-compatible δ. Thus the model is both the input that shapes the measurement and the output used to interpret it, making the 'complexity' index partly self-definitional.

  2. fitted input called prediction [Section 2.3.1, text following Fig. 7; Results, Section 3]
    "However, in subsequent analyses while testing the generalizability of these findings to other subjects in the experiment, we observed that the scaling of the RESP time series does not systematically synchronize with the EEG and ECG scaling when stripe size and linear fit parameters are chosen algorithmically as described above. ... Consequently, we high-pass filtered the RESP signal at 2 Hz ... This approach revealed a significant improvement in synchronization of RESP with EEG and ECG as can be seen in Fig. 10 (see Table 4)."

    The 2 Hz high-pass filter was introduced only after the authors observed that unfiltered RESP did not synchronize with EEG and ECG under the automated parameter selection. The filter cutoff was therefore chosen with knowledge of the outcome it was meant to produce, and the paper then reports the post-filter synchronization (24/54 datasets for three-way CS) as an empirical finding rather than as a consequence of the preprocessing decision. Because the preprocessing was selected on the basis of the result it later 'predicts,' the EEG-RESP and ECG-RESP complexity synchronization is partly constructed by that choice rather than independently measured. The authors themselves flag interpretability concerns, but the reported headline result still rests on this outcome-dependent filter choice.

full rationale

The paper's central claim is that MDEA-based complexity synchronization detects significant co-variation of scaling indices δ across EEG, ECG, and high-pass-filtered RESP. This claim is not a closed-form derivation, but it depends on two data-driven choices that make the result partially circular. First, the automated stripe-size selection minimizes the KS statistic against an IPL CCDF (Eq. 5), and the resulting δ is interpreted through the IPL relation μ = 1 + 1/δ (Section 2.1). This means the measurement pipeline is built on the very IPL assumption it is used to support; without a goodness-of-fit threshold or rejection of non-IPL signals, the δ series is produced for any input. Second, the RESP signal was high-pass filtered at 2 Hz only after the unfiltered RESP failed to synchronize with EEG/ECG, and the post-filter synchronization is then reported as the main three-way finding. This is a post-hoc (outcome-dependent) preprocessing choice, not an independent prediction. The paper's synthetic validation uses Mittag-Leffler data (a positive control) but provides no negative controls showing that uncoupled or non-IPL signals yield low CS, so the specificity of the pipeline is unestablished. The EEG-ECG coupling (46/54 datasets) is robust to the RESP filter and gives the work independent empirical content, and the self-citations to prior SOTC work are hypotheses rather than a uniqueness argument. However, the headline three-way CS result is partly built from the IPL-enforced estimator and the result-informed filter, warranting a partial-circularity score of 6 rather than a higher score.

Assumptions & free parameters 6 free parameters · 6 assumptions · 0 invented entities

All free parameters are empirically fitted or chosen; the IPL and filtering assumptions carry the main burden of the analysis. No new physical entities are introduced; 'complexity synchronization' and 'SOTC' are concepts from prior work by the same group.

free parameters (6)
  • stripe size s_l = per-window estimates; median per subject (e.g., EEG 10.3, ECG 33.3, RESP 500.0, RESP filtered 146.6)
    Coarse-graining threshold for event detection; chosen by minimizing KS statistic to an IPL CCDF.
  • IPL index μ in KS fit = range 1-3, selected via grid search
    Fitted together with stripe size to match empirical CCDF; later connected to MDEA δ via μ=1+1/δ.
  • linear fit region endpoints = per-subject median start and end indices from ischange()
    Determines δ from the slope of S(w) vs log(w); selected using an adaptive threshold.
  • respiratory high-pass filter cutoff = 2 Hz, Kaiser filter order 8192
    Post-hoc choice applied after unfiltered RESP failed to show CS; central to the reported three-way CS.
  • window length and overlap = 30 s windows, 20 s overlap
    Chosen analysis parameters that affect statistical dependence and effective sample size.
  • ischange threshold adjustment increment = not specified exactly
    Controls sensitivity of change-point detection for fit regions; described as small increments but not quantified.
assumptions (6)
  • domain assumption Crucial event waiting times follow an IPL distribution ψ(τ) ∝ τ^{-μ} with 1 ≤ μ ≤ 3.
    Invoked in Eq. 1 and used by the KS stripe-size selection; if false, MDEA's δ does not measure temporal complexity.
  • standard math MDEA scaling δ and waiting-time IPL index satisfy μ = 1 + 1/δ.
    Standard result from diffusion entropy analysis literature (Scafetta and Grigolini 2002), used to convert δ to μ.
  • ad hoc to paper The stripe size minimizing the KS statistic against an IPL CCDF yields the correct stripe size for the underlying SOTC process.
    Core assumption of the new automated selection method; not proven, only demonstrated on Mittag-Leffler synthetic data.
  • domain assumption The linear region detected by Matlab ischange() corresponds to the scaling regime of the diffusion entropy.
    Used in Section 2.3.2 to choose fit intervals for δ; assumes change points mark the valid scaling range.
  • ad hoc to paper High-pass filtering RESP at 2 Hz removes only non-IPL deterministic oscillation and preserves complexity-carrying fluctuations.
    Load-bearing for the reported RESP synchronization; the authors acknowledge this is not yet theoretically justified.
  • domain assumption δ estimates from overlapping 30 s windows with 20 s overlap can be treated as independent for correlation significance testing.
    Needed for the p-values in Tables 3 and 4; violated in practice because windows share 20 s of data, inflating effective sample size.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Complexity synchronization analysis of neurophysiological data: Theory and methods." pith.science (2026). https://pith.science/paper/S6RVAHJM

@misc{pith2026241114602,
  author       = {Pith},
  title        = {Pith review of: Complexity synchronization analysis of neurophysiological data: Theory and methods},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/S6RVAHJM}},
  note         = {Machine review of arXiv:2411.14602}
}
read the original abstract

We apply modified diffusion entropy analysis (MDEA) to assess multifractal dimensions of ON time series (ONTS) and complexity synchronization (CS) analysis to infer information transfer among ONs that are part of a network of organ networks (NoONs). The purpose of this paper is to advance the validation, standardization, and repeatability of MDEA and CS analysis of heterogeneous neurophysiological time series data. Results from processing these datasets show that the complexity of brain, heart, and lung ONTS significantly co-vary over time during cognitive task performance but that certain principles, guidelines, and strategies for the application of MDEA analysis need consideration.

Figures

Figures reproduced from arXiv: 2411.14602 by the authors.

Figure 2
Figure 2. Amplitudes and temporal dynamics of EEG, ECG, and RESP within one breath cycle (4 s), showing the need of having a method to capture the different relevant features of each time series. In MDEA analysis, amplitudes of all time series are normalized to the interval [0, 1], but the original amplitude and frequency scales are shown here to illustrate the different orders of magnitude scales of EEG with respect to ECG a… view at source ↗
Figure 3
Figure 3. The impact of stripe size selection in the CE waiting-time PDF. Left column shows the no-IPL shape of the empirical waiting times PDF (blue curve) when the stripe size is not properly selected. Clearly, the blue curve corresponding to Femp(τ) [cf. Eq. 2] does not fit any of the theoretical IPL PDFs indicated by the red curves for values of µ starting at µ = 1 and increasing by 0.4 as the red curves approach the hori… view at source ↗
Figure 4
Figure 4. Bias and variance of KS-based and MDEA-based estimation of IPL parameter µ versus number of samples Nsamp utilized for the estimation. size estimate is used as input. However, this is not the case for MDEA when the stripe size is not correctly selected, in which case both the estimated bias and variance deviate as Nsamp increases [PITH_FULL_IMAGE:figures/full_fig_p010_4.png] view at source ↗
Figures from the paper (3 more)
Figure 6
Figure 6. Figure 6: Distributions of stripe sizes determined across 30 sec moving windows for EEG, ECG, RESP, and filtered RESP across 27 subjects in low and high time stress conditions. 2.3.2. Automated fit parameter region for estimating δ - scaling To address the issue of automatically…
Figure 7
Figure 7. Figure 7: RESP time series zoomed to 20 sec (upper) and spectra (lower) unfiltered (left) and high-pass filtered (right) over entire 675 sec task period. 8 (top right) S (w) vs. log(w) plot for ECG reveals three clear segments, contrasting with the relatively linear EEG patterns…
Figure 8
Figure 8. Figure 8: Entropy (S (w)) vs. log window length (log(w)) over sliding windows for EEG, ECG, and unfiltered and filtered RESP. Red lines demarcate the median fit regions applied for estimating delta scaling indices [PITH_FULL_IMAGE:figures/full_fig_p014_8.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

8 extracted references · 8 canonical work pages

  1. [5]

    Neuroimage 85, 853–872

    Phase transfer entropy: a novel phase -based measure for directed connectivity in networks coupled by oscillatory interactions. Neuroimage 85, 853–872. Mahmoodi, K., Kerick, S.E., Grigolini, P., Franaszczuk, P.J., West, B.J., 2023a. Complexity synchronization: a measure of interaction between the brain, heart and lungs. Scientific Reports 13, 11433. Mahmo...

    work page 2023

  2. [7]

    Shock and Vibration 2016, 5836717

    Fault diagnosis system of induction motors based on multiscale entropy and support vector machine with mutual information algorithm. Shock and Vibration 2016, 5836717. Pernice, R., Zanetti, M., Nollo, G., De Cecco, M., Busacca, A., Faes, L.,

    work page 2016

  3. [2014]

    Coexisting forms of coupling and phase-transitions in physiological networks, in: Nonlinear Dynamics of Electronic Systems: 22nd International Conference, NDES 2014, Albena, Bulgaria, July 4-6,

    work page 2014

  4. [2015]

    Non-linear analysis of eeg and hrv signals during sleep, in: 2015 37th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC), IEEE. pp. 4174–4177. Marwan, N., Romano, M.C., Thiel, M., Kurths, J.,

    work page 2015

  5. [2016]

    Journal of Computational and Applied Mathematics 296, 181–211

    On mittag-leffler distributions and related stochastic processes. Journal of Computational and Applied Mathematics 296, 181–211. doi:https://doi.org/10.1016/j.cam.2015.09.031. Ivanov, P.C.,

    work page doi:10.1016/j.cam.2015.09.031 2015

  6. [2019]

    Mutual information analysis of brain -body interactions during different levels of mental stress, in: 2019 41st Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC), IEEE. pp. 6176–6179. 1 Petzschner, F.H., Weber, L.A., Wellstein, K.V., Paolini, G., Do, C.T., Stephan, K.E.,

    work page 2019

  7. [2021]

    Mutual information analysis of brain-heart interactions in epileptic children, in: 2021 Signal Processing Symposium (SPSympo), IEEE. pp. 133–137. Kumar, M., Singh, D., Deepak, K.,

    work page 2021

  8. [2023]

    Eeg-based emotion recognition using sub-band time-delay correlations, in: 2023 45th Annual International Conference of the IEEE Engineering in Medicine & Biology Society (EMBC), IEEE. pp. 1–4. Bartsch, R.P., Ivanov, P.C.,

    work page 2023

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.