{"id":"24c8dc48-2c7c-4a13-bcd4-2a0316bd0af4","arxiv_id":"2412.15018","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"Respiratory self-gated TOF PET can be reconstructed into a single motion-free, attenuation-artifact-free activity image by jointly estimating gate-matched attenuation and motion with MLACF and demons.","lead":"PET scans take several minutes, and patients keep breathing during the scan, which blurs the images. A mismatch between the breath-hold CT and the moving PET data also creates artifacts, and this paper's algorithms fix both without motion-tracking hardware.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The attenuation-artifact-free claim rests on an unvalidated CT-prior dependence: gate-matched attenuation is estimated only as multiplicative corrections to a breath-hold CT sinogram, regularized toward unity by a hand-tuned gamma.","rationale":"The reader's weakest-assumption analysis correctly identifies the CT-based static attenuation sinogram and the multiplicative correction-factor model with a hand-tuned gamma prior as the load-bearing element of the attenuation estimation. My read agrees with that assessment: the phantom result (contrast 2.0 to 5.2) is quantitative and reproducible only if the CT prior is an accurate baseline and if gamma is neither too strong nor too weak. The paper's own limitations section acknowledges the static scatter assumption and the sensitivity of the ADMM parameter rho, but it does not test robustness to CT-prior errors. Because these issues are exactly what the CONDITIONAL verdict already conditions on, I do not see a reason to move the verdict. No fatal or internally inconsistent step was found; the appendix derivations are sufficiently detailed to reimplement, and the phantom experiment provides a meaningful benchmark. The secondary observation that the quantitative phantom evaluation replaced the data-driven gating signal with the actor-based signal reinforces the conditionality without overturning the paper's overall contribution.","tokens_in":13705,"tokens_out":8534,"duration_ms":63335,"concrete_test":"On the Wilhelm phantom, rerun Hybrid JRMA (and, if feasible, ADMM JRMA) with deliberately perturbed CT priors: shift the CT attenuation sinogram by one respiratory phase, scale it by 10%, and add a localized bias mimicking truncation or metal artifacts, all while varying gamma over, e.g., 0.05, 0.1, 0.2, and 0.4 of the nominal value. If liver-dome lesion contrast remains near 5.2 and the banana artifact does not reappear across these perturbations, the CT-prior concern is resolved; if contrast or artifact appearance degrades by a clinically visible margin, the central claim depends on the accuracy of the CT baseline and on gamma tuning.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim that the methods produce an activity image 'free from attenuation artifacts' depends on the correctness of the CT-based static attenuation sinogram and on the ability of the multiplicative correction-factor model (Eqs. 21-24) to capture all gate-dependent attenuation changes. In the MLACF update (Eq. 23), the estimated correction factors g_k are pulled toward 1 by the intensity prior with hand-tuned strength gamma = 0.2 mean(emission sinogram). If the CT prior is biased, truncated, or otherwise inaccurate, the only mechanism available to correct it is g_k, and that mechanism is explicitly regularized toward the prior; a biased CT baseline therefore biases the estimated gate-matched attenuation and, through the final JR OS-MLEM, the reported lesion contrasts and artifact reduction. The paper provides no robustness analysis over CT-prior errors or over gamma, and the scatter estimate is static (Section 5.2), so the same MLACF update inherits any gate-dependent scatter mismatch. A secondary evidentiary gap is that the quantitative Wilhelm phantom validation used the actor-based gating signal rather than the data-driven signal (Section 4), so the 'without external hardware' aspect is not directly tested in the only ground-truth experiment. These issues do not reveal an internal inconsistency, but they make the headline claim conditional on the CT prior and on the tuning of gamma.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript proposes two data-driven algorithms for jointly estimating activity, gate-dependent attenuation, and respiratory motion in respiratory self-gated TOF PET, with the goal of producing a single motion- and attenuation-corrected activity image without external respiratory hardware. The hybrid JRMA method estimates gate-by-gate MLACF attenuation sinograms and motion warping operators from gate-by-gate MLACF reconstructions, then performs a final JR OS-MLEM reconstruction; the ADMM JRMA method couples these estimates inside a modified ADMM framework. The methods are evaluated on a moving Wilhelm thorax phantom with a matching static reference acquisition and on three clinical FDG PET/CT datasets. The phantom experiment shows lesion-to-background contrast improving from 2.0 (no motion correction) to 5.2/5.3 with the proposed methods, matching the 5.2 contrast of the static reference, while patient datasets show reduced banana artifacts and improved lesion contrast on visual inspection.","tokens_in":14005,"tokens_out":5486,"duration_ms":51514,"significance":"If fully validated, this work would make respiratory motion and attenuation correction more practical by removing the need for external respiratory hardware and by estimating gate-matched attenuation without reconstructing attenuation images. The phantom experiment with an independent static reference is the strongest element, and the evaluation on two different scanner platforms supports generalizability. The appendix derivations are coherent, the MLACF-based updates are transparent, and the authors honestly recommend the simpler hybrid method over the more complex ADMM variant. The principal quantitative result, contrast recovery matching the static phantom, is compelling. However, the headline claims are not yet directly supported: the only ground-truth phantom experiment used the actor-based gating signal rather than the data-driven signal, and the attenuation model is a regularized multiplicative correction to a breath-hold CT sinogram, so the 'free from attenuation artifacts' claim is conditional on the CT prior and on the manually tuned regularization strength. The paper is a solid proof-of-concept but needs additional validation before the stated guarantees can be accepted.","major_comments":[{"comment":"The quantitative Wilhelm phantom experiment uses the actor-based ground-truth gating signal for gate definition rather than the PCA data-driven signal, as stated in Section 4: 'we decided to use actor-based ground-truth signal for the gate definition in the Wilhelm acquisition.' Because the headline claim is operation without external hardware, the only ground-truth experiment does not test that claim. Please repeat the phantom evaluation with data-driven gating, or quantitatively assess the sensitivity of lesion contrast and artifact reduction to the choice of gating signal.","section":"Section 4, Fig. 2"},{"comment":"The gate-dependent attenuation is modeled exclusively as nonnegative multiplicative correction factors g_k applied to the breath-hold CT attenuation sinogram, regularized toward unity with gamma = 0.2 mean(emission sinogram). The claim that the reconstructed images are 'free from attenuation artifacts' is therefore conditional on the CT-derived baseline being accurate and on gamma being appropriately tuned. No experiment perturbs the CT prior (e.g., misregistration, truncation, or bias) or varies gamma. Because a biased CT prior cannot be fully corrected by g_k under the strong regularization, please add robustness tests over both the CT prior and gamma, or explicitly reformulate the claim to reflect this dependence.","section":"Section A.2, Eqs. (21)-(24); Table 2"},{"comment":"The estimated peak displacement between gate 1 and gate 6 is 12.2 mm against the true 20 mm, a 39% underestimation, and the corresponding JR MLEM estimate is 9.3 mm. The paper offers plausible explanations (intra-gate motion, 5 mm slices, regularization) but provides no quantitative evidence that this residual motion error does not affect the final activity image or the reported contrast recovery. Please quantify residual motion after correction in the static-reference phantom, or provide a motion-error sensitivity analysis, to support the claim that the methods produce a motion-corrected image.","section":"Section 4, Fig. 4"},{"comment":"The MLACF attenuation update in Eq. (23) uses the expected scattered and random coincidences r_k in the numerator, but the scatter estimate is static and not gate-matched, as acknowledged in Section 5.2. Since the attenuation correction factors g_k are estimated jointly with the emission data, a gate-dependent scatter mismatch could bias the estimated attenuation sinograms. Please add a sensitivity analysis or a brief quantitative argument showing that smooth, static scatter does not materially bias g_k in the tested count statistics.","section":"Section 5.2; Eq. (23)"}],"minor_comments":[{"comment":"The acronym is used inconsistently: 'JMRA' appears in Figure 1 and in some text passages, while the title and most of the manuscript use 'JRMA.' Please standardize the acronym throughout.","section":"Abstract and Fig. 1"},{"comment":"The caption contains a typo: 'reconstructon' should be 'reconstruction.'","section":"Fig. 5 caption"},{"comment":"The injected activity is written as '4.25 ˙MBq/kg' with an unusual dot before 'MBq'; this should be '4.25 MBq/kg.'","section":"Section 3.2.2"},{"comment":"The text says the algorithm is divided into 'four simpler subproblems' but then lists five subproblems (subproblems 1 through 5). Please renumber or reword the introduction to this subsection.","section":"Section 2.4"},{"comment":"The replacement of the exact ADMM subproblem (25) with the denoising problem (26)-(27) using the average of inverse-warped gate images is an approximation that is not covered by standard ADMM convergence theory. Because this is a heuristic modification, it should be flagged as such in the main text rather than only in the appendix.","section":"Appendix A.3, Eqs. (25)-(27)"},{"comment":"The paper does not state whether the reconstruction code or phantom data will be made available. A data/code availability statement would improve reproducibility, especially given the complexity of the proposed pipelines.","section":"Section 3.3"}],"recommendation":"major_revision","confidential_remarks":null},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Two things you should know. First: this is a genuinely useful methods paper, not a breakthrough but a solid engineering contribution. It combines MLACF-based gate-specific attenuation correction factors with demons registration and a modified ADMM to produce a single motion-corrected, attenuation-corrected TOF PET image from data-driven respiratory gates. The Wilhelm phantom experiment is the strongest part: lesion contrast goes from 2.0 (no motion correction) to 5.2/5.3 with both proposed methods, matching the 5.2 static reference, while JR MLEM with static attenuation only reaches 3.5. That is an external benchmark, not a self-fulfilling fit. The appendix derivations are detailed enough to reimplement, and the authors honestly report limitations, including ADMM's rho sensitivity, motion underestimation (12.2 vs 20 mm), static scatter, and single-bed processing.\n\nThe soft spots are real but not fatal. The 'purely data-driven' label overstates things: for the Wilhelm phantom—the only ground-truth experiment—they used the actor-based gating signal rather than the data-driven one, so the 'without external hardware' claim is not directly tested quantitatively. The patient evaluation is three datasets, visual assessment, no error bars or repeated measurements. And the attenuation estimation is a multiplicative correction to a breath-hold CT sinogram, regularized toward unity by a hand-tuned gamma; if the CT prior is biased, the correction factor has limited room to fix it, and there is no robustness analysis over gamma or CT errors. Those are conditional-not-fatal caveats, and the authors partially acknowledge them.\n\nWho is this for? Researchers working on motion correction or joint reconstruction in PET/CT. They will get a clear, reproducible-from-the-appendix algorithm description and a useful benchmark. It deserves a serious referee—conditional acceptance with requests for uncertainty quantification on the patient data and a robustness check on gamma/CT-prior mismatch would be appropriate. I'd take it.","headline":"Useful methods paper: the phantom benchmark shows the joint MLACF+motion approach recovers static-reference lesion contrast, but the end-to-end 'no hardware' claim is only partially tested.","tokens_in":14509,"tokens_out":2164,"would_cite":true,"duration_ms":19873,"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":"Respiratory self-gated TOF PET data contain enough information to jointly estimate activity, gate-matched attenuation, and motion, yielding a single motion- and attenuation-artifact-free image without external hardware.","keywords":["respiratory motion correction","time-of-flight PET","joint activity and attenuation estimation","MLACF","ADMM","data-driven gating","attenuation artifacts","motion-compensated reconstruction"],"falsifier":"Acquire a moving phantom whose true attenuation is known at each respiratory gate (for example, by fast CT at each phase) alongside the PET list-mode data. If the jointly estimated gate-matched attenuation sinograms do not match the true phase-matched CT sinograms up to the known scaling ambiguity, or if the final activity image does not match a reconstruction that used the true gated attenuation, the central claim that the data-driven estimates are free of attenuation bias is contradicted.","tokens_in":13503,"feed_emoji":"🩻","tokens_out":6742,"duration_ms":48508,"temperature":0.7,"pith_summary":"The paper proposes two fully data-driven reconstruction algorithms for respiratory self-gated time-of-flight (TOF) PET: a simple Hybrid JRMA and a more complex ADMM-based JRMA. Both jointly estimate a single activity image, gate-matched attenuation sinograms, and respiratory motion vector fields, so that a single motion-corrected, attenuation-artifact-free image can be reconstructed without external gating hardware. In the Wilhelm thorax phantom, both methods raised a liver-dome lesion's contrast from 2.0 to 5.2, matching the static acquisition reference, while motion correction with static CT-based attenuation reached only 3.5. The methods also reduced motion blur and the liver-dome 'banana' attenuation artifact in three patient datasets acquired on a second scanner. The authors recommend the hybrid method for practical use because the ADMM variant is sensitive to its penalty parameter.","feed_headline":"Joint PET reconstruction removes motion and attenuation artifacts","feed_subtitle":"Two data-driven methods restore lesion contrast to static-scan levels without gating belts.","key_machinery":"The central objects are the forward model $\\bar{y}^k_{it}(\\lambda) = a^k_i \\big(P(S^k\\lambda)\\big)_{it} + r^k_{it}$ for each respiratory gate $k$, the factorization $a^k = g^k \\tilde{a}$ of gate-matched attenuation sinograms into the static CT-based sinogram $\\tilde{a}$ and gate-specific multiplicative correction factors $g^k$, and the MLACF update that estimates $g^k$ under an intensity prior favoring values close to 1. The factorization converts the ill-posed joint estimation of activity and attenuation into a well-scaled multiplicative correction problem, and the MLACF estimates are used both to define non-rigid image warps (via diffeomorphic demons registration of gate-by-gate reconstructions) and to provide gate-matched attenuation for the final joint reconstruction.","core_discovery":"The central claim is that respiratory self-gated TOF PET data contain enough information to estimate, purely from the emission data, the respiratory phase-matched attenuation sinograms and the non-rigid motion between respiratory gates, and to combine all gates into a single high-quality activity image. The load-bearing idea is to write each gate's attenuation sinogram as a product of a breath-hold CT-based static attenuation sinogram and nonnegative gate-specific correction factors, estimated with MLACF updates and an intensity prior that keeps the correction factors close to unity; this resolves the scale ambiguity inherent in joint activity/attenuation estimation. Attenuation and motion estimates obtained either separately (Hybrid) or jointly (ADMM) are then used in a final joint-reconstruction OS-MLEM that uses all acquired counts. The paper reports that this restores phantom lesion contrast to the static-scan value and visibly removes motion and attenuation artifacts in patients.","pith_inferences":["If the multiplicative correction model and the intensity prior hold for other attenuation mismatches (for instance, patient positioning differences in PET/MRI), the same factorization could be applied as long as a reliable baseline attenuation image is available.","The motion vector fields were underestimated (12.2 mm vs. the true 20 mm in the phantom), so residual bias in small moving-structure quantification is likely; clinical use would need to tolerate or compensate for that bias.","The method's reliance on a static scatter estimate may limit accuracy at high scatter fractions; replacing it with gated scatter estimates is a testable extension.","The hybrid pipeline's dependence on non-rigid registration of MLACF reconstructions means that registration errors propagate into the final image; learning-based motion priors could reduce that error."],"forward_implications":["A single activity image can be reconstructed from all respiratory gates, using all counts without discarding data, which minimizes noise while removing motion blur.","The data-driven PCA-based respiratory signal correlates strongly with the true motion signal (Pearson correlation 0.91 in the phantom), so external gating hardware can be omitted.","Correcting attenuation per gate removes the liver-dome banana artifact and improves lesion-to-background contrast beyond motion correction with static attenuation alone.","The comparable performance of both methods across two scanner types (Siemens mCT and GE DMI) suggests the approach generalizes to different TOF PET/CT systems.","The authors recommend Hybrid JRMA for clinical translation over ADMM JRMA because the latter is sensitive to the Lagrangian penalty parameter."],"supporting_citations":[{"why":"Supplies the MLACF algorithm used to estimate gate-dependent multiplicative attenuation correction factors.","marker":"[5]"},{"why":"Establishes that TOF PET data determine the attenuation sinogram up to a constant, the theoretical basis for joint activity-attenuation estimation.","marker":"[3]"},{"why":"Demonstrates simultaneous reconstruction of activity and attenuation in TOF PET, the foundation for the joint estimation strategy.","marker":"[4]"},{"why":"Presents prior joint reconstruction and motion estimation with a single attenuation map, which this work extends to gate-matched attenuation.","marker":"[13]"},{"why":"Shows that TOF accelerates joint reconstruction and motion estimation with misaligned attenuation, motivating the clinical TOF evaluation here.","marker":"[15]"},{"why":"Provides evidence that MLACF-based gated reconstructions yield the most accurate motion estimates compared to static or no attenuation correction.","marker":"[12]"},{"why":"Describes device-less PCA-based gating for PET/CT, the basis for the data-driven respiratory signal extraction used in this paper.","marker":"[2]"},{"why":"Introduces the anthropomorphic Wilhelm thorax phantom used to simulate realistic respiratory motion for quantitative validation.","marker":"[7]"},{"why":"Supplies the diffeomorphic demons registration algorithm used to estimate non-rigid motion warping operators.","marker":"[22]"},{"why":"Provides the iterative algorithm used to solve the ADMM subproblem combining Poisson likelihood and quadratic penalty.","marker":"[17]"}],"fun_headline_variants":["Data-driven PET clears motion and attenuation artifacts","Self-gated TOF PET jointly estimates activity, motion, attenuation","Respiratory-gated PET matches static scan contrast sans hardware","Joint PET estimation fixes motion and attenuation without belts"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The whole pipeline presumes that the breath-hold CT-based static attenuation sinogram is a valid baseline and that every respiratory attenuation change can be represented by nonnegative multiplicative correction factors pulled toward 1 by a hand-tuned intensity prior, with a static scatter estimate treated as adequate.","fun_headline_variants_meta":{"raw":{"variants":["Data-driven PET clears motion and attenuation artifacts","Self-gated TOF PET jointly estimates activity, motion, attenuation","Respiratory-gated PET matches static scan contrast sans hardware","Joint PET estimation fixes motion and attenuation without belts"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000247,"raw_usage":{"total_tokens":1585,"prompt_tokens":1032,"completion_tokens":553,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":648,"completion_tokens_details":{"reasoning_tokens":489}},"tokens_in":648,"tokens_out":553,"duration_ms":4908,"temperature":1.0,"reasoning_tokens":489,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T11:42:48.947609+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Acquire a moving phantom whose true attenuation is known at each respiratory gate (for example, by fast CT at each phase) alongside the PET list-mode data. If the jointly estimated gate-matched attenuation sinograms do not match the true phase-matched CT sinograms up to the known scaling ambiguity, or if the final activity image does not match a reconstruction that used the true gated attenuation, the central claim that the data-driven estimates are free of attenuation bias is contradicted.","supporting_citations":[{"cited_title":"ML-reconstruction for TOF-PET with simultaneous estima- tion of the attenuation factors,","cited_arxiv_id":null,"evidence_quote":"Supplies the MLACF algorithm used to estimate gate-dependent multiplicative attenuation correction factors."},{"cited_title":"Maximum-likelihood joint image reconstruction/motion estimation in attenuation-corrected respiratory gated pet/ct using a single attenuation map,","cited_arxiv_id":null,"evidence_quote":"Presents prior joint reconstruction and motion estimation with a single attenuation map, which this work extends to gate-matched attenuation."},{"cited_title":"Maximum-likelihood joint image reconstruction and motion estimation with misaligned attenuation in TOF-PET/CT,","cited_arxiv_id":null,"evidence_quote":"Shows that TOF accelerates joint reconstruction and motion estimation with misaligned attenuation, motivating the clinical TOF evaluation here."},{"cited_title":"Respiratory motion compensation for PET/CT with motion information derived from matched attenuation-corrected gated PET data,","cited_arxiv_id":null,"evidence_quote":"Provides evidence that MLACF-based gated reconstructions yield the most accurate motion estimates compared to static or no attenuation correction."},{"cited_title":"Device-less gating for PET/CT using PCA,","cited_arxiv_id":null,"evidence_quote":"Describes device-less PCA-based gating for PET/CT, the basis for the data-driven respiratory signal extraction used in this paper."},{"cited_title":"Anthropomorphic thorax phantom for cardio-respiratory motion simulation in tomographic imaging,","cited_arxiv_id":null,"evidence_quote":"Introduces the anthropomorphic Wilhelm thorax phantom used to simulate realistic respiratory motion for quantitative validation."},{"cited_title":"Diffeomorphic demons: Efficient non- parametric image registration,","cited_arxiv_id":null,"evidence_quote":"Supplies the diffeomorphic demons registration algorithm used to estimate non-rigid motion warping operators."},{"cited_title":"A modified expectation maximization algorithm for penalized likelihood estima- tion in emission tomography,","cited_arxiv_id":null,"evidence_quote":"Provides the iterative algorithm used to solve the ADMM subproblem combining Poisson likelihood and quadratic penalty."}],"review_version":1}