{"id":"c6c84829-7c8d-4a8c-996b-5b05c59302f8","arxiv_id":"2608.10331","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":6,"one_line_summary":"History-dependent softening in a full-head finite element model amplifies brain deformation and predicted injury risk under repeated head impacts.","lead":"A computer model of the human head that lets brain tissue soften after each impact shows that repeated hits can amplify predicted brain strain and injury risk beyond what standard models predict. The effect builds over successive impacts and can make a weaker later punch look riskier than an earlier harder one.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Rate-independent quasi-static Mullins parameters are applied to 15 ms impact loading without viscoelastic damping; the reported amplification may largely be an artifact of this untested transfer.","rationale":"The reader's weakest assumption identifies the same load-bearing concern: quasi-static Mullins parameters are transferred rate-independently to impact-rate loading. My stress-test sharpens this into a specific mechanism: the progressive amplification is a dynamic overshoot that depends on the absence of viscous dissipation, and the paper's own validation (Section S1.2) shows oscillations in the Mullins-based model that are absent experimentally, with only Marginal strained-volume-fraction agreement. This makes the rate transfer the single step on which the abstract's quantitative claim rests. I also considered other candidate concerns, including injury risk function transfer, the hand-chosen 0.25 scaling of multiaxial kinematics, single-anatomy generality, and absence of released code; these are real but either acknowledged in the paper or secondary to the constitutive transfer. Because the study is a controlled computational demonstration with honest limitations, and because the central claim is explicitly framed as 'these findings demonstrate' rather than as a clinical prediction, conditional acceptance remains appropriate. No verdict change is warranted, but the paper should require the proposed viscoelastic/rate-dependent sensitivity check or equivalent impact-rate experimental justification before the quantitative amplification values are used.","tokens_in":37685,"tokens_out":7552,"duration_ms":80178,"concrete_test":"Augment the Mullins formulation with the same Prony-series viscoelasticity already used in the LVHE model (Table 3) and rerun the three single-axis five-cycle protocols and the twenty-cycle multiaxial sequence, keeping geometry, mesh, loading, and post-processing identical. Then compare cycle-wise CMPS95, CSDM15, CMPSR95, and sequence-level P(at least one injury) between the Mullins-viscoelastic and original Mullins models, and between the Mullins-viscoelastic and HE/LVHE models. If the Cycle 1-5 CMPS95 increases shown in Table 5 fall below roughly 20% of their current values, or if the final CMPS95-based probability gap (77.0% versus 50.0%) narrows by more than half, then the rate-independent, damping-free Mullins formulation is the load-bearing cause of the reported amplification, and the central claim fails at impact rates.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim requires that quasi-static cyclic softening, calibrated from Franceschini et al. (2006) (Table 1: r=1.2, m=0.05 kPa), remains operative at the 15 ms impact rates used in the simulations. This assumption enters at Section 2.1 and is applied uniformly to all seven parenchymal regions under a common brain-bulk response (Section 2.2.2). The Mullins-based model contains no viscoelasticity, so the softened unloading branch is undamped; the validation in Section S1.2 already shows that this formulation produces oscillations absent in the tagged-MRI data and only 'Marginal' strained-volume-fraction agreement (MPS>0.01 CORA=0.283). Mechanistically, the progressive CMPS95/CSDM15 amplification is a dynamic overshoot effect: at the instant Wdev=Wdev,max, eta returns to 1 in Eq. (3), so each cycle's new peak is achieved on the intact envelope only after a softer, less-damped approach. Omitting rate-dependent damage and viscous dissipation can therefore amplify the overshoot substantially. The paper acknowledges this in Section 4 ('The rate-independent Ogden-Roxburgh formulation and common Mullins parameters assigned to the seven parenchymal brain substructures therefore reflect the available calibration evidence'), but the central abstract claim is quantitative: 'substantially amplify tissue deformation and injury-risk estimates.' Without impact-rate cyclic data or a rate-dependent model, the magnitude of the claimed effect, including the 77.0% versus 50.0% CMPS95-based sequence-level probability gap, remains unsupported.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper implements an Ogden–Roxburgh Mullins damage model in a high-fidelity, subject-specific finite element head model and applies repeated head-loading histories derived from MMA impact data. Under five identical single-axis cycles, the Mullins-based model shows progressive increases in CMPS95, CSDM15, strain-rate metrics, and strain-based AIS2 injury probabilities, while a hyperelastic (HE) comparison model remains nearly constant and UBrIC-based predictions are unchanged. Under a randomized twenty-cycle multiaxial sequence, the Mullins-based model produces larger late-sequence deformation and the highest probability of at least one injury over the sequence (77.0% by CMPS95 vs 50.0% HE and 67.9% LVHE), despite lower initial estimates. The study also reports regional differences in damage-evolution coefficients and higher absolute CMPS in gyral than sulcal elements, with larger proportional amplification in sulci. The authors conclude that short-term damage-mediated softening can substantially amplify tissue deformation and injury-risk estimates relative to damage-free models under the same prescribed loading histories, while noting that rate-dependent, region-specific, and recovery behavior remain experimentally uncharacterized.","tokens_in":37955,"tokens_out":2417,"duration_ms":26249,"significance":"If the central claim holds, the paper identifies a mechanistic pathway through which repeated head impacts could produce escalating tissue deformation and injury risk even when kinematics are identical or declining, which is a meaningful departure from conventional isolated-impact head modeling. The strengths of the study are the clean controlled comparisons that isolate the Mullins damage effect, the subject-specific tagged-MRI validation effort, and the unusually candid treatment of limitations. The paper makes its assumptions explicit, provides detailed constitutive equations in the supplement, and does not overstate the mechanistic link between the constitutive damage variable and biological injury. However, the quantitative magnitude of the claimed amplification rests on an unverified transfer of quasi-static Mullins parameters to impact-rate loading, and the strained-volume-fraction validation is marginal; these issues affect the strength of the central quantitative claim rather than the existence of the qualitative effect.","major_comments":[{"comment":"The quasi-static Mullins parameters r=1.2 and m=0.05 kPa from Franceschini et al. (2006) are applied rate-independently to 15 ms impact pulses, and the model contains no viscoelastic dissipation in the brain bulk. This transfer is load-bearing for the abstract's quantitative claim that softening can 'substantially amplify' deformation and injury risk. The validation in §S1.2 shows that this formulation produces oscillations absent in the tagged-MRI data, and Section 4 acknowledges that the rate-independent formulation reflects available calibration evidence rather than demonstrated rate behavior. The magnitude of the reported amplification, including the 77.0% vs 50.0% sequence-level CMPS95-based probabilities, is therefore conditional on an untested assumption. The authors should either add a rate-dependent damage formulation or clearly reframe the quantitative results as an upper-bound scenario pending impact-rate cyclic tissue data.","section":"§2.1 (Table 1) and §2.2.2"},{"comment":"The strained-volume-fraction validation for the Mullins model receives only 'Marginal' CORA scores (e.g., MPS>0.01 CORA=0.283), yet CSDM15-based injury probabilities are reported as primary results in Figures 7, 8, and 11. The paper itself notes in Section 4 that greater confidence should be placed in CMPS95 trends than CSDM15 trends. This is a direct conflict between the validation evidence and the weight given to CSDM15-based sequence-level probabilities (99.9% vs 88.1%). The manuscript should either demote CSDM15-based probability estimates to a secondary role in the abstract and conclusions or provide additional justification for why the marginal strained-volume-fraction agreement is sufficient for quantitative CSDM15 risk predictions.","section":"§S1.2, Table S1; results in §3.1 and §3.2"},{"comment":"The multiaxial sequence uses a hand-chosen scaling factor of 0.25 applied to the sampled angular velocities, chosen to prevent 'immediate saturation of injury metrics.' The sequence-level probability comparisons in §3.2, including the 77.0% vs 50.0% CMPS95-based result, depend directly on this scaling factor, and no sensitivity analysis is provided. Because the relative ordering of the models changes with accumulated damage, it is plausible that a different scaling factor or a different random seed could alter not only the absolute probabilities but also the cycle at which the Mullins-based model overtakes the damage-free models. The authors should report sensitivity to the scaling factor and to the stochastic sampling, or explicitly characterize the reported sequence-level probabilities as illustrative single-realization outcomes.","section":"§2.3.1, multiaxial loading protocol"}],"minor_comments":[{"comment":"The choice of a 200 ms rest period is justified by strain decay, but the sensitivity of the results to this inter-cycle interval is not examined; given that Budday et al. (2020) report recovery over ~60 min, a brief sentence on how longer rest periods would diminish the predicted amplification would help readers interpret the time scale of the claimed effect.","section":"§2.3.1"},{"comment":"The regional damage evolution coefficients k are obtained from exponential fits to only five data points per region and direction, but no goodness-of-fit statistics are reported; the inset figure suggests some fits may be poorly constrained, so the authors should report R² or confidence intervals for the fitted k values.","section":"§3.3, Eq. (26)"},{"comment":"The text states that standard deviations are not used in the single-axis cases, but the axial and coronal mean values are used without a discussion of whether these cases are representative of typical MMA impacts given the large standard deviations reported in Table 4.","section":"§2.3.1"},{"comment":"There are minor typographical issues, such as 'T able 1' in the text preceding Table 1, and inconsistent hyphenation of 'Mullins-based' (e.g., 'Mullins-based' vs 'Mullins based'); these should be corrected during revision.","section":"Throughout"},{"comment":"The discussion of the higher absolute gyral CMPS compared with previous sulcal-strain studies is fair, but the authors could strengthen it by noting that the present analysis uses only 75 manually selected pairs, and a sensitivity check on the pairing procedure would increase confidence in the reported p<0.001 differences.","section":"§4, gyral–sulcal analysis"}],"recommendation":"major_revision","confidential_remarks":"The paper is well-aligned with the journal's scope and the simulation study is carefully executed, but the gap between the central quantitative claim and the unverified rate-transfer assumption is the main barrier to acceptance. I would advise the editor that the manuscript could become acceptable after the authors either (a) add a rate-dependent damage formulation or a sensitivity analysis over plausible rate-dependent parameter ranges, and (b) recalibrate the abstract and conclusions to match the strength of the validation evidence for CSDM15-based predictions. The current version is not rejectable because the qualitative mechanism is a direct and reproducible consequence of the Mullins formulation, but the quantitative 'substantial amplification' claim needs to be better supported or more carefully bounded."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Worth a look: this is the first organ-scale FE head model with Ogden-Roxburgh Mullins softening under repeated loading, and the controlled single-axis comparisons cleanly separate the damage effect from the hyperelastic baseline. The qualitative conclusion - that prior loading history can make identical head kinematics produce larger tissue strains and higher strain-based injury estimates - is well supported within the model. The paper is also honest about its constraints; the limitations section explicitly flags the rate-independent formulation, the common brain-bulk parameters, and the transfer of injury risk functions.\n\nThe soft spots are real but not fatal. The Mullins parameters r=1.2 and m=0.05 kPa come from quasi-static cyclic tests (Franceschini 2006) and are applied to 15 ms impact pulses with no rate-dependent damage and no viscoelastic damping in the brain. The model's own validation shows oscillations absent in the tagged-MRI data and only 'Marginal' strained-volume-fraction agreement. So the progressive amplification (e.g., CMPS95-based sequence probability 77% vs 50% for HE) is a prediction of an unverified constitutive transfer, not a measured effect. The qualitative direction may still hold if cyclic softening operates at impact rates, but the magnitude is not yet grounded. The hand-chosen 0.25 scaling of the multiaxial kinematics and the Bernoulli independence assumption for sequence-level probabilities add more uncertainty. All of this is acknowledged, which I appreciate.\n\nThe math and the computational setup look sound. The mesh is large (1.9M nodes), the comparison models are consistent, and the CORA score for MPS95 is good (0.724) even if volume-fraction scores are marginal. The citation pattern is appropriate; prior 2D/small-volume damage models are properly credited. No code or data released, which limits independent reproduction.\n\nWho benefits: anyone working on computational head injury biomechanics, sports-related concussion risk, or constitutive modeling of brain tissue. It deserves a serious referee - the novelty is clear and the limitations are foreseeable rather than disqualifying. I'd send it out with a request for the authors to either add a sensitivity analysis on the damage parameters or to reframe the quantitative risk claims as illustrative rather than predictive.","headline":"First full-head Mullins-damage repeated-loading study; the qualitative amplification claim holds up within the model, but the quantitative risk numbers rest on quasi-static damage parameters transferred to impact-rate loading.","tokens_in":38563,"tokens_out":1937,"would_cite":true,"duration_ms":20221,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Repeated head impacts can soften brain tissue enough to push deformation and injury-risk estimates above what single-impact models predict.","keywords":["traumatic brain injury","repeated head impacts","Mullins effect","Ogden-Roxburgh damage model","finite element head model","injury risk prediction","brain tissue softening","mixed martial arts head impacts"],"falsifier":"Run cyclic tension-compression tests on human brain tissue at strain rates near 100 per second and fit the Ogden–Roxburgh parameters to the observed second-cycle softening; if the measured $r$ and $m$ differ strongly from $r=1.2$ and $m=0.05$ kPa, the amplification predictions do not transfer to impact conditions. An organ-scale check would deliver two identical head rotations a few seconds apart in tagged-MRI experiments and compare the second strain field to the first.","tokens_in":37417,"feed_emoji":"🧠","tokens_out":8747,"duration_ms":77205,"temperature":0.7,"pith_summary":"The paper takes on a blind spot in computational head-injury modeling: impacts are usually simulated one at a time, even though real exposure comes in sequences. It builds a high-fidelity finite-element head model in which the brain bulk softens across loading cycles through a Mullins-type damage law (a material's stress response weakening after repeated loading to the same level) calibrated to cyclic human brain tissue, and runs both controlled five-cycle single-axis loadings and a randomized twenty-cycle multiaxial sequence derived from mixed martial arts head impacts. The result is that prior loading changes the mechanical response: strain, strain rate, and tissue-based injury probabilities grow from cycle to cycle in the damage model while a damage-free model stays nearly flat. A sympathetic reader would take the paper to be establishing that the mechanical state of the brain before an impact is as important as the kinematics of the impact itself, and that ignoring history could make single-impact models underestimate repeated-impact injury risk.","feed_headline":"Repeated hits soften brain tissue and raise injury risk","feed_subtitle":"Prior impacts amplify strain and injury probability beyond what single-impact head models predict.","key_machinery":"The load-carrying object is the Ogden–Roxburgh pseudoelastic Mullins damage model, implemented as a history-dependent multiplier on the deviatoric stress of a neo-Hookean brain tissue. The damage variable is $\\eta = 1 - \\frac{1}{r}\\,\\mathrm{erf}\\left(\\frac{W_{\\mathrm{dev,max}} - W_{\\mathrm{dev}}}{m}\\right)$, where $W_{\\mathrm{dev,max}}$ is the maximum distortional strain energy the material point has ever reached and $W_{\\mathrm{dev}}$ is the current value; the error-function form makes softening rapid at first and saturating later. Because $\\eta$ remains below one until the tissue exceeds its previous maximum, each element remembers earlier cycles, and this memory is what turns identical kinematics into different strains. The parameters $r=1.2$ and $m=0.05$ kPa come from cyclic human brain tissue tests, making the softening organ-specific rather than generic rubber behavior.","core_discovery":"The paper's central claim is that loading history itself is a mechanical variable in head injury: when brain tissue undergoes short-term Mullins-type softening, identical head kinematics produce progressively larger tissue deformation, so injury-risk estimates that ignore prior softening can understate the danger. In five-cycle single-axis simulations, the Mullins-based model showed cycle-to-cycle increases in the 95th-percentile cumulative maximum principal strain (CMPS95), the cumulative strain damage measure at a 0.15 threshold (CSDM15), strain-rate metrics, and strain-based AIS2 injury probabilities, with the largest jump between cycles 1 and 2, while the hyperelastic model stayed within a few percent of its first-cycle values and UBrIC-based probabilities, which depend only on kinematics, did not change. In the twenty-cycle multiaxial sequence, cycles of similar kinematic intensity produced different predictions depending on accumulated softening; the Mullins-based model generally produced the largest estimates in later cycles and the highest probability of at least one injury by cycle 20 (CMPS95-based 77.0% versus 50.0% and 67.9% for HE and LVHE; CSDM15-based 99.9% versus 95.3% and 88.1%). Regional results show amplification is spatially heterogeneous and direction-dependent, with white matter evolving faster than grey matter in all tested directions, gyral elements carrying higher absolute strain, and sulcal elements showing larger proportional amplification.","pith_inferences":["A testable extension the paper does not run is varying the rest interval between cycles: the model uses a 200 ms rest while recovery of cyclic softening is reported to take roughly an hour, so the predicted amplification should shrink as the interval lengthens if the mechanism is physical.","Applying the same damage-aware pipeline to football, soccer, or military blast exposure sequences could quantify how much of the variability in subconcussive outcomes is mechanical history rather than per-impact severity.","If high-rate cyclic tests show weaker or rate-dependent softening, the early-cycle dominance of LVHE and the later-cycle dominance of the Mullins-based model could reorder; a coupled nonlinear visco-hyperelastic damage formulation would be the direct next model to test."],"forward_implications":["Under identical single-axis loading, five cycles push CMPS95, CSDM15, and strain-rate metrics upward in the Mullins-based model while a damage-free hyperelastic model stays nearly flat, so repeated impacts with the same kinematics are not mechanically equivalent to one impact.","Kinematics-only criteria such as UBrIC cannot register prior softening; in the multiaxial sequence, the Mullins-based CMPS95 injury probability rose from cycle 1 to 2 and from cycle 10 to 11 even as the UBrIC-based probability fell, so repeated-exposure risk needs tissue-state-aware measures.","Damage-free HE and LVHE models can dominate early-cycle predictions, but the Mullins-based model overtakes them in later cycles and yields the highest probability of at least one injury over the 20-cycle sequence (CMPS95-based 77.0% versus 50.0% and 67.9%; CSDM15-based 99.9% versus 95.3% and 88.1%).","Regional and gyral–sulcal patterns depend on loading direction and accumulated softening: white matter evolves faster than grey matter under all directions, gyral elements carry higher absolute CMPS, and sulcal elements show larger relative amplification.","Because the bounded damage variable approaches saturation, further identical cycles produce diminishing increases, so the amplification is a rapid early effect rather than an unbounded one."],"supporting_citations":[{"why":"Supplies the pseudoelastic Mullins framework that makes the stress response depend on prior maximum strain energy.","marker":"Ogden and Roxburgh (1999)"},{"why":"Provides the cyclic human brain tissue data that calibrate the damage parameters r=1.2 and m=0.05 kPa.","marker":"Franceschini et al. (2006)"},{"why":"Supplies brain-bulk shear and bulk moduli and the MRI-based modeling approach used in the finite-element head model.","marker":"Alshareef et al. (2021)"},{"why":"Provides the mixed martial arts head-impact exposure scenario and the 20-impact count behind the loading sequences.","marker":"O'Keeffe et al. (2020)"},{"why":"Reports directional mean and standard deviation angular velocities from MMA impacts used to sample the multiaxial cycles.","marker":"Laksari et al. (2020)"},{"why":"Reports punching impact durations used to set the 15 ms pulse and its variability.","marker":"Adamec et al. (2020)"},{"why":"Supplies the Weibull injury risk functions that turn CMPS95, CSDM15, and UBrIC into AIS2 probabilities.","marker":"Wu et al. (2022)"},{"why":"Defines UBrIC and the short-duration impact window that justifies using peak angular velocity as the severity measure.","marker":"Gabler et al. (2018)"},{"why":"Defines CMPS95 and CSDM as strain-based injury metrics and recommends post-loading observation windows.","marker":"Takhounts et al. (2008)"}],"fun_headline_variants":["Repeated hits amplify brain tissue strain and injury risk","Prior impacts soften brain, raising injury risk beyond single-hit models","History matters: repeated head impacts increase injury risk","Brain softening from repeated hits elevates injury risk","Damage from prior hits amplifies strain and injury risk"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that brain tissue softens between rapid impacts the same way it does in slow laboratory cycling tests, and equally across all brain regions; if the softening is weaker, slower, or uneven at impact speeds, the predicted amplification and its spatial pattern would change.","fun_headline_variants_meta":{"raw":{"variants":["Repeated hits amplify brain tissue strain and injury risk","Prior impacts soften brain, raising injury risk beyond single-hit models","History matters: repeated head impacts increase injury risk","Brain softening from repeated hits elevates injury risk","Damage from prior hits amplifies strain and injury risk"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.0002,"raw_usage":{"total_tokens":1457,"prompt_tokens":1106,"completion_tokens":351,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":722,"completion_tokens_details":{"reasoning_tokens":275}},"tokens_in":722,"tokens_out":351,"duration_ms":4518,"temperature":1.0,"reasoning_tokens":275,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T14:21:50.748955+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run cyclic tension-compression tests on human brain tissue at strain rates near 100 per second and fit the Ogden–Roxburgh parameters to the observed second-cycle softening; if the measured $r$ and $m$ differ strongly from $r=1.2$ and $m=0.05$ kPa, the amplification predictions do not transfer to impact conditions. An organ-scale check would deliver two identical head rotations a few seconds apart in tagged-MRI experiments and compare the second strain field to the first.","supporting_citations":[{"cited_title":"A pseudo-elastic model for the Mullins effect in filled rubber , volume =","cited_arxiv_id":null,"evidence_quote":"Supplies the pseudoelastic Mullins framework that makes the stress response depend on prior maximum strain energy."}],"review_version":1}