Pith. sign in

REVIEW 4 major objections 6 minor 1 cited by

Contextual Minimum-Norm Estimates (CMNE): A Deep Learning Method for Source Estimation in Neuronal Networks

T0 review · 4 major / 6 minor · reviewed 2026-08-14 · deepseek-v4-flash

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

Pith's one-line read LSTM-learned temporal context sharpens MEG/EEG source estimates beyond dSPM.

desk verdict A fresh but unproven combination of LSTM prediction and dSPM correction; the Bayesian framing doesn't hold, but the core idea deserves a serious referee. read the letter →

arxiv 1909.02636 v1 pith:VTPI7LLQ submitted 2019-09-05 q-bio.QM cs.LGstat.ML

Christoph Dinh , John GW Samuelsson , Alexander Hunold , Matti S Hämäläinen , Sheraz Khan This is my paper · ORCID
classification q-bio.QMcs.LGstat.ML
keywords MEGEEGsourceestimationinverseproblemLSTMdSPMMarkovlocalizationDempster-Shafertheory
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

The paper introduces Contextual Minimum-Norm Estimates (CMNE), a source-localization method for MEG and EEG that replaces sample-by-sample estimation with a prediction-correction loop. An LSTM network trained on past dSPM estimates predicts the current source distribution, and the current dSPM estimate corrects that prediction through a Dempster-Shafer product. The authors report that CMNE gives the highest signal-to-noise ratio in space and time, keeps source maps focal over long epochs, and raises SNR by an order of magnitude on real auditory steady-state responses compared with dSPM alone. If this holds, temporal context becomes a practical tool for reducing the ill-posedness of the MEG/EEG inverse problem.

What carries the argument

The central mechanism is a Markov chain built from an LSTM transition model and a dSPM measurement model. The LSTM, a recurrent neural network with forget gates that learns long-range dependencies, maps the last $k$ posterior beliefs $b(\hat{q}_{t-k:t-1})$ to a predicted belief $\bar{b}(\hat{q}_t)$, and the current dSPM estimate corrects that prediction by the product rule of Eq. (8). The LSTM cell state $S_t$ stores context over arbitrarily long time scales without being printed into every output, which is what lets the method carry history beyond the explicit $k$-sample window. This prediction-correction loop has the same shape as a discrete Bayes filter, but with the transition model learned from data rather than hand-specified.

What would settle it

Run CMNE on data with known ground-truth sources while replacing the LSTM prediction with the paper's 80-sample averaging control; if the SNR gap over dSPM disappears, the claimed contextual gain is not coming from the learned LSTM transition. As a second check, inspect the dSPM maps entering Eq. (8): if their entries are negative or do not sum to one, they are not probabilities, and the Dempster-Shafer product is not a valid belief update.

Watch

Extended reading notes

Core claim

CMNE frames source estimation as grid-based Markov localization. At each time $t$, a dSPM estimate $\hat{q}_t$ is computed from the current measurement, while an LSTM network fed with the previous $k$ posterior beliefs produces a predicted belief $\bar{b}(\hat{q}_t)$; the two are combined into the posterior belief $b(\hat{q}_t)=\eta\, p(\hat{q}_t|y_t)\,\bar{b}(\hat{q}_t)$ following Dempster-Shafer theory, and the posterior is fed back into the memory queue. Against dSPM, pure LSTM prediction, and an 80-sample averaging control, CMNE produces the highest SNR in space and time on simulated propagating epileptiform discharges and the most focal activation maps, and on human auditory steady-state data its SNR is an order of magnitude above dSPM. The authors note that the LSTM alone has a lowpass character and misses fast rises, so the dSPM correction is a necessary part of the loop rather than a formality. They conclude that contextual information reduces the ill-posedness of the inverse problem and that the LSTM cell state can be read as an abstract state of underlying brain activity.

Load-bearing premise

The load-bearing premise is that the dSPM output $\hat{q}_t$ can be treated as a probability distribution over source locations in Eq. (8), even though dSPM values are noise-normalized current amplitudes that can be negative and are not normalized; if that premise fails, CMNE's correction is an ad hoc product-and-normalize step rather than a principled Bayesian update.

Editorial extensions

If this is right

  • If CMNE's reported gains hold, MEG/EEG reconstructions of propagating activity, such as epileptiform discharges, will stay focal over the whole event instead of smearing, improving presurgical epilepsy evaluation.
  • For auditory steady-state responses, CMNE's order-of-magnitude SNR gain over dSPM suggests that weak or distributed responses become easier to detect in averaged or single epochs.
  • The control experiment with an 80-sample averaging predictor implies that the learned LSTM transition, not temporal averaging, carries the improvement.
  • The LSTM hidden state, interpreted as an abstract brain state, could be used to relate or integrate different imaging modalities, a direction the paper raises explicitly.

Reading between the lines

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

  • Beyond the paper: the Dempster-Shafer product in Eq. (8) requires treating dSPM output as a probability distribution, which is not guaranteed; replacing it with a calibrated likelihood model would put CMNE on firmer Bayesian ground.
  • Beyond the paper: the same prediction-correction loop could be tested with measurement models other than dSPM, such as beamformers or sparse priors, to see whether the contextual gain transfers.
  • Beyond the paper: the LSTM cell state is a low-dimensional summary of cortical dynamics; if it tracks the underlying state, it could serve as a neural-state signal for brain-computer interfaces or cross-subject comparison, but the paper does not test this.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 6 minor

Summary. The paper proposes CMNE, a source estimation method that combines dSPM estimates with predictions from an LSTM network trained on past dSPM estimates. The method is framed as grid-based Markov localization: the LSTM provides a predicted belief distribution, which is corrected by the current dSPM estimate via a product rule (Eq. 8). The authors evaluate CMNE against dSPM, pure LSTM prediction, and a simple moving-average control, on simulated epileptiform discharges and on auditory steady-state response (ASSR) data from a single human subject. The central claim is that CMNE yields more accurate and more focal source estimates than dSPM alone.

Significance. If the claim holds, the combination of a trained recurrent network with a classical distributed inverse solution would be a novel and potentially useful contribution to MEG/EEG source imaging, especially for exploiting temporal context in tasks such as tracking propagating activity. The paper also contributes a concrete framework for integrating deep learning with linear inverse solvers. However, the theoretical justification is questionable and the empirical validation is thin: single-subject data, no error bars, hyperparameters selected on the same data used for evaluation, and a post hoc SNR metric. The method's core novelty—the contextual correction—rests on an unsupported probabilistic interpretation, so the significance depends on whether the heuristic can be justified or reformulated.

major comments (4)
  1. [Sec. 2.2, Eq. (8)] Equation (8) defines the corrected belief as b(q̂_t)=η p(q̂_t|y_t) b̄(q̂_t), presented as a Dempster–Shafer correction. This combination is only valid if both factors are nonnegative and normalized probability mass functions over the same frame. However, p(q̂_t|y_t) is the dSPM estimate, a signed noise-normalized current amplitude (Sec. 2.1), and b̄(q̂_t) is the LSTM output trained on z-scored dSPM maps (Sec. 3.4), also signed. The product can therefore contain negative entries, and the normalization scalar η cannot produce a probability distribution. Consequently, the grid-based Markov localization / Bayes-filter interpretation collapses, and Eq. (8) reduces to an ad hoc product-and-normalize heuristic. This does not by itself disprove the empirical claim of improved source estimation, but it removes the stated theoretical grounding for the correction step, and the reported improvements could be artifacts of the particular way signed values are combined. Please either justify the product form with a non-probabilistic rationale, or reformulate the method with proper probability maps (e.g., softplus or normalized outputs) and demonstrate that the improvement persists.
  2. [Sec. 4.1 and Sec. 4.3] The LSTM hyperparameters (number of hidden units d and look-back k) were selected empirically using the ASSR data set (Sec. 4.1), and the same ASSR data set, from the same single subject, is then used for validation in Sec. 4.3. This constitutes tuning on the evaluation data, which can inflate the reported SNR gains. The text does not clarify whether the 248 validation epochs used in Sec. 4.3 were part of the data used for hyperparameter selection in Sec. 4.1. Even if they were disjoint, the selection procedure is not independent of the subject. Please report results from a nested cross-validation or from an independent subject, and provide error bars across epochs or subjects.
  3. [Sec. 4.2 and Sec. 4.3, SNR definition] The SNR metric is defined post hoc on manually selected time windows (green intervals in Figures 7 and 9) and uses the maximal dipole amplitude in a predefined label region. This metric is not objective: the window selection is not automated, and the maximum-amplitude criterion is sensitive to the method's own spatial smoothing. No error bars, confidence intervals, or statistical tests are provided for the SNR comparisons, and the simulation results are based on only 20 averaged epochs. Please either report a fully automated, pre-registered SNR definition (e.g., defined over the entire epoch or over a fixed window independent of the estimate) or provide distributions over many single epochs, along with standard localization metrics such as dipole localization error.
  4. [Sec. 3.4 and Sec. 4.2] The LSTM is trained to predict dSPM source estimates, and the final CMNE estimate is a multiplicative combination of the LSTM output and the current dSPM. Because the LSTM is trained on dSPM labels, the improvement over dSPM alone may reflect a self-consistency artifact rather than genuine contextual information. The ground-truth simulation provides an external check, but the simulation training uses only 100 minibatch iterations (Sec. 4.2) and the evaluation is on 20 averaged epochs, which is underpowered. Please include a comparison against a purely linear predictor (e.g., an autoregressive model) trained on the same dSPM data and combined with dSPM in the same way, to demonstrate that the LSTM's nonlinear contextual model is responsible for the improvement.
minor comments (6)
  1. [Sec. 2.2] The notation b̄(q̂t) is used for both the predicted belief distribution and the LSTM output, but the text also introduces u_t as activation change; clarify how u_t is computed from the LSTM and how the Markov-chain formulation connects to the network architecture.
  2. [Sec. 3.3] The ASSR data are described as "the same ASSR data that were used in (Samuelsson et al., n.d.)"; this reference is unpublished and the dependence is unclear.
  3. [Figures 7 and 9] The green time windows used for SNR computation are not visible in grayscale; add distinct line styles or annotations.
  4. [Sec. 4.2] The "estimation error" is described as "the distance between the activated dipole location and the location of the dipole with the greatest amplitude" without a formula; define it quantitatively.
  5. [Sec. 2.1, Eq. (4)] Equation (4) is written as an unconstrained minimization; specify the norm used in the first term and the conditions under which the closed-form solution holds.
  6. [Sec. 2.2] The paper invokes Dempster–Shafer theory but does not define the frame of discernment, the mass functions, or the combination rule; either provide these definitions or drop the terminology and describe Eq. (8) as a heuristic.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: CMNE's improvement claim is checked against ground-truth simulation and an LSTM-free control, not forced by Eq. (8) or by self-citation.

full rationale

The claimed derivation chain is not circular. The LSTM is trained to predict sample-wise dSPM maps (Sec. 3.4), and Eq. (8) combines that prediction with the current dSPM map, but this is a stated modeling choice rather than an identity: b(q̂_t) = η p(q̂_t|y_t) b̄(q̂_t) is not equal to p(q̂_t|y_t) by construction unless the LSTM output is identically one, which is not the case. The empirical claim is anchored to an external ground-truth simulation (Sec. 4.2, Fig. 8) and to a control condition that uses the same past dSPM data without an LSTM (Sec. 3.2), so the reported SNR advantage could fail and is not statistically forced. The only self-citation (Dinh 2015, Fig. 1 caption) is used to label a Markov chain and is not load-bearing. Concerns about the Dempster–Shafer interpretation of Eq. (8) are validity issues, not circularity.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The central method rests on two ad hoc probabilistic assumptions (dSPM as likelihood, LSTM output as prior) and on hyperparameters tuned on the same data used for evaluation. No new physical entities are introduced.

free parameters (3)
  • LSTM lookback window k = 80 samples
    Determined empirically on the ASSR data (Sec 4.1, Figure 6) and used for both simulation and ASSR evaluation.
  • LSTM hidden units d = 1280
    Chosen empirically on the ASSR data as trade-off between prediction accuracy and training time (Sec 4.1, Figure 5).
  • dSPM regularization parameter lambda = 0.1
    Fixed at 0.1 in Sec 3.1, not derived from SNR as in Lin et al. (2006); influences the noise-normalized estimates used as labels and measurements.
assumptions (4)
  • domain assumption The MEG/EEG forward model y_t = G q_t + n_t (Eq. 1) is an accurate description of the measurement process.
    Invoked in Sec 2.1 as the basis for all source estimation.
  • ad hoc to paper The dSPM estimate p(q_t|y_t) can be treated as a likelihood in the Dempster-Shafer correction (Eq. 8).
    Sec 2.2 assumes dSPM outputs are probability-like, but they are noise-normalized amplitudes that can be negative and are not normalized.
  • ad hoc to paper The LSTM output represents a valid predicted belief distribution b_bar(q_t).
    Sec 2.3 uses the LSTM output directly as the prior in Eq. 8 without establishing that it is a probability distribution over source configurations.
  • domain assumption A cortical grid of 5124 dipoles with loose orientation constraint 0.2 gives a sufficient source space.
    Sec 3.1 describes the source grid and orientation constraint as standard preprocessing.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Contextual Minimum-Norm Estimates (CMNE): A Deep Learning Method for Source Estimation in Neuronal Networks." pith.science (2026). https://pith.science/paper/VTPI7LLQ

@misc{pith2026190902636,
  author       = {Pith},
  title        = {Pith review of: Contextual Minimum-Norm Estimates (CMNE): A Deep Learning Method for Source Estimation in Neuronal Networks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/VTPI7LLQ}},
  note         = {Machine review of arXiv:1909.02636}
}
read the original abstract

Magnetoencephalography (MEG) and Electroencephalography (EEG) source estimates have thus far mostly been derived sample by sample, i.e., independent of each other in time. However, neuronal assemblies are heavily interconnected, constraining the temporal evolution of neural activity in space as detected by MEG and EEG. The observed neural currents are thus highly context dependent. Here, a new method is presented which integrates predictive deep learning networks with the Minimum-Norm Estimates (MNE) approach. Specifically, we employ Long Short-Term Memory (LSTM) networks, a type of recurrent neural network, for predicting brain activity. Because we use past activity (context) in the estimation, we call our method Contextual MNE (CMNE). We demonstrate that these contextual algorithms can be used for predicting activity based on previous brain states and when used in conjunction with MNE, they lead to more accurate source estimation. To evaluate the performance of CMNE, it was tested on simulated and experimental data from human auditory evoked response experiments.

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Artificial Neural Networks for Magnetoencephalography: A review of an emerging field

    q-bio.NC 2025-01 conditional novelty 6.0 of 10

    A systematic review of 119 ANN-MEG studies shows rapid growth across decoding, BCI, clinical, modeling, and source-localization applications, with recurring reproducibility gaps.

Reference graph

Works this paper leans on

7 extracted references · 2 canonical work pages · cited by 1 Pith paper

  1. [34]

    A New Approach to Linear Filtering and Prediction Problems

    https://doi.org/10.1002/hbm.20068. Kalman, R. E. 1960. “A New Approach to Linear Filtering and Prediction Problems.” Journal of Basic Engineering 82 (1): 35. https://doi.org/10.1115/1.3662552. Kingma, Diederik P., and Jimmy Ba. 2014. “Adam: A Method for Stochastic Optimization,” 1–15. http://arxiv.org/abs/1412.6980. Lamus, Camilo, Matti S Hamalainen, Simo...

    arXiv 1960

  2. [97]

    Interpreting Magnetic Fields of the Brain: Minimum Norm Estimates

    https://doi.org/10.1103/RevModPhys.65.413. Hamalainen, Matti S, and Risto J Ilmoniemi. 1994. “Interpreting Magnetic Fields of the Brain: Minimum Norm Estimates.” Medical & Biological Engineering & Computing 32 (1): 35–42. https://doi.org/10.1007/BF02512 476. Hamalainen, Matti S, and Jukka Sarvas. 1989. “Realistic Conductivity Geometry Model of the Human H...

    doi:10.1103/revmodphys.65.413 1994

  3. [711]

    From The Cover: The Human Brain Is Intrinsically Organized into Dynamic, Anticorrelated Functional Networks

    https://doi.org/10.1038/nrn2201. 12 Dinh, Samuelsson et al. Fox, Michael D., Abraham Z. Snyder, Justin L. Vincent, Maurizio Corbetta, David C. Van Essen, and Marcus E. Raichle. 2005. “From The Cover: The Human Brain Is Intrinsically Organized into Dynamic, Anticorrelated Functional Networks.” Proceedings of the National Academy of Sciences 102 (27): 9673–...

    arXiv 2005

  4. [2002]

    A Generalized framework of weak value amplification in path interference of polarized light for the enhancement of all possible polarization anisotropy effects

    “Spatiotemporal Forward Solution of the EEG and MEG Using Network Modeling.” IEEE Transactions on Medical Imaging 21 (5): 493–504. https://doi.org/10.1109/TMI.2002.1009385. Jun, Sung Chan, and Barak A. Pearlmutter. 2005. “Fast Robust Subject - Independent Magnetoencephalographic Source Localization Using an Artificial Neural Network.” Human Brain Mapping ...

    work page Pith review arXiv 2002

  5. [2007]

    Hierarchical Bayesian Estimates of Distributed MEG Sources: Theoretical Aspects and Comparison of Variational and MCMC Methods

    “Hierarchical Bayesian Estimates of Distributed MEG Sources: Theoretical Aspects and Comparison of Variational and MCMC Methods.” NeuroImage 35 (2): 669–85. https://doi.org/10.1016/ j.neuroimage.2006.05.001. Nunez, P. L. 1995. Neocortical Dynamics and Human EEG Rhythms . 1st ed. Oxford: Oxford University Press. Olah, Christopher. 2015. “Understanding LSTM...

    work page arXiv 2006

  6. [2014]

    Learning Phrase Representations Using RNN Encoder - Decoder for Statistical Machine Translation,

    “Learning Phrase Representations Using RNN Encoder - Decoder for Statistical Machine Translation,” June. http://arxiv.org/abs/1406.1078. Chollet, François. 2015. “Keras.” GitHub. https://keras.io. Cichy, Radoslaw Martin, Aditya Khosla, Dimitrios Pantazis, and Aude Oliva. 2017. “Dynamics of Scene Representations in the H uman Brain Revealed by Magnetoencep...

    arXiv 2015

  7. [2015]

    Real-Time MEG Source Localization Using Reg ional Clustering

    “Real-Time MEG Source Localization Using Reg ional Clustering.” Brain Topography , 1–14. https://doi.org/10.1007/s10548 -015-0431-9. Engemann, Denis A., and Alexandre Gramfort. 2015. “Automated Model Selection in Covariance Estimation and Spatial Whitening of MEG and EEG Signals.” NeuroImage 108: 328–42. https://doi.org/10.1016/j.neuroimage.2014.12.040. F...

    arXiv 2015

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.