{"id":"1c81a84a-c114-4eeb-98c4-1f1835fe0afe","arxiv_id":"2411.11572","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"The Twin Peak Method matches two curves extracted from the frequency-wavenumber spectrum of shear waves to estimate both tissue elasticity and viscosity, with tests on simulated, ex vivo, and in vivo data.","lead":"Researchers present a new method for estimating tissue stiffness and viscosity from shear wave elastography, which uses sound waves to probe tissue. The approach may make viscosity measurement more robust to noise, potentially improving non-invasive diagnosis of liver and other diseases.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"TPM's accuracy rests on an unvalidated push-profile assumption: a small error in the Field II ARF model would shift both twin peaks and bias G0 and tau, yet no sensitivity analysis is provided.","rationale":"The reader's weakest assumption is the same as the load-bearing concern identified here. TPM is a simulation-matched inversion: it does not use the closed-form dispersion curves for estimation; it minimizes the distance between peaks from a forward-model simulation and peaks from measurements. Any unmodeled physical difference between the simulation and the measurement therefore enters directly as systematic error. The paper's in silico tests show robustness to simple additive and multiplicative Gaussian noise, and the 3D-to-2D approximation is checked, but no test perturbs the ARF push profile even though the push is a required forward-model input. The ex vivo and in vivo data were acquired through tissue paths in which acoustic attenuation and aberration are not known to match Field II in a homogeneous medium; if the actual push differs, the extracted peak loci shift and the inversion will compensate by biasing G0 and tau. The ex vivo validation with Rheospectris is genuine independent support, but the post hoc choice of frequency/wavenumber ranges after inspecting scatter and reference agreement makes the validation vulnerable to overfitting a biased forward model. The proposed sensitivity test settles the issue: perturb the push profile in silico and measure the resulting parameter shift relative to the noise-induced error. If the shift is small, the concern is resolved; if large, the accuracy claim must be qualified as conditional on push-model validity, which supports the reader's CONDITIONAL verdict.","tokens_in":15313,"tokens_out":10016,"duration_ms":101780,"concrete_test":"Use a synthetic 3D truth model with G0=2 kPa, tau=0.5 ms and the Field II push profile including 0.5 dB/cm/MHz attenuation. Run the TPM pipeline with altered push profiles: (a) attenuation 0.3 dB/cm/MHz, (b) attenuation 0.7 dB/cm/MHz, (c) focal depth shifted by +/-10%, and (d) no attenuation. Compare recovered G0 and tau to the truth and to the noise-level error bars in Fig. 6. If any perturbation changes G0 by more than about 5% or tau by more than about 10%, the twin-peak inversion is not robust to realistic push uncertainty, and the reported accuracy is conditional on an unvalidated input.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim is that TPM estimates G0 and tau by matching simulated twin peaks to measured twin peaks, so accuracy requires the forward model's peaks to coincide with the experimental peaks for the true tissue parameters. The forward model assumes a homogeneous, unbounded medium and a known ARF push profile computed with Field II. The in silico tests vary noise but never vary the push profile; the ex vivo and in vivo analyses use the same Field II push with no independent check. Because the peaks are maxima of |F(k,omega)/(rho*omega^2 - Gbar(k,omega)*k^2)|, their locations depend on the spectral shape of F. If the actual push differs from simulation due to overlying tissue attenuation, aberration, or transducer modeling error, the measured peaks shift relative to simulated peaks, and the least-squares match will trade off G0 and tau to compensate. The paper's Summary acknowledges the homogeneity limitation but does not quantify it; the Field II push assumption is not listed as a limitation at all. The ex vivo validation provides independent mechanical-testing support, but the frequency/wavenumber ranges were selected post hoc using data scatter and reference agreement, so it cannot fully arbitrate this bias. Thus the accuracy claim is only as strong as the push-profile and homogeneity assumptions.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper introduces the Twin Peak Method (TPM) for estimating tissue viscoelasticity (storage modulus G0 and relaxation time tau) from shear wave elastography (SWE) data. The method extracts two peak curves in the frequency-wavenumber domain: the f(k) peaks (frequency of maximum response for each wavenumber) and the k(f) peaks (wavenumber of maximum response for each frequency). These peaks diverge with increasing viscosity, and the paper argues that their slopes and separation are sensitive to elasticity and viscosity. TPM inverts for viscoelastic parameters by iteratively matching simulated peaks, obtained from a forward model with a Field II ARF push profile and a homogeneous unbounded medium assumption, to experimental peaks via least-squares minimization (BFGS). The method is tested on noise-free and noise-laden synthetic data, compared to AMUSE and fw2DFT approaches, validated against ex vivo porcine liver mechanical spectroscopy, and applied to in vivo human liver data.","tokens_in":15594,"tokens_out":4276,"duration_ms":42207,"significance":"If TPM performs as claimed, it offers a useful alternative for SWE-based viscoelasticity estimation that is robust to noise because it relies on peak locations rather than amplitude or attenuation measures. The paper provides several concrete strengths: independent ex vivo validation against Rheospectris mechanical spectroscopy, a systematic comparison with AMUSE and fw2DFT, a 2D-versus-3D forward model study that supports computational efficiency, and explicit acknowledgment of the homogeneity limitation in the Summary. The use of consistent forward and experimental processing (same sampling, truncation, and windowing) is methodologically sound. The twin-peak concept is novel and could stimulate further work on peak-based elastography inversion. However, the current evidence for the central accuracy claim is weakened by manual and partly post hoc selection of inversion frequency/wavenumber ranges, the unvalidated ARF push-profile assumption, and an ad hoc noise model.","major_comments":[{"comment":"The ex vivo validation is compromised by post hoc range selection. The paper reports that the initial inversion of the remaining eight datasets overpredicted viscosity, and only after reducing the frequency and wavenumber ranges to 100-200 Hz and 10-1000 m^-1 was an 'improved match' with Rheospectris obtained (Section 'Ex vivo Validation', Figures 10b and 10d). Since the same data were used to choose the ranges that produce the reported agreement, the validation is not independent. The paper acknowledges that automation of range selection is future work, but the present manuscript should at least provide a sensitivity analysis over plausible ranges and a pre-specified range-selection rule to establish that the reported accuracy is not a consequence of tuning.","section":"Ex vivo Validation and Choosing the frequency and wavenumber ranges"},{"comment":"The forward model assumes a homogeneous, unbounded medium and a Field II computed ARF push profile. The in silico tests never vary the ARF profile or the medium heterogeneity, so they cannot detect bias from these assumptions. Since the peak locations depend on the spectral shape of the forcing (Eq. 14), an error in the assumed push profile due to overlying tissue attenuation, aberration, or transducer modeling error would shift both twin peaks and bias G0 and tau. The paper needs a sensitivity analysis of the inversion to perturbations of the ARF profile and to heterogeneous inclusions, and should quantify the homogeneity approximation beyond the sentence in the Summary that it 'needs to be investigated.'","section":"Forward modeling for response computation and In silico Verification"},{"comment":"The noise robustness claim is based on a single ad hoc noise model. Equation (24) with alpha_m=0.30 and alpha_a=0.007 is introduced as 'informed by visual comparisons' with real data, but no quantitative validation is provided. The reported inversion errors (approximately 5% in G0 and 5-10% in tau) are conditional on this specific noise model, and it is unclear whether the noise parameters match the ex vivo or in vivo measurement conditions. The authors should calibrate the noise model against the actual noise statistics of the experimental data, or at least sweep over a range of alpha_m and alpha_a to demonstrate robustness to the noise assumption.","section":"Inversion with Noise-laden Synthetic Data"},{"comment":"The in silico verification is partly circular because the forward model used for inversion is the same wave equation and rheological model used to generate the synthetic data. While this verifies the numerical implementation and the inversion algorithm, it does not validate the physical assumptions of homogeneity, unboundedness, and the Field II ARF profile. The independent test of these assumptions is the ex vivo comparison, but that test is weakened by the post hoc range selection noted above. The paper should distinguish between numerical verification and physical validation, and the ex vivo section should be presented as the primary evidence for the physical assumptions.","section":"In silico Verification"}],"minor_comments":[{"comment":"In Eq. (25), the reference frequency omega_0 is described as arbitrary, but the reported G0 value is then dependent on the chosen omega_0. The paper should state explicitly that G0 is the modulus evaluated at the selected reference frequency, and that changing omega_0 rescales G0 by the factor (omega_0)^alpha.","section":"Example Inversion with Spring-Pot Model"},{"comment":"The processing details for peak extraction are described qualitatively (padding, windowing, truncation). For reproducibility, the exact parameter values (padding length, window type, any tapering) should be specified, as these affect peak locations and hence the inversion.","section":"Identifying Peaks"},{"comment":"The acronym AMUSE is used in Figures 5 and 7 without being defined in the text; please define it at first use (Attenuation Measuring Ultrasound Shearwave Elastography).","section":"Introduction"},{"comment":"The axis labels and units of the f-k domain plots in Figure 2 are not described in the text or caption; adding them would help readers understand the peak definitions.","section":"Figure 2"},{"comment":"The derivation of the scalar wave equation (1) from incompressible elastodynamics is only briefly justified; a sentence explaining why the scalar displacement in the z-direction suffices and why compressibility is neglected would be helpful.","section":"Methods, Equation (1)"}],"recommendation":"major_revision","confidential_remarks":"The core idea is promising and the ex vivo data provide some independent support, but the post hoc range selection and the unquantified ARF/homogeneity assumptions are load-bearing for the claimed accuracy. I would encourage the editor to seek a revision that adds sensitivity analyses and a pre-specified range-selection protocol, rather than rejecting the manuscript outright."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The actual new thing here is the twin-peak observation: in the f-k domain, peaks obtained by sweeping k at fixed f and peaks obtained by sweeping f at fixed k diverge with viscosity, and matching both against a forward model lets you estimate both G0 and tau. That is a genuinely different idea from the usual dispersion/attenuation inversions, and it is well motivated—peaks are maxima, so they should be more robust to noise than attenuation estimates. Equations 11 and 12 make the mechanism transparent.\n\nThe paper does a lot right. The forward model is described clearly, and they are careful to process simulated peaks exactly like measured peaks, which addresses truncation and windowing artifacts. The in silico verification is honest, with error tables and a comparison against AMUSE and fw2DFT. The ex vivo validation against independent mechanical spectroscopy is a real strength—it gives the method credibility beyond simulated data. The in vivo application is exploratory, which is fine.\n\nThe soft spots are real but not fatal. The main one, which the stress-test note correctly identifies, is the unvalidated push-profile assumption. The forward model uses a Field II ARF profile, and the inversion minimizes the difference between simulated and measured peaks. If the actual acoustic push differs from the simulation—due to overlying tissue attenuation, aberration, or transducer modeling error—both peaks shift, and the inversion will trade off G0 and tau to compensate. There is no sensitivity analysis on this. The in silico tests vary noise but never the push profile, so that part is partly circular. The ex vivo validation provides an independent check, but the frequency/wavenumber ranges were manually tuned post hoc to reduce scatter and improve agreement with the reference, which weakens its ability to arbitrate this bias. The paper does acknowledge the homogeneity limitation in the summary, but it does not list the push-profile uncertainty as a limitation at all. That is the gap I would want closed.\n\nThe range-selection issue is real but minor—they explicitly call it future work, and the method still performs reasonably without automation. Lack of shared code/data is a minor annoyance, not a scientific flaw. The comparison with AMUSE and fw2DFT is a bit apples-to-oranges because TPM assumes a rheological model and the others are model-free, but the authors note this.\n\nSo: the central idea is sound, the paper is well within the norms of the field, and the limitations are the standard ones for forward-model-based inversion. This deserves a serious referee. I would recommend sending it to peer review, and I would ask the authors to add a sensitivity study varying the push profile and to be more explicit about how the inversion ranges were chosen. If those are addressed, this could be a useful contribution to SWE methodology.","headline":"Novel twin-peak inversion for SWE viscoelasticity; the core idea is clean and the ex vivo validation is real, but the unexamined push-profile assumption and post hoc range tuning need sensitivity analysis before I'd trust clinical numbers.","tokens_in":16112,"tokens_out":2020,"would_cite":true,"duration_ms":22463,"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":"By matching twin peaks in the frequency-wavenumber domain, shear wave elastography can estimate both tissue elasticity and viscosity, even from noisy in vivo liver data.","keywords":["shear wave elastography","tissue viscoelasticity","twin-peak method","frequency-wavenumber domain","Kelvin-Voigt model","spring-pot model","acoustic radiation force","liver elastography"],"falsifier":"Run TPM on a tissue-mimicking phantom with a stiff inclusion just off the measurement line, with independent mechanical characterization of both regions; if the estimated elasticity and viscosity change systematically as the inclusion is moved closer to the line or made stiffer, the homogeneous-unbounded assumption fails. Alternatively, insert an attenuating layer between the transducer and the phantom and rerun TPM; if the results shift with layer thickness, the assumption of a known simulated push profile is the weak link.","tokens_in":15112,"feed_emoji":"🩺","tokens_out":11610,"duration_ms":103547,"temperature":0.7,"pith_summary":"Tissue viscosity is a useful disease biomarker, but shear wave elastography usually estimates it from attenuation, which is exactly the part of the signal that noise destroys. This paper proposes the twin-peak method (TPM), which instead reads viscosity from the separation between two peak curves in the frequency-wavenumber plane: the $f(k)$ peak, the frequency of maximum response for each wavenumber, and the $k(f)$ peak, the wavenumber of maximum response for each frequency. In a viscoelastic medium these curves diverge, and the paper shows the divergence is controlled by the relaxation time while the curve slope is controlled by the elastic modulus. The method estimates both parameters by matching simulated and measured peaks with a least-squares inversion, and reports accurate reconstructions on noise-free and noisy synthetic data, ex vivo porcine liver, and in vivo human liver.","feed_headline":"Twin peaks separate elasticity from viscosity in ultrasound","feed_subtitle":"A new inversion matches simulated and measured frequency-wavenumber peaks, surviving noisy in vivo liver data.","key_machinery":"The twin peaks are two ridge-location curves extracted from the magnitude of the 2D Fourier transform of the measured particle velocity: the $f(k)$ peak is the frequency of maximum response at each wavenumber, and the $k(f)$ peak is the wavenumber of maximum response at each frequency. Their slope tracks the shear-wave speed, hence elasticity, and their separation tracks the relaxation time, hence viscosity. The inversion carries these peaks through a forward model that solves the 2D viscoelastic wave equation on the same space-time grid as the measurement, using a simulated acoustic-radiation-force push profile obtained from an ultrasound transducer simulation program; simulated and measured peaks are compared with a relative least-squares objective minimized by a quasi-Newton optimizer. Applying identical sampling, truncation, windowing, and padding to simulated and measured data prevents signal-processing artifacts from shifting the peaks.","core_discovery":"The central discovery is that viscosity leaves a clean, noise-robust signature in where the response ridge peaks, not just in how wide it is. In the 2D Fourier transform of the shear-wave particle velocity, the frequency $\\omega$ that maximizes the response at a fixed wavenumber $k$ defines one peak curve, while the wavenumber that maximizes the response at a fixed frequency defines another; for a purely elastic medium the two curves coincide, and for a Kelvin-Voigt medium they separate by an amount governed by the relaxation time $\\tau$. The paper argues that matching both curves through a simulated forward model, rather than fitting the full ridge or its width, yields estimates of both the storage modulus $G_0$ and $\\tau$ that are stable under noise. Verification with 3D-simulated data inverted through a 2D forward model, validation against mechanical spectroscopy on ex vivo porcine liver, and an in vivo liver application with a bowl-shaped objective function all support the claim.","pith_inferences":["Inference: the twin-peak separation is essentially a geometric measure of how far the response ridge departs from a linear dispersion relation, so the same idea could be ported to other wave-based elastography modalities, such as magnetic resonance elastography or transient elastography, wherever two peak definitions are available.","Inference: the repeatability of peaks across replicate experiments, which the paper uses informally to set frequency and wavenumber ranges, could be turned into a formal bootstrap uncertainty estimate for the inverted elasticity and viscosity.","Inference: because TPM needs only point measurements along a line, it could be applied retrospectively to existing clinical SWE datasets collected with standard transducers, without new hardware or pulse sequences."],"forward_implications":["A single acoustic-radiation-force push can yield point estimates of both storage modulus and viscosity, not just elasticity, making viscoelasticity a more practical clinical biomarker.","Because TPM uses ridge locations rather than amplitudes, it stays accurate under the tested noise levels (30 percent multiplicative and 0.7 percent additive noise), where attenuation-based AMUSE and fw2DFT degrade.","The 2D forward model is sufficient: inverting 3D-simulated data with the 2D model introduces negligible error and cuts computation time from about 1250 seconds to about 5.5 seconds, or about 0.8 seconds with the time-domain Kelvin-Voigt solver.","The method is not tied to one rheology: inversion with the spring-pot model recovered its two parameters to within about 2 percent, supporting flexible model choice for different tissues.","The in vivo liver estimate (storage modulus 1.45 kPa, relaxation time 0.63 ms) falls in the range expected for healthy liver, and the objective function retains a clear minimum despite high noise."],"supporting_citations":[{"why":"Simulation of the acoustic radiation force push profile, which is the known input to the forward model.","marker":"[38,39]"},{"why":"AMUSE method, a model-free attenuation-based baseline that TPM is compared against and that supplies the diffraction-correction context.","marker":"[28]"},{"why":"fw2DFT method, the finite-window baseline that motivated TPM's decision to process simulated and experimental peaks identically.","marker":"[30]"},{"why":"Hyper-frequency viscoelastic spectroscopy, the independent mechanical test used to validate ex vivo porcine liver estimates.","marker":"[42]"},{"why":"Coherent plane-wave compounding, the acquisition scheme that provides the high-frame-rate shear wave motion data.","marker":"[40]"},{"why":"Autocorrelation method used to estimate particle velocity from ultrasound IQ data in the ex vivo and in vivo experiments.","marker":"[41]"}],"fun_headline_variants":["Twin peaks separate viscosity from elasticity","Shear wave twin peaks yield viscosity despite noise","New peak matching estimates tissue viscosity robustly","Twin peak approach overcomes noise for viscosity","Frequency-wavenumber twin peaks map viscoelasticity"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The forward model assumes the tissue is homogeneous and unbounded and that the push profile is known from simulation of the ultrasound transducer's acoustic field; if the real tissue is heterogeneous or the real push differs, for example because overlying tissue attenuates the acoustic beam, the simulated peaks will be systematically offset from the measured ones and the inversion will be biased.","fun_headline_variants_meta":{"raw":{"variants":["Twin peaks separate viscosity from elasticity","Shear wave twin peaks yield viscosity despite noise","New peak matching estimates tissue viscosity robustly","Twin peak approach overcomes noise for viscosity","Frequency-wavenumber twin peaks map viscoelasticity"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000606,"raw_usage":{"total_tokens":2821,"prompt_tokens":934,"completion_tokens":1887,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":550,"completion_tokens_details":{"reasoning_tokens":1817}},"tokens_in":550,"tokens_out":1887,"duration_ms":15468,"temperature":1.0,"reasoning_tokens":1817,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T18:21:44.473462+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run TPM on a tissue-mimicking phantom with a stiff inclusion just off the measurement line, with independent mechanical characterization of both regions; if the estimated elasticity and viscosity change systematically as the inclusion is moved closer to the line or made stiffer, the homogeneous-unbounded assumption fails. Alternatively, insert an attenuating layer between the transducer and the phantom and rerun TPM; if the results shift with layer thickness, the assumption of a known simulated push profile is the weak link.","supporting_citations":[{"cited_title":"Attenuation measuring ultrasound shearwave elastography and in vivo application in post-transplant liver patients","cited_arxiv_id":null,"evidence_quote":"AMUSE method, a model-free attenuation-based baseline that TPM is compared against and that supplies the diffraction-correction context."},{"cited_title":"Accounting for the Spatial Observation Window in the 2-D Fourier Transform Analysis of Shear Wave Attenuation","cited_arxiv_id":null,"evidence_quote":"fw2DFT method, the finite-window baseline that motivated TPM's decision to process simulated and experimental peaks identically."},{"cited_title":"Hyper-frequency viscoelastic spectroscopy of biomaterials","cited_arxiv_id":null,"evidence_quote":"Hyper-frequency viscoelastic spectroscopy, the independent mechanical test used to validate ex vivo porcine liver estimates."},{"cited_title":"Coherent plane-wave compounding for very high frame rate ultrasonography and transient elastography","cited_arxiv_id":null,"evidence_quote":"Coherent plane-wave compounding, the acquisition scheme that provides the high-frame-rate shear wave motion data."},{"cited_title":"Real-time two-dimensional blood flow imaging using an autocorrelation technique","cited_arxiv_id":null,"evidence_quote":"Autocorrelation method used to estimate particle velocity from ultrasound IQ data in the ex vivo and in vivo experiments."}],"review_version":1}