{"id":"ec35ce44-dcb2-4f04-8159-b50bb309d207","arxiv_id":"2607.17602","paper_version":1,"verdict":"UNVERDICTED","confidence":"HIGH","novelty_score":0.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"A tutorial review of how EEG/MEG source imaging and connectivity metrics are used to estimate human brain networks.","lead":"This paper is a methods chapter that explains how EEG and MEG recordings can be turned into maps of brain networks. It reviews the physics, software, and statistical measures used to avoid artifacts when studying connections between brain regions.","discovery_kind":"review","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Source-space recommendation rests on an unquantified extrapolation from localization accuracy to connectivity accuracy; head-model errors and residual source leakage may undermine the default even with leakage-robust metrics.","rationale":"The reader identified the same underlying assumption: subject-specific head-model accuracy is load-bearing for the central claim. I agree that this is the weakest point. However, I refine the concern: the chapter's argument extrapolates from source-localization accuracy to connectivity accuracy without evidence, and it underplays the source-leakage literature it cites. This is a genuine gap in the argument, but it does not change the overall verdict. The work is a review chapter, not an original research claim; its value is didactic, and the central recommendation is consistent with current consensus (even if that consensus is itself based on indirect evidence). The chapter already includes caveats ('no single method completely resolves the issue'), and the missing quantitative comparison does not make the chapter misleading—it makes it incomplete as a standalone justification. The editorial leftover ('Several complementary metrics were used in this study') and the uncited companion-chapter reference are quality issues that further support the UNVERDICTED classification, but they are not load-bearing for the methodological argument. The proposed simulation is a concrete way to test whether the extrapolation holds; if it fails, the chapter would need to soften its 'defensible default' phrasing, but the verdict for a review would still not be acceptance or rejection of a research result.","tokens_in":18151,"tokens_out":4326,"duration_ms":45857,"concrete_test":"Simulate a known 20-node cortical network with known ground-truth edges using an FEM head model with true conductivities; generate 5 minutes of EEG/MEG with realistic noise. Reconstruct sources with (a) the correct head model, (b) the same model with skull conductivity perturbed by ±30%, and (c) an MNI template head model. Compute wPLI and imaginary coherence in source space for each reconstruction, plus sensor-space wPLI, and compare area under the ROC curve for detecting true edges. If source-space connectivity with perturbed or template head models does not outperform sensor-space wPLI, the chapter's stated default—source-space plus leakage-robust metrics—needs a significant caveat about head-model dependence.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The chapter's central claim is that interpretable EEG/MEG network analysis requires source reconstruction plus zero-lag-suppressing metrics. This claim depends on the premise that source estimates from subject-specific head models are accurate enough for connectivity estimation. However, the cited support ([19,20]) is about source localization accuracy, not connectivity accuracy. A forward model that improves dipole localization does not automatically improve bivariate connectivity estimates, because connectivity is sensitive to correlated errors across locations, not just absolute localization error. Moreover, linear inverse solutions (MNE, dSPM, beamformers) produce spatial leakage that creates 'ghost interactions' even with perfect anatomy (Palva et al. 2018, cited as [29] but not integrated). wPLI and imaginary coherence suppress zero-lag components but do not remove all leakage artifacts; residual phase-delayed leakage can still bias connectomes. If conductivity values or tissue segmentation are wrong, localization biases propagate nonlinearly into phase-delay estimates, and no leakage-robust metric can recover the true network. The chapter acknowledges that no method completely resolves volume conduction, but its 'defensible default' claim would require evidence that, under realistic head-model error, source-space wPLI/iCoh actually recovers true edges better than sensor-space metrics or experimental contrasts. That evidence is not provided, and the cited literature does not directly test it. The argument is therefore plausible but not secured by the chapter's own evidence.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The chapter is a review of EEG/MEG-based brain network analysis. It covers the physical principles of EEG and MEG, forward and inverse modeling, volume conduction and leakage, functional and effective connectivity measures, software pipelines (with emphasis on Brainstorm and BrainSuite), and emerging topics such as time-varying and cross-frequency connectivity. The central methodological claim is that interpretable EEG/MEG network analysis requires solving the inverse problem to obtain source-space time series, and that connectivity measures should suppress zero-lag, volume-conduction-related coupling (e.g., imaginary coherence, wPLI, orthogonalized amplitude envelope correlation). The chapter is framed as a tutorial/reference for advanced students and researchers, and it includes practical recommendations for reproducible workflows and data standards.","tokens_in":971,"tokens_out":1195,"duration_ms":82349,"significance":"If used as a reference chapter, it provides a broad and mostly accurate synthesis of current EEG/MEG connectivity methodology. Its strengths are the clear explanation of volume conduction, the balanced enumeration of functional and effective connectivity measures, the emphasis on source-space interpretation, and the integration of modern reproducible pipelines and BIDS. The paper does not present new empirical results or a novel methodological derivation, so its value is pedagogical and organizational. The stress-test concern that source-space connectivity is recommended without direct quantitative validation does not fully land: the chapter explicitly acknowledges that source-space leakage is not completely resolved and recommends confirmatory controls, and the key claims are supported by standard review citations rather than presented as new quantitative findings. A few editorial revisions would improve clarity and balance.","major_comments":[],"minor_comments":[{"comment":"The sentence 'Several complementary metrics were used in this study' appears to be a leftover from a data-analysis manuscript and is confusing in a review chapter. Please rephrase, e.g., 'The main categories of metrics discussed below include...'.","section":"Functional Connectivity Measures (Undirected)"},{"comment":"The software name is inconsistently capitalized ('Brainsuite' in one passage, 'BrainSuite' elsewhere). Standardize throughout, including in the reference description.","section":"BrainSuite for MRI-Based Head Modeling, Structural Connectivity and beyond"},{"comment":"The recommendation that subject-specific models are preferred 'for best accuracy' is reasonable, but it is stated without a direct supporting citation. Adding a specific reference (e.g., Vorwerk et al. 2014 [15] or Darvas et al. 2006 [23]) would help readers locate the evidence.","section":"Template head models vs. subject-specific models"},{"comment":"To preempt a common objection, consider adding one sentence citing Palva et al. 2018 [29] that residual phase-delayed leakage can still create ghost interactions in source space, and that leakage-robust metrics reduce but do not eliminate this risk. The current text already notes that no method completely resolves the issue, but making the residual-leakage point explicit would strengthen the best-practice recommendation.","section":"Sensor vs. Source Space, Volume Conduction and Its Impact on Connectivity"},{"comment":"Reference [61] is a Zenodo tutorial/repository. If the chapter permits, update to the most recent version with a stable DOI so that the reference is fully permanent.","section":"References"},{"comment":"The caption states that edges are 'derived from the white matter fiber tracks'. If this is just a visualization overlay rather than tractography-based structural connectivity, the wording should be clarified to avoid implying that EEG/MEG connectivity edges are computed from structural connectivity.","section":"Figure 6 caption"}],"recommendation":"minor_revision","confidential_remarks":"The chapter has a promotional tone for the authors' own tools: Brainstorm, BrainSuite, and Brainstorm-DUNEuro are described in detail, with a high density of self-citations. This is not a correctness problem, but an editor may want the authors to add a sentence acknowledging that other toolkits (FieldTrip, MNE-Python, EEGLAB) have comparable capabilities and are also described as widely used, so that the chapter does not read as primarily a software advertisement."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"This is a review chapter, not a research paper. It synthesizes established EEG/MEG source imaging and connectivity methods accurately and reads well as a primer. The main things to know: it's a didactic summary with no new methods or data, and its strongest practical advice—source-space connectivity with leakage-robust metrics—is plausible field consensus but is asserted rather than demonstrated here.\n\nThe chapter does a good job covering the full pipeline: forward models (BEM/FEM, DTI-informed conductivity), inverse methods (MNE, dSPM, sLORETA, MEM), volume conduction problems, functional and effective connectivity measures, and software workflows (Brainstorm, BrainSuite). The descriptions are accurate and appropriately hedged in places—for example, noting that no method completely resolves volume conduction. The practical guidance on BIDS, reproducibility, and experimental contrasts is sensible and would help advanced students.\n\nThe soft spots are real but not fatal for a review. First, the chapter leans heavily on the authors' own Brainstorm and BrainSuite, with frequent references to those tools and no disclosure statement. That's not circular in a logical sense, but it reads promotional. Second, there are editing leftovers: a sentence in the connectivity section says \"Several complementary metrics were used in this study\"—an artifact of a prior empirical paper—and there's an uncited reference to a companion chapter. These should be fixed before publication. Third, the stress-test concern is legitimate: the chapter moves from \"subject-specific FEM improves source localization\" (supported by refs 19, 20) to the recommendation that source-space connectivity with wPLI/iCoh is the defensible default. But localization accuracy is not connectivity accuracy; spatial leakage creates ghost interactions even with good anatomy (Palva et al., cited but not engaged). The chapter acknowledges leakage limitations, but it doesn't quantify or justify the leap. For a review, I'd accept that as a reasonable statement of current practice, but it should be phrased as practice rather than settled evidence.\n\nWho is this for? Advanced students and researchers new to EEG/MEG network analysis. I'd recommend it as reading material, not as a research contribution. It deserves a serious referee to catch the promotional tone, the leftovers, and to insert caveats about the source-space default. I'd send it to peer review in that spirit.","headline":"A competent, clearly written review chapter on EEG/MEG network analysis; useful as a teaching resource, but it carries promotional self-citation and an unquantified leap from source localization to connectivity accuracy.","tokens_in":18888,"tokens_out":1951,"would_cite":false,"duration_ms":19572,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"The chapter argues that mapping brain networks with EEG/MEG requires moving from sensor-level signals to source-space estimates via forward and inverse modeling, and then using connectivity metrics that suppress zero-lag volume conduction.","keywords":["EEG","MEG","brain networks","functional connectivity","effective connectivity","source reconstruction","volume conduction","connectivity measures"],"falsifier":"A ground-truth test with simultaneous intracranial EEG and scalp EEG/MEG, in which the known implanted-network connectivity is compared with sensor-level and source-level estimates, would settle the claim: if leakage-robust source-space connectivity does not outperform sensor-level connectivity, the chapter's central premise is weakened.","tokens_in":17997,"feed_emoji":"🧠","tokens_out":7331,"duration_ms":55954,"temperature":0.7,"pith_summary":"EEG and MEG record brain activity at millisecond resolution, but their sensor-level signals are a mixture of many underlying sources, so raw sensor correlations do not reveal real brain networks. This chapter argues that trustworthy network analysis must first solve the forward problem—modeling how currents in the head produce the measured fields—and then the inverse problem, reconstructing cortical source activity. Once source time series are obtained, connectivity should be computed with measures that down-weight or ignore zero-lag correlations (such as imaginary coherence, the phase-lag index, and its weighted version), because volume conduction and field spread create spurious zero-phase coupling. The chapter also surveys directed (effective) connectivity methods and presents an integrated, reproducible analysis pipeline for doing all of this. For a sympathetic reader, the value is a defensible methodological default: source space plus leakage-robust metrics, built from subject-specific head models.","feed_headline":"EEG/MEG networks need source reconstruction, not sensor data.","feed_subtitle":"Sensor-level correlations mislead; source-space phase-lag metrics give defensible networks.","key_machinery":"The central machinery is the combined forward-inverse pipeline that converts scalp/magnetometer measurements into cortical source time series. The forward problem uses a volume-conductor model of the head (boundary-element or finite-element) to compute a lead-field matrix linking current dipoles to sensors; the inverse problem estimates the dipole currents, typically with minimum-norm or beamforming methods. This is what makes connectivity interpretable. A second mechanism is the use of zero-lag-discounting connectivity metrics—imaginary coherence, phase-lag index (and weighted PLI), and source orthogonalization before amplitude-envelope correlation—which remove instantaneous coupling that v","core_discovery":"The chapter's central claim is that interpretable EEG/MEG brain-network analysis depends on source reconstruction and on connectivity metrics that suppress zero-lag volume conduction. Sensor-space correlations are ambiguous: a single source can appear at many sensors with no time delay, so strong coherence between channels may reflect a common source rather than an interaction. The authors therefore advance a workflow in which the forward problem is solved with a volume-conductor model of the head (ideally subject-specific), the inverse problem is solved to estimate cortical currents, and connectivity is then computed on those reconstructed sources using measures such as imaginary coherence,","pith_inferences":["If the chapter's argument is correct, a large portion of the existing sensor-level coherence literature may need to be reinterpreted or reanalyzed, since many reported 'connections' could be volume-conduction artifacts.","A testable extension would be to run the same connectivity analysis at sensor and source levels on data with concurrent intracranial recordings; if source-space metrics do not track the known network better than sensor-space metrics, the central claim would be challenged.","The emphasis on zero-lag artifacts suggests that combined EEG+MEG source reconstruction, which uses complementary sensitivity patterns, may further reduce leakage and produce more accurate network estimates than either modality alone.","The chapter's best-practice framework implies that when individual MRIs are unavailable, template models should be used with caution and results should be cross-checked with at least two leakage-resistant metrics."],"forward_implications":["Sensor-level connectivity should not be used as evidence for brain networks, because volume conduction creates spurious zero-lag correlations between channels that do not reflect interactions.","The recommended default for EEG/MEG network analysis is source-space connectivity using metrics that suppress zero-lag coupling (imaginary coherence, PLI/wPLI, or orthogonalized envelope correlation).","Subject-specific head models built from individual MRIs are preferred over template models, since anatomical mismatches degrade source localization and contaminate downstream connectivity.","Directed (effective) connectivity methods such as Granger causality, dynamic causal modeling, and transfer entropy should be applied to reconstructed source signals, not to sensor recordings, and interpreted conservatively.","Integrated, reproducible pipelines that combine preprocessing, head modeling, source estimation, and connectivity computation make group-level and clinical network studies feasible."],"fun_headline_variants":["Sensor correlations deceive; reconstructed sources reveal true brain networks","Zero-lag leak: why sensor-phase coherence misleads network analysis","For real brain networks, ditch sensor data and reconstruct cortical sources","Volume conduction blurs sensor connectivity; source-space metrics fix it","Source reconstruction is non-negotiable for trustworthy EEG/MEG networks"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The argument rests on the premise that subject-specific, anatomically accurate head models yield source estimates accurate enough that the reconstructed regional time series reflect real brain activity rather than leakage artifacts.","fun_headline_variants_meta":{"raw":{"variants":["Sensor correlations deceive; reconstructed sources reveal true brain networks","Zero-lag leak: why sensor-phase coherence misleads network analysis","For real brain networks, ditch sensor data and reconstruct cortical sources","Volume conduction blurs sensor connectivity; source-space metrics fix it","Source reconstruction is non-negotiable for trustworthy EEG/MEG networks"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.0003,"raw_usage":{"total_tokens":1571,"prompt_tokens":746,"completion_tokens":825,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":490,"completion_tokens_details":{"reasoning_tokens":738}},"tokens_in":490,"tokens_out":825,"duration_ms":7443,"temperature":1.0,"reasoning_tokens":738,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-01T17:30:04.559466+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A ground-truth test with simultaneous intracranial EEG and scalp EEG/MEG, in which the known implanted-network connectivity is compared with sensor-level and source-level estimates, would settle the claim: if leakage-robust source-space connectivity does not outperform sensor-level connectivity, the chapter's central premise is weakened.","supporting_citations":[],"review_version":1}