{"id":"dfc25925-6e34-4fe8-9015-61df5f7cb957","arxiv_id":"2608.12169","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":5,"one_line_summary":"Averaging frequency power ratio decay curves across multiple transmit frequencies and steering angles reduces spectral oscillations and improves ultrasound attenuation coefficient estimation accuracy and repeatability.","lead":"Ultrasound attenuation measurements of the liver can help estimate fat content, but the signal curves used for the estimate often wiggle and distort the result. This paper shows that sending several frequencies and angles, then averaging the curves, reduces the wiggling and improves accuracy in phantoms and in a 15-patient pilot study.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Frequency/angle increments are chosen and validated on the same phantoms, so the reported ACE gains are not an independent test of the method's key parameter choice.","rationale":"The reader's weakest assumption is that the decorrelation hypothesis is only tested on the same phantoms used for validation and that generalization to in vivo liver remains an assumption. My review identifies the same selection-validation overlap as the most load-bearing concern: the parameter choice is optimized and evaluated on the same data, so the phantom component of the central claim is not an independent demonstration. The paper does have real supporting evidence: a matched transmission-count control, consistent improvements in median R², mean ACE, and CV across two phantoms, robustness with an interposed pork-belly layer, and a pilot patient study showing improved correlation with MRI-PDFF and reduced pooled SD. These are genuine reasons not to reject the method. However, the lack of independent parameter validation and the absence of significance testing for the in vivo correlation increase the risk that the reported effect size is optimistic. The appropriate disposition remains CONDITIONAL: the claim is plausible and well-supported in-sample, but should be verified with out-of-sample parameter selection or a sensitivity analysis and a larger, statistically powered clinical cohort. My verdict is therefore UNCHANGED relative to the reader's CONDITIONAL assessment.","tokens_in":17310,"tokens_out":12897,"duration_ms":131417,"concrete_test":"On the existing phantom data, run a sensitivity sweep over Δf in {0.2, 0.3, ..., 0.8 MHz} and Δθ in {2°, 4°, ..., 12°}, computing the proposed averaged-FPDC ACE and median R² for both phantoms at every grid point and comparing with conventional RFM. If the proposed method's improvement over conventional RFM persists across the full grid rather than only at the selected 0.5 MHz/8° point, the in-sample selection is not the driver. For the in vivo result, compute the bootstrap 95% confidence interval for the difference between the two correlation coefficients (R = 0.89 vs 0.83) across the 15 patients; if the interval includes zero, the reported superiority in vivo is not statistically supported.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim is that averaging FPDCs acquired at multiple transmit frequencies and steering angles suppresses interference-induced oscillations, improving ACE accuracy and repeatability. For this to hold, the chosen frequency and angle increments must produce sufficiently decorrelated FPDCs while preserving a common attenuation slope. Section III-A selects Δf = 0.5 MHz and Δθ = 8° precisely by maximizing the linearity (R²) and minimizing the NCC of averaged FPDCs on the same two calibrated phantoms that are then used in Section III-B to report the headline improvements (0.51 vs 0.56 dB/cm/MHz; 0.77 vs 0.70 dB/cm/MHz). The validation is therefore in-sample for the parameter choice: the reported accuracy gain is partly a consequence of tuning to these phantoms, not an out-of-sample test of the method's robustness. The in vivo pilot (n=15) does use the phantom-selected parameters and shows R = 0.89 vs 0.83 for correlation with MRI-PDFF, but no confidence intervals or significance testing are provided, and the paper itself acknowledges the small cohort as a limitation. If the decorrelation benefit is sensitive to Δf/Δθ or to tissue-specific scatterer statistics, the improvement could shrink or vanish outside the calibration phantoms, which would make the central clinical claim unsubstantiated.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a multi-frequency, multi-angle transmission scheme for reference frequency method (RFM)-based ultrasound attenuation coefficient estimation (ACE) in harmonic imaging. The authors hypothesize that FPDCs acquired at different transmit frequencies and steering angles exhibit weakly correlated interference-induced oscillations while sharing the same attenuation-dependent slope, so that averaging them suppresses oscillations and improves the linearity and accuracy of ACE. Two calibrated tissue-mimicking phantoms are used first to select frequency and angle spacings (Δf = 0.5 MHz, Δθ = 8°) from FPDC correlation curves, and then to show that the proposed method improves median R² (0.99 vs 0.89–0.90), mean ACE accuracy (0.51 and 0.77 vs 0.56 and 0.70–0.71 dB/cm/MHz), and inter-block CV compared with conventional RFM, both with and without an overlying pork-belly layer. A pilot patient study (n=15) reports improved correlation with MRI-PDFF (R = 0.89 vs 0.83) and a reduced pooled SD (0.0378 vs 0.0667 dB/cm/MHz).","tokens_in":17559,"tokens_out":7963,"duration_ms":68144,"significance":"If the central claim holds, the proposed method offers a simple, physically motivated improvement to RFM-based ACE, a technique already considered promising for noninvasive liver fat assessment. Strengths of the work include a fair matched-transmission-count control between the proposed and conventional acquisition schemes, consistent improvements across multiple metrics in two phantoms (including under simulated phase aberration), and a clear falsifiable hypothesis about decorrelation of spectral oscillations. The work is, however, currently limited by the fact that the key acquisition parameters were selected and evaluated on the same experimental data, and by the absence of statistical inference for the in vivo results; the quantitative magnitude of the improvement over conventional RFM is therefore not yet established out-of-sample.","major_comments":[{"comment":"The optimal frequency spacing Δf = 0.5 MHz and steering-angle spacing Δθ = 8° are selected from NCC/R² correlation curves computed on the same two calibrated phantoms (0.5 and 0.76 dB/cm/MHz) that are then used in Section III-B to report the headline improvements (e.g., mean ACE 0.51 vs 0.56 dB/cm/MHz, CV 0.84% vs 2.16%). The parameter choice is therefore optimized and evaluated on the same data, so the reported gains over conventional RFM are not an independent test of the method's key design choice. Please provide an out-of-sample validation (e.g., a separate calibration phantom, or leave-one-phantom-out selection) or demonstrate that the conclusions are robust across a plausible range of Δf and Δθ values. Without this, the abstract's quantitative claims overstate the support.","section":"III-A and III-B"},{"comment":"The in vivo result in the abstract ('R = 0.89 vs. 0.83' and reduced pooled SD) is based on only 15 patients, yet no confidence intervals, significance tests, or uncertainty estimates are reported for the correlation difference or for the pooled SD reduction. Given the small cohort, the observed differences may be within sampling variability. Please add bootstrap confidence intervals for R for each method and for the difference in R, and report the uncertainty of the pooled SD estimates (or a formal test). The paper's own discussion acknowledges the small sample size as a limitation, but the abstract's 'clinical potential' claim requires at least interval estimates to avoid overinterpretation.","section":"III-C and IV"}],"minor_comments":[{"comment":"The legends list '5.5 MHz' for some FPDCs, whereas the text consistently states the three harmonic frequencies are 4.0, 4.5, and 5.0 MHz. Please correct the figure legends for consistency.","section":"Fig. 5 and Fig. 6 legends"},{"comment":"There is a typo in Eq. (5): 'Stan dard' should be 'Standard'.","section":"Section II-E, Eq. (5)"},{"comment":"The abbreviation 'Frequecny' appears in the figure axis labels; it should be 'Frequency'.","section":"Fig. 2 labels"},{"comment":"The MRI-PDFF measurement is cited as reference [35], but reference [35] is a paper on multi-core DSP beamforming, not on the IDEAL IQ MRI sequence. Similarly, reference [34] does not appear to support the statement about 'previous work [34]' regarding median ACE. Please correct these citations.","section":"Section II-D"},{"comment":"The phrase 'In in-vitro experiments using calibrated phantoms (0.5 and 0.76 dB/cm/MHz) demonstrated' is grammatically incomplete; suggest 'In in-vitro experiments with calibrated phantoms (0.5 and 0.76 dB/cm/MHz), the proposed method demonstrated...'.","section":"Abstract and Section III-A"},{"comment":"The selection of Δf = 0.5 MHz and Δθ = 8° is described as a 'balance' between R², NCC, and bandwidth/directivity, but no formal optimization criterion is defined. Clarify whether the choice is unique or whether adjacent spacings give similar performance; this is related to the out-of-sample concern in the first major comment.","section":"Section III-A"}],"recommendation":"major_revision","confidential_remarks":"The in-sample parameter selection is the central concern: Section III-A picks Δf and Δθ from the same phantoms used in Section III-B, so the reported improvement is partly a tuning artifact. The in vivo statistical reporting is also thin (n=15, no CIs). The idea is sound and the matched-transmission-count control is a real strength, so a major revision with out-of-sample validation and better in vivo statistics could make this acceptable. I would also check the reference mis-citations before sending to production."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Here's my read. The paper applies a well-known idea—frequency and angular compounding to decorrelate speckle—to the reference frequency method (RFM) for ultrasound attenuation estimation, and it works. The phantom results are clean and the matched-transmission-count control is fair. The main caveat is that the frequency and angle spacings (Δf=0.5 MHz, Δθ=8°) are chosen from correlation curves measured on the same two phantoms that later show the headline improvements, so those numbers are partly in-sample. The patient pilot (n=15) uses those spacings and still shows a better correlation with MRI-PDFF (0.89 vs 0.83), which is some independent evidence, but it's underpowered and the correlation difference isn't significance-tested.\n\nWhat's genuinely new here is the explicit application to RFM-based ACE: averaging FPDCs across transmit frequencies and steering angles to suppress oscillations while preserving the attenuation slope. That's not a new physical principle, and the paper should have cited the compounding literature, but it's a legitimate and well-executed adaptation for a clinically relevant problem.\n\nThe soft spots in proportion: the in-sample parameter choice is the biggest one. It would be a stronger paper if they'd tested sensitivity to those parameters on an independent phantom or tissue-mimicking target. The phantom study also uses a single acquisition per measurement, so there are no inter-acquisition error bars. No code or data is provided, which limits reproducibility. The small cohort and lack of significance testing on the R values are honest limitations and clearly acknowledged.\n\nNone of this is fatal. The mechanism is physically plausible—speckle patterns do decorrelate with frequency and angle—and the phantom evidence is consistent across two phantoms and with an overlay of pork belly. The paper would benefit from revision, but it deserves a serious referee. I'd probably send it to review with a request for an independent validation of the parameter choice and some repeated acquisitions.","headline":"A solid, well-executed extension of frequency/angular compounding to RFM-based attenuation estimation, with a tuning caveat on the same phantoms.","tokens_in":18164,"tokens_out":4918,"would_cite":true,"duration_ms":43593,"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":"The paper establishes that averaging frequency power-ratio decay curves from multiple transmit frequencies and steering angles suppresses interference-induced oscillations, improving ultrasound attenuation coefficient accuracy and…","keywords":["ultrasound attenuation coefficient estimation","reference frequency method","frequency power-ratio decay curve","harmonic imaging","hepatic steatosis","oscillation suppression","multifrequency transmission","steering angle averaging"],"falsifier":"In a homogeneous phantom with known attenuation, compute the residual standard deviation of detrended FPDCs after averaging 1, 2, 4, 9, and 16 transmissions; under the weakly-correlated-oscillation hypothesis the residual should decrease roughly as the square root of the number averaged and the fitted slope should converge to the calibrated value. If the residual variance plateaus above zero or the slope systematically departs from calibration as more curves are averaged, the decorrelation premise fails.","tokens_in":17108,"feed_emoji":"🩻","tokens_out":9003,"duration_ms":77650,"temperature":0.7,"pith_summary":"Ultrasound attenuation coefficient estimation (ACE) is a promising noninvasive way to quantify liver fat, but the reference frequency method (RFM) suffers from oscillations in its frequency power-ratio decay curves (FPDCs) caused by constructive and destructive interference among backscattered echoes. This paper proposes transmitting at several harmonic frequencies and steering angles, computing one FPDC per transmit setting, and averaging them; the premise is that the oscillation patterns are weakly correlated across settings while the attenuation-dependent slope is the same. In calibrated phantoms the averaged curves are markedly more linear, the estimated attenuation coefficients move closer to the calibrated values, and inter-block variability shrinks. In a pilot patient study the correlation with MRI-PDFF rises from R=0.83 to R=0.89 and repeated-measurement variability falls, suggesting the approach makes ACE more robust. If the underlying decorrelation premise holds, the method is a simple, system-independent way to stabilize RFM without needing a reference phantom.","feed_headline":"Averaging nine ultrasound passes sharpens liver fat readings","feed_subtitle":"Three frequencies times three steering angles cancels interference oscillations, improving accuracy and repeatability.","key_machinery":"The central object is the frequency power-ratio decay curve (FPDC), formed by taking the natural logarithm of the ratio of backscattered power at adjacent frequency components and plotting it against depth; under the RFM model its slope is proportional to the attenuation coefficient. The mechanism is transmit diversity plus averaging. Each frequency-angle combination produces a different interference pattern, so the oscillatory parts of the FPDCs are weakly correlated, but the attenuation decay term, the slope, is common. Averaging across all M×N curves therefore cancels oscillations while preserving the slope, and linear regression on the averaged curve yields a more stable ACE estimate. A secondary design choice is selecting Δf = 0.5 MHz and Δθ = 8° as the largest separations that keep all transmissions within the transducer bandwidth and beam directivity.","core_discovery":"The central claim is that averaging FPDCs obtained under transmit diversity improves ACE because the interference-induced oscillations are weakly correlated between different harmonic transmit frequencies and steering angles, while the attenuation-dependent decay slope is common to all of them. With three transmit frequencies (4.0, 4.5, 5.0 MHz) and three steering angles (−8°, 0°, 8°), the average of nine FPDCs has higher linearity than any single curve: median R² rises from 0.87–0.90 to 0.99 in the two calibrated phantoms, and the estimated attenuation coefficients move closer to the calibrated values (0.51 vs 0.56 dB/cm/MHz on the 0.5 dB/cm/MHz phantom; 0.77 vs 0.70–0.71 dB/cm/MHz on the 0.76 dB/cm/MHz phantom). The same advantage appears when a pork-belly layer simulates phase aberration, and in 15 patients the median ACE correlates with MRI-PDFF at R=0.89 versus 0.83 for conventional RFM, with the pooled standard deviation of repeated measurements dropping by nearly half.","pith_inferences":["Beyond the paper, the decorrelation premise suggests an adaptive scheme: since FPDC correlation falls as Δf and Δθ grow, picking the largest separations the transducer bandwidth allows, per patient, should suppress oscillations even further than the fixed 3×3 grid tested here.","A testable extension is to apply the same averaging logic to other spectrum-based estimators such as spectral-shift or spectral-difference methods, which share the interference problem but not RFM's adjacent-frequency normalization; if the decorrelation is a property of the scattering medium rather than the estimator, those methods should improve as well.","The paper's pilot cohort is small and the decorrelation assumption was validated on phantoms rather than liver tissue; an untested consequence is that performance may be tissue-dependent, with strongly structured or coherent scatterers showing less benefit from averaging or even a biased slope."],"forward_implications":["Phantom ACE estimates become both more accurate and more precise: median FPDC R² reaches 0.99 and inter-block coefficient of variation falls to roughly 1% or below in the tested phantoms.","The benefit survives phase aberration: with a 17 mm pork-belly layer interposed, median R² stays above 0.96 versus 0.85–0.86 for conventional RFM.","In the pilot clinical data, median ACE correlates more strongly with MRI-PDFF (R=0.89 vs 0.83) and is more repeatable across ten repeated measurements, with pooled SD dropping from 0.0667 to 0.0378 dB/cm/MHz.","Because the total number of transmissions is matched to conventional RFM, the improvement is attributed to frequency-angle diversity rather than to additional acquisition time.","The method inherits RFM's system independence, so it needs no calibrated reference phantom and can be implemented on research ultrasound scanners with a sequence change."],"supporting_citations":[{"why":"Defines the conventional RFM in harmonic imaging that the proposed method is compared against and improves.","marker":"[14]"},{"why":"Introduces the system-independent spectral-normalization RFM and the FPDC slope relation that the proposed averaging feeds into.","marker":"[20]"},{"why":"Noise suppression method incorporated to prevent low-SNR distortion of the FPDC.","marker":"[16]"},{"why":"Prior method targeting non-uniform-structure oscillations, complementary to the interference oscillations this paper addresses.","marker":"[17]"},{"why":"Supplies the power-spectrum model used to derive the FPDC attenuation relation.","marker":"[19]"},{"why":"Pulse-inversion technique that enables the harmonic imaging acquisition used in both RFM variants.","marker":"[21]"},{"why":"Confirms the benefit of pulse-inversion harmonic imaging, supporting the harmonic-mode acquisition.","marker":"[22]"},{"why":"Establishes MRI-PDFF as the reference standard for the in-vivo correlation analysis.","marker":"[18]"}],"fun_headline_variants":["Nine passes cancel interference, improve liver fat ultrasound","Multi-angle multi-frequency averaging refines liver fat readings","Transmit diversity quells spectral noise for liver fat assessment","Averaging nine harmonic beams reduces liver fat variation"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that FPDCs acquired at different transmit frequencies and steering angles carry weakly correlated oscillation patterns while sharing the same attenuation-dependent slope, so that averaging cancels the oscillations without biasing the slope.","fun_headline_variants_meta":{"raw":{"variants":["Nine passes cancel interference, improve liver fat ultrasound","Multi-angle multi-frequency averaging refines liver fat readings","Transmit diversity quells spectral noise for liver fat assessment","Averaging nine harmonic beams reduces liver fat variation"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000364,"raw_usage":{"total_tokens":2054,"prompt_tokens":1129,"completion_tokens":925,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":745,"completion_tokens_details":{"reasoning_tokens":862}},"tokens_in":745,"tokens_out":925,"duration_ms":9475,"temperature":1.0,"reasoning_tokens":862,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-16T00:13:56.199275+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"In a homogeneous phantom with known attenuation, compute the residual standard deviation of detrended FPDCs after averaging 1, 2, 4, 9, and 16 transmissions; under the weakly-correlated-oscillation hypothesis the residual should decrease roughly as the square root of the number averaged and the fitted slope should converge to the calibrated value. If the residual variance plateaus above zero or the slope systematically departs from calibration as more curves are averaged, the decorrelation premise fails.","supporting_citations":[{"cited_title":"Ultrasound Attenuation Estimation in Harmonic Imaging for Robust Fatty Liver Detection,","cited_arxiv_id":null,"evidence_quote":"Defines the conventional RFM in harmonic imaging that the proposed method is compared against and improves."},{"cited_title":"Improved Ultrasound Attenuation Estimation with Non-uniform Structure Detection and Removal,","cited_arxiv_id":null,"evidence_quote":"Prior method targeting non-uniform-structure oscillations, complementary to the interference oscillations this paper addresses."},{"cited_title":"Simultaneous backscatter and attenuation estimation using a least squares method with constraints,","cited_arxiv_id":null,"evidence_quote":"Supplies the power-spectrum model used to derive the FPDC attenuation relation."},{"cited_title":"Pulse inversion Doppler: a new method for detecting nonlinear echoes from microbubble contrast agents,","cited_arxiv_id":null,"evidence_quote":"Pulse-inversion technique that enables the harmonic imaging acquisition used in both RFM variants."},{"cited_title":"Improvement of tissue harmonic imaging using the pulse -inversion technique,","cited_arxiv_id":null,"evidence_quote":"Confirms the benefit of pulse-inversion harmonic imaging, supporting the harmonic-mode acquisition."},{"cited_title":"Tada et al","cited_arxiv_id":null,"evidence_quote":"Establishes MRI-PDFF as the reference standard for the in-vivo correlation analysis."}],"review_version":1}