{"id":"dc8257e4-e9d9-42e6-bf00-2217f2294c4d","arxiv_id":"2501.10544","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":8,"one_line_summary":"In simulations, the nonlinear correlation H2, alone or with graph metrics, best tracks electrical diffusion, while H2 with Strength, R2(PageRank), and ICOH(Strength) best track mechanotransduction shifts.","lead":"This paper uses a computer model of the uterus to generate simulated electrohysterogram signals and ranks which connectivity features best detect changes in uterine synchronization. The motivation is to improve early detection of preterm labor.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Admitted limitation: 4x4 electrode matrix (<10 cm) cannot fully capture mechanotransduction, so the claimed 'best features for detecting mechanotransduction shifts' are not supported as stated.","rationale":"The reader's weakest assumption is the fidelity of the simulated EHG to real uterine connectivity, and the paper's own Discussion concedes that the electrode matrix is too small to capture mechanotransduction. This is the most load-bearing point because the central claim is a feature ranking for mechanotransduction, and the signals used to produce that ranking are admitted not to capture the phenomenon. I would keep the verdict CONDITIONAL rather than REJECT: the simulation framework is a plausible methodological contribution, and the issue is testable by rerunning with a larger electrode array. A secondary statistical concern—no reported p-values or correction for multiple comparisons in the slope selection—reinforces the need for the conditional, and the absence of released code/data makes independent verification harder. No evidence in the manuscript contradicts this assessment; the limitation statement in the Discussion is explicit and should be treated as a scope restriction, not a minor caveat.","tokens_in":15479,"tokens_out":5619,"duration_ms":56099,"concrete_test":"Using the same uterine simulation model and identical EDM parameter sweeps, increase electrode coverage—e.g., an 8x8 grid or multiple 4x4 patches spaced >20 cm—and recompute Table 6's mean/median slope rankings. If H2(Str), R2(PR), and ICOH(Str) are no longer the top features, the original ranking is an artifact of the 4x4 electrode matrix. Additionally, compute long-range connectivity (e.g., ICOH/H2 between electrodes >15 cm apart) in the enlarged array; if these values remain near zero, the model itself lacks the long-distance synchronization needed to validate any mechanotransduction feature claim.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central claim is that H2(Str), R2(PR), and ICOH(Str) are the best features for detecting mechanotransduction shifts. However, the Discussion explicitly states that the electrode matrix is 'less than 10cm x 10cm' and 'does not allow for a thorough investigation of the mechanotransduction process associated with long-distance diffusion. As a result, the mechanotransduction process is not fully captured in the EHG signals collected or simulated in this study.' Since mechanotransduction is the long-distance synchronization mechanism the EDM parameter sweeps are supposed to represent, a feature ranking computed from signals that do not fully capture that mechanism cannot establish which features detect mechanotransduction. The ranking may instead reflect local electrodiffusion artifacts of the small electrode grid. No independent validation shows that the model's parameter-induced EHG changes correspond to real uterine mechanotransduction; the comparison to real EHG rankings in Table 6 is qualitative and confounded by the same electrode-size issue. The 'student test' for significance is mentioned in Section 2.6.1 but no p-values or multiple-comparison corrections are reported, so the slopes in Tables 2-5 are not shown to be statistically distinguishable from noise. The admitted limitation alone is sufficient to narrow the conclusion: the framework identifies features sensitive to the simulated model's local mechanotransduction parameters, not to mechanotransduction in the uterus.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a framework for identifying which connectivity features and graph metrics are most sensitive to the two modeled drivers of uterine synchronization: electrical diffusion (ED) and mechanotransduction (EDM). Simulated EHG signals are generated with a multiscale electromechanical uterine model, using a 4x4 electrode matrix, in two groups: one varying tissue resistance (ED) and one varying five mechanotransduction parameters (EDM). Four connectivity measures (H2, R2, FW_h2, ICOH) and five graph metrics (Str, CC, Eff, PR, BC) are computed, and feature sensitivities are assessed by the slopes of feature values versus model parameters after normalization to the first parameter value. The abstract and discussion claim that the best features for detecting mechanotransduction shifts are H2 alone or combined with Str, R2(PR), and ICOH(Str), and that the best features for electrical diffusion shifts are H2, Eff, PR, and BC. The paper also compares these simulated rankings with rankings from real EHG signals using Fscore and AUC.","tokens_in":15802,"tokens_out":3805,"duration_ms":36415,"significance":"If the claims were fully supported, the paper would offer a useful model-based approach for selecting a minimal set of EHG connectivity features to monitor uterine synchronization, with potential implications for preterm-labor monitoring. The study has notable strengths: it systematically varies model parameters, uses 50 simulations per setting, evaluates four connectivity methods and five graph metrics, and compares simulated and real EHG feature rankings. The central claim, however, is only partially supported because the admitted limitation that the 4x4 electrode matrix does not fully capture mechanotransduction directly undermines the stated conclusion about mechanotransduction feature detection. With a revised, narrower claim and proper statistical reporting, the framework could still be a valuable contribution for comparing feature sensitivities in simulated uterine EHG signals.","major_comments":[{"comment":"The admitted limitation that the electrode matrix is 'less than 10cm x 10cm' and 'does not allow for a thorough investigation of the mechanotransduction process associated with long-distance diffusion' directly undermines the paper's central claim that the best features for detecting mechanotransduction shifts are H2(Str), R2(PR), and ICOH(Str). Since the simulated EHG signals are also generated with the same small electrode matrix, the feature rankings for the EDM group may reflect only local diffusion artifacts rather than true long-distance mechanotransduction. The authors should either reframe the claim to state that the identified features are sensitive to the modeled local mechanotransduction parameters in the current electrode configuration, or they should simulate with a larger electrode array that can capture long-distance synchronization.","section":"Section 4 (Discussion, final paragraph)"},{"comment":"The Student's test mentioned in Section 2.6.1 is never reported with p-values, confidence intervals, effect sizes, or multiple-comparison corrections. The slope values in Tables 2-5, many of which are on the order of 0.001-0.01 after normalization to the first parameter value, are therefore not shown to be statistically distinguishable from noise. This is load-bearing because the rankings in Table 6 are based entirely on the magnitudes of these slopes. The authors should provide the test statistics, p-values, or bootstrap confidence intervals for the slopes, and they should account for the fact that 24 features are compared across six parameter scenarios.","section":"Section 2.6.1 and Section 3 (Tables 2-5)"},{"comment":"The FW_h2 filter was changed from the real-EHG band of 0.1-0.3 Hz to 0.1-0.7 Hz for the simulated EHG signals, based on the simulated signals' PSD. This makes the FW_h2 features in the simulated-signal rankings (Table 6, columns 2-3) not directly comparable to the FW_h2 features in the real-EHG rankings (columns 4-5). The claim that FW_h2 'appears to be important in characterizing the mechanotransduction process and uterine synchronization' in both real and simulated signals is therefore confounded by the filter adaptation. The authors should either justify the band choice with a sensitivity analysis or restrict the comparison to features computed with the same filter.","section":"Section 2.6.2 and Table 6"},{"comment":"The external validity of the simulated feature rankings is asserted rather than demonstrated. The model encodes the assumption that lower tissue resistance increases diffusion-based synchronization, so the observed decrease of H2 with increasing resistance (Figure 6) is a consistency check of the model, not an independent validation of the feature ranking. The comparison with real-EHG rankings in Table 6 is qualitative, and no agreement metric (e.g., overlap proportion, rank correlation, permutation test) is provided. The authors should state this limitation explicitly and quantify the agreement between simulated and real rankings if they wish to claim that the simulated results can identify features useful for real EHG monitoring.","section":"Section 2.3 and Section 4"}],"minor_comments":[{"comment":"In the description of FW_h2, the text says 'filtering the Electrogastrogram (EHG) signal'; this should be 'electrohysterogram' to match the EHG abbreviation.","section":"Section 2.4"},{"comment":"The equation for PageRank is referenced as Equation (7), but the displayed equation number is missing and the text jumps from Equation (6) to Equation (8).","section":"Section 2.4"},{"comment":"The text says 'student test' in Section 2.6.1 but the correct term is 'Student's t-test'; also, the caption of Figure 6 says 'significative differences' which should be 'significant differences'.","section":"Section 3"},{"comment":"The caption states that features indicated in blue are selected by Fscore, but no blue color appears in the printed table; the table should be formatted so that this annotation is visible.","section":"Section 3, Table 6"},{"comment":"References [21] and [28] appear to be the same article (Verwaerde et al., 'Statistical shape analysis of gravid uteri...') but are cited as two distinct references; this should be corrected.","section":"References"},{"comment":"The abbreviation FW_h2 is defined as 'Filtered Windowed H2' but the exact formula or algorithmic steps for the filtering and windowing are not given; since the paper modifies the filter band, the definition should be more explicit.","section":"Section 2.4"}],"recommendation":"major_revision","confidential_remarks":"The paper addresses a clinically relevant question and contains a substantial simulation framework, but the admitted electrode-matrix limitation is central to the main conclusion about mechanotransduction features. I recommend major revision rather than rejection because the authors could reframe the claim to the local mechanotransduction setting and add the missing statistical analysis. If the authors cannot simulate with a larger electrode array or provide independent validation, the revised paper should clearly limit its claims to the simulated model's behavior rather than to real uterine mechanotransduction."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear [Colleague],\n\nThis is a useful, honest parameter sweep, and the main thing to know is that it is more modest than the abstract makes it sound. The authors simulate EHG signals with a previously published multiscale electromechanical uterine model, vary tissue resistance (electrical diffusion) and five mechanotransduction parameters, compute four connectivity measures (H2, R2, FW_h2, ICOH) with and without five graph metrics, and rank which features are most sensitive to each driver. The rankings are new, and the comparison with Fscore/AUC rankings from real EHG signals is a reasonable sanity check. The paper is also transparent about its simplifications.\n\nThe soft spot is the one the authors themselves admit in the Discussion: the 4x4 electrode matrix (less than 10 cm x 10 cm) cannot fully capture mechanotransduction, which is a long-distance synchronization process. So the abstract's claim that H2, Str, R2(PR), and ICOH(Str) are 'the best features for detecting mechanotransduction shifts' is too strong. What is actually shown is that these features are sensitive to the mechanotransduction parameters in this model given this small electrode grid; the ranking may partly reflect local electrodiffusion rather than long-distance mechanotransduction. That is a real limitation and it should be fixed either by rephrasing the claims or by simulating a larger electrode array that covers the long-distance scale.\n\nThe statistical support is also thinner than it looks. The student test is mentioned in Section 2.6.1, but no p-values, confidence intervals, or multiple-comparison corrections appear anywhere; the slope tables are reported without error bars. The normalization to the first parameter value is defensible but makes all slopes relative, so the absolute magnitudes in Tables 2-5 should be interpreted carefully. No code or data are released, which makes it hard to verify the rankings independently.\n\nI do not see a fatal circularity. The paper is a model-based feature-selection exercise, not a physiological validation. The circularity concern applies if you read the rankings as evidence about real uterine physiology; the authors should be clearer that this is model sensitivity analysis. For a reader working on EHG monitoring or on this uterine model, the paper has value as a feature ranking and a caution about electrode coverage. It deserves a serious referee, but the authors should be asked to tone down the mechanistic language, report uncertainty, and ideally share their simulation pipeline.","headline":"Useful, honest parameter sweep of connectivity/graph features on simulated EHG, but the abstract's mechanotransduction claim outruns the paper's own admitted electrode-size limitation.","tokens_in":16259,"tokens_out":3227,"would_cite":true,"duration_ms":31895,"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":"Simulated EHG signals rank the features that best track each driver of uterine synchronization.","keywords":["preterm labor","electrohysterography (EHG)","uterine synchronization","connectivity methods","graph theory","mechanotransduction","electrical diffusion","simulated EHG signals"],"falsifier":"Record EHG on a pregnant uterus with an electrode array larger than 10 cm x 10 cm and measure H2(Str), H2(Eff), H2(PR), and H2(BC) during preterm-labor contractions; if these features do not track the expected synchronization increase, or if the simulated ranking reverses, the paper's feature-selection claim is falsified.","tokens_in":15319,"feed_emoji":"🤰","tokens_out":4312,"duration_ms":41174,"temperature":0.7,"pith_summary":"Preterm labor prediction would be stronger if clinicians knew which electrohysterography (EHG) features actually reflect the two physiological drivers of uterine synchronization: short-range electrical diffusion and long-range mechanotransduction. This paper claims that simulated EHG signals from a multiscale electromechanical uterine model can identify those features. Ranking features by how their connectivity values change as model parameters shift, it finds that H2-based measures are the most sensitive to both drivers, with H2 combined with Strength best for mechanotransduction and H2 with Efficiency, PageRank, or Betweenness Centrality best for electrical diffusion. If correct, the result defines a compact feature set for monitoring contraction efficiency and, ultimately, for improving preterm labor detection.","feed_headline":"H2-based features best track uterine synchronization shifts","feed_subtitle":"A 16-electrode simulated uterus ranks connectivity and graph features for each driver of labor.","key_machinery":"The key machinery is the conversion of multichannel EHG signals into connectivity matrices and then into weighted graphs: each electrode is a node and each connectivity value is an edge weight, from which graph metrics are extracted. The connectivity measures are the nonlinear correlation coefficient H2, the linear cross-correlation R2, the filtered windowed H2 (FW_h2), and the imaginary part of coherence (ICOH); the graph metrics are Strength, Clustering Coefficient, Efficiency, PageRank, and Betweenness Centrality. Feature sensitivity is quantified by the slope of each normalized feature against the varied model parameter, with 50 simulations per parameter setting and a student test for significance.","core_discovery":"Using a 4x4 electrode matrix over a simulated uterus, the paper generates EHG signals under two regimes: electrical diffusion alone (ED), by varying tissue resistance, and electrical diffusion plus mechanotransduction (EDM), by varying five mechanotransduction parameters. For each regime it computes four connectivity methods (H2, R2, FW_h2, ICOH), with and without five graph metrics, and orders the features by the slope of their normalized response. The paper's central claim is that the best features for detecting mechanotransduction shifts are H2 alone or combined with Str, R2(PR), and ICOH(Str), while the best features for detecting electrical diffusion shifts are H2 alone and combined with Eff, PR, and BC. The paper further claims that FW_h2 is prominent for mechanotransduction and that a simplified electromechanical model suffices to monitor uterine synchronization from simulated EHGs, despite acknowledged simplifications.","pith_inferences":["Because real labor changes electrical diffusion and mechanotransduction together, the single-parameter perturbation design may underestimate interactions between the two; a combined-parameter sweep would test this.","The authors note the 10 cm x 10 cm electrode grid is too small to capture long-distance mechanotransduction, so the H2(Str) ranking might change with a larger electrode array.","The slope-ranking method could be reused to validate other biophysical models of uterine activity, not just this one.","A direct testable extension is to record real EHG with a wider electrode matrix and check whether H2(Str) increases in labor as the simulated ranking predicts."],"forward_implications":["A minimal feature set (H2 with Str for mechanotransduction; H2 with Eff, PR, or BC for diffusion) could be prioritized in future preterm-labor classifiers.","FW_h2's strong showing supports filtering EHG to the low-frequency band when targeting mechanotransduction.","The rankings give modelers a target: simulated EHG features should respond to the same parameter shifts that real labor does.","The 0.1-0.7 Hz passband identified for simulated EHG FW_h2 could guide filter settings for real recordings.","H2's dual sensitivity indicates nonlinear correlation captures both local diffusion and long-distance synchronization."],"supporting_citations":[{"why":"Supplies the mechanotransduction theory that defines the long-distance synchronization process the EDM simulations target.","marker":"[12]"},{"why":"Introduces the nonlinear correlation coefficient H2 for 16-electrode EHG, the core connectivity measure of the study.","marker":"[13]"},{"why":"Provides the graph-theory analysis of real EHG signals that serves as the baseline for comparing simulated feature rankings.","marker":"[17]"},{"why":"Previous real-signal feature selection results that the simulated EHG rankings are directly compared against.","marker":"[20]"},{"why":"The multiscale uterine model with mechanotransduction that generates the simulated EHG signals.","marker":"[26]"},{"why":"Supplies the uterine geometry and mechanical sub-models used in the simulation pipeline and the expected parameter effects.","marker":"[28]"},{"why":"Defines FW_h2, the filtered windowed H2 method that emerges as important for mechanotransduction.","marker":"[31]"},{"why":"Provides the Fscore feature selection on real EHG signals that is the comparison standard for the simulated best features.","marker":"[44]"}],"fun_headline_variants":["H2 and graph features best track uterine synchronization shifts","Simulated uterus: H2 ranks top for tracking contraction drivers","H2 plus graph metrics pinpoint mechanotransduction and diffusion","Connectivity and graph features rank H2 as key for uterine sync"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The rankings are only as trustworthy as the simulation: simulated EHG signals must change with tissue resistance and mechanotransduction parameters the way real uterine electrical activity does, and the paper's own electrode grid is too small to fully capture mechanotransduction.","fun_headline_variants_meta":{"raw":{"variants":["H2 and graph features best track uterine synchronization shifts","Simulated uterus: H2 ranks top for tracking contraction drivers","H2 plus graph metrics pinpoint mechanotransduction and diffusion","Connectivity and graph features rank H2 as key for uterine sync"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000218,"raw_usage":{"total_tokens":1440,"prompt_tokens":946,"completion_tokens":494,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":562,"completion_tokens_details":{"reasoning_tokens":425}},"tokens_in":562,"tokens_out":494,"duration_ms":4920,"temperature":1.0,"reasoning_tokens":425,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T19:08:10.800526+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Record EHG on a pregnant uterus with an electrode array larger than 10 cm x 10 cm and measure H2(Str), H2(Eff), H2(PR), and H2(BC) during preterm-labor contractions; if these features do not track the expected synchronization increase, or if the simulated ranking reverses, the paper's feature-selection claim is falsified.","supporting_citations":[{"cited_title":"Mechanotransduction mechanisms for coordinating uterine contractions in human labor,","cited_arxiv_id":null,"evidence_quote":"Supplies the mechanotransduction theory that defines the long-distance synchronization process the EDM simulations target."},{"cited_title":"Uterine electromyography: A critical review,","cited_arxiv_id":null,"evidence_quote":"Introduces the nonlinear correlation coefficient H2 for 16-electrode EHG, the core connectivity measure of the study."},{"cited_title":"Inter-electrode delay estimators for electrohysterographic propagation analysis,","cited_arxiv_id":null,"evidence_quote":"Provides the graph-theory analysis of real EHG signals that serves as the baseline for comparing simulated feature rankings."},{"cited_title":"Multi-scale and multi-physics model of the uterine smooth muscle with mechanotransduction,","cited_arxiv_id":null,"evidence_quote":"The multiscale uterine model with mechanotransduction that generates the simulated EHG signals."},{"cited_title":"A continuum model for excitation–contraction of smooth muscle under finite deformations,","cited_arxiv_id":null,"evidence_quote":"Supplies the uterine geometry and mechanical sub-models used in the simulation pipeline and the expected parameter effects."},{"cited_title":"Uterine Synchronization Analysis During Pregnancy and Labor Using Graph Theory, Classification Based on Neural Network and Deep Learning,","cited_arxiv_id":null,"evidence_quote":"Provides the Fscore feature selection on real EHG signals that is the comparison standard for the simulated best features."}],"review_version":1}