Pith. sign in

REVIEW 3 major objections 4 minor 125 references

FlowPET: Physics-Informed Symplectic Flow Matching for Low-Count PET Reconstruction

T0 review · 3 major / 4 minor · reviewed 2026-07-14 · grok-4.5

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read Volume-preserving symplectic transport prevents weak PET lesions from being washed out by generative reconstruction.

desk verdict Clean symplectic construction plus solid multi-dataset gains; the “immunizes lesions by Liouville” claim is stronger than the isolation evidence supports. read the letter →

arxiv 2607.11104 v1 pith:VNMTJ2JY submitted 2026-07-13 cs.CV

Zheng Zhang , Hao Tang , Yingying Hu , Zhanli Hu , Jing Qin This is my paper · ORCID
classification cs.CV
keywords low-countPETsymplecticflowmatchingHamiltoniandynamicsvolumepreservationRange-Nulldecompositionlesionrecoverymedicalinverseproblems
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

Low-count PET reconstruction is hard because standard generative models contract probability mass and therefore erase faint lesion signals along with noise. FlowPET lifts the problem into a symplectic phase space whose dynamics are generated by a separable Hamiltonian; by construction the vector field is divergence-free, so phase-space volume is conserved and weak signals cannot be numerically extinguished. Data fidelity is enforced only in the range of the PET operator while stochastic texture is injected only in the null space; the resulting flow is trained by symplectic flow matching and integrated by a leapfrog scheme that preserves volume exactly. On simulated, pediatric and clinical low-count data the method improves both global metrics and the recovery of low-contrast lesions relative to deterministic and dissipative baselines. The practical claim is that Hamiltonian structural constraints supply a geometric safeguard for inverse problems in which information conservation matters as much as noise removal.

What carries the argument

Separable Hamiltonian flow matching: the energy is split into a potential U(x) and kinetic K(p) so that the induced vector field is automatically divergence-free (Liouville), while Range-space momentum embeds the data-score restoring force and Null-space momentum injects orthogonal noise; a leapfrog integrator then realises exact discrete volume preservation.

What would settle it

On a controlled phantom of known low-contrast lesions, measure the mean and total SUV ratios along the reverse trajectory; if the ratios still collapse below unity under the symplectic leapfrog while remaining near unity under a deliberately dissipative integrator of the same trained networks, the conservation claim fails.

Watch

Extended reading notes

Core claim

Parameterizing the posterior dynamics of low-count PET as a separable Hamiltonian system on an augmented phase space yields a divergence-free vector field that conserves probability mass of weak lesions; conjugate Range-Null momentum boundaries then steer this conservative flow so that data consistency and stochastic texture generation remain orthogonal.

Load-bearing premise

That keeping phase-space volume constant, together with a single scalar that balances the Range and Null momenta, is enough to keep weak lesions from being treated as noise.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 4 minor

Summary. FlowPET reformulates low-count PET reconstruction as volume-preserving transport on a symplectic phase space. It parameterizes the posterior via a separable Hamiltonian (potential U_ψ(x) + kinetic K_φ(p)), which by construction yields a divergence-free vector field (Proposition 4.1, Eq. 5–6) and is integrated with a unit-Jacobian leapfrog scheme (Theorem A.2). Conjugate boundary conditions embed range-space data consistency into the source momentum p0 (Eq. 7) and confine stochastic injection to the null-space momentum p1 (Eq. 8), scaled by a compression factor γ. Training uses symplectic flow matching (Eq. 9–10); inference reverses the flow from an FBP prior. Experiments on BrainWeb (20 %), pediatric whole-body (1 %), and UDPET Brain (1 %) claim superior SSIM/PSNR and, crucially, better recovery of low-contrast lesions versus deterministic and dissipative generative baselines, supported by ablations (Tables 2–3) and a synthetic-lesion SUV-ratio trajectory analysis (Section 5.4, Figure 5).

Significance. If the geometric-safeguard narrative holds, the work supplies a useful inductive bias for ill-posed medical inverse problems in high-noise regimes: volume preservation plus operator-aware boundaries can protect sparse pathological signals that dissipative flows extinguish. Strengths that raise the paper above a pure engineering contribution include the explicit proofs of continuous and discrete Liouville properties (Appendix A.2, A.4), the clean Range-Null decomposition of the phase-space boundaries, multi-dataset evaluation with both simulated and clinical low-count data, component ablations, and public code. These elements make the method falsifiable and reusable. The practical impact would be improved lesion detectability under dose reduction, a clinically relevant goal.

major comments (3)
  1. [Abstract, §1, Prop. 4.1, §5.4] Abstract, §1 and Proposition 4.1 claim that the separable Hamiltonian “theoretically immunizes weak signals against probability mass collapse.” Liouville’s theorem (and the unit-Jacobian leapfrog of Theorem A.2) only guarantees that the measure of any set is conserved; it does not select which sets remain coherent lesions versus being redistributed into background. The paper itself notes that volume preservation removes the denoising effect of contraction (§4.2) and therefore introduces the Range-Null momentum boundaries (Eqs. 7–8) plus a tuned γ (Table 3). The wash-out analysis (Section 5.4) and the integrator ablation (Appendix B.1) compare full systems, never isolating pure volume preservation from the physics-informed boundaries or network capacity. The causal attribution therefore remains an empirical correlation, not a theorem; the language of theoretical immunization should be sof
  2. [Eqs. (7)–(8), (10), (14); Algorithms 1–2] Placement of the momentum compression factor γ is inconsistent between the main text and the algorithms. Main-text Eqs. (7)–(8) and (14) fold γ into the definitions of p0 and p1, while the matching loss (Eq. 10) uses the already-scaled Δp. Algorithm 1 instead defines unscaled p0/p1 and multiplies γ only inside the potential loss; Algorithm 2 likewise omits γ from the inference initialization. This discrepancy is load-bearing for reproducibility and for the claimed balance between restoring force and thermal noise; the equations and pseudocode must be aligned.
  3. [Appendix B.1, Table 4; cf. §5.4] Appendix B.1 reports that, with identical trained networks, Leapfrog improves lesion contrast only marginally over RK4 (0.8267 vs 0.8244) while SSIM/PSNR are essentially identical. Given that the central narrative rests on the geometric safeguard of symplectic integration, a more decisive isolation experiment (same Range-Null boundaries, same capacity, symplectic vs non-symplectic integrators, reported on the clinical low-count sets) is needed to quantify how much of the claimed lesion recovery is truly attributable to volume preservation rather than to the boundary design or network expressivity.
minor comments (4)
  1. [Table 1, §5.2] Table 1: on BrainWeb, FourierPET reports higher SSIM (0.9859) than FlowPET (0.9838); the text should acknowledge the second-place ranking rather than claiming uniform superiority.
  2. [§4.2.2] Notation for the system matrix adjoint and pseudo-inverse is used without explicit definition of the discrete implementation (e.g., whether A† is the exact Moore-Penrose or an FBP approximation); a short clarifying sentence would help.
  3. [Figure 5, §5.4] Figure 5 caption and surrounding text use both “Mean SUV Ratio” and “Total SUV Ratio”; the precise ROI aggregation formula should be stated once for reproducibility.
  4. [§2] Several recent PET diffusion / flow-matching baselines (e.g., the 2025 works already cited) appear only in the related-work survey; a short discussion of why their dissipative character is expected to produce the same wash-out would strengthen the positioning.

Circularity Check

1 steps flagged · score 1.0 of 10

No load-bearing circularity: divergence-free transport follows by construction from separable Hamiltonian geometry (standard Liouville), Range-Null boundaries are operator-derived, and metrics are external held-out comparisons; only mild hyperparameter tuning of gamma.

  1. fitted input called prediction [Section 5.3, Table 3 (Momentum Compression Factor gamma)]
    "Table 3 investigates the reconstruction sensitivity on Whole-Body scans across logarithmic scales, and we observe a distinct performance peak at gamma=10^{-2}. ... Consequently, we adopt gamma=10^{-2} as the canonical setting to ensure optimal transport stability."

    Gamma is selected by maximizing the same SSIM/PSNR metrics later reported as method performance. This is standard hyperparameter tuning rather than a true circular prediction (the geometric claims and Range-Null construction do not depend on the fitted value), hence only a minor, non-load-bearing instance.

full rationale

The core derivation chain is self-contained and non-circular. Proposition 4.1 and Eqs. 5-6 establish that a separable Hamiltonian H = U(x) + K(p) yields a vector field whose Jacobian has vanishing diagonal blocks, hence divergence zero by direct differentiation (proved in A.2 without external data or self-referential definitions). This is ordinary symplectic geometry (Liouville), not a claim that reduces to the target lesion-recovery metrics. Conjugate boundaries (Eqs. 7-8) are constructed from the PET operator's range-null projectors and a data-fidelity gradient; they are inputs that steer the flow, not outputs redefined as predictions. Training (SFM loss Eq. 9-10) regresses the known linear path between those fixed boundaries; inference uses the unit-Jacobian leapfrog (Theorem A.2), again a standard discrete conservation property. Empirical superiority (Table 1, Fig. 5 SUV ratios) is measured against independent baselines on held-out BrainWeb/pediatric/UDPET slices; no fitted parameter is renamed as a first-principles prediction of those metrics. The sole minor element is the scalar gamma (Table 3), chosen by validation SSIM/PSNR peak at 10^{-2}; this is ordinary hyperparameter selection, not a circular 'prediction' of a closely related quantity, and does not underwrite the geometric claims. No self-citation supplies a uniqueness theorem or ansatz that forces the result; prior author papers appear only as baselines. Score remains near zero.

Assumptions & free parameters 3 free parameters · 4 assumptions · 2 invented entities

The geometric guarantees rest on classical symplectic geometry and the PET forward model; the practical performance rests on a handful of hand-chosen scalars and architectural choices that are not derived from first principles.

free parameters (3)
  • Momentum compression factor gamma = 1e-2
    Scalar that rescales both range-space data score and null-space noise; chosen by grid search on validation SSIM/PSNR (Table 3), peak at 1e-2.
  • Kinetic and Potential network channel widths = 128 / 64 base channels
    Asymmetric U-Net capacities (128 vs 64 base channels) selected for speed/accuracy trade-off; not derived.
  • Leapfrog step count N = 100 (analysis)
    Number of reverse integration steps (typically 100 in wash-out analysis); free discretization parameter.
assumptions (4)
  • standard math Liouville's theorem: Hamiltonian vector fields are divergence-free and preserve phase-space volume
    Invoked in Section 3 and Proposition 4.1 to claim immunity to wash-out.
  • domain assumption PET measurements follow Poisson(Ax) and the system matrix A admits a well-defined range-null decomposition
    Standard PET physics used to define conjugate boundaries (Section 4.2).
  • domain assumption Linear interpolation path zt = (1-t)z0 + t z1 yields a valid target vector field for flow matching
    Taken from the Flow Matching literature (Lipman et al.) and used without further justification for the symplectic case.
  • ad hoc to paper L2 data-fidelity gradient is an adequate surrogate for the Poisson score in extreme low-count regimes
    Explicitly adopted in Eq. 7 to avoid singularities; not derived from the true likelihood.
invented entities (2)
  • Conjugate phase-space boundaries (range-space restoring momentum p0 and null-space thermal momentum p1)
    purpose: Steer the conservative Hamiltonian flow so data consistency and stochastic texture remain orthogonal
    Defined in Section 4.2.2; no independent physical measurement of these momenta exists outside the paper.
  • Symplectic Flow Matching (SFM) objective that decouples kinetic and potential matching losses
    purpose: Train the separable Hamiltonian nets without simulating trajectories
    Eqs. 9-10; a paper-specific specialization of ordinary flow matching.

how reviews work

0 comments
Cite this review

Pith. "Pith review of FlowPET: Physics-Informed Symplectic Flow Matching for Low-Count PET Reconstruction." pith.science (2026). https://pith.science/paper/VNMTJ2JY

@misc{pith2026260711104,
  author       = {Pith},
  title        = {Pith review of: FlowPET: Physics-Informed Symplectic Flow Matching for Low-Count PET Reconstruction},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/VNMTJ2JY}},
  note         = {Machine review of arXiv:2607.11104}
}
read the original abstract

Low-count Positron Emission Tomography (PET) reconstruction is severely hindered by the dissipative nature of prevailing generative models, where the inherent phase-space contraction leads to the numerical extinction (``wash-out'') of weak but diagnostically critical lesion signals. To overcome this geometric limitation, we propose \textbf{FlowPET}, a physics-informed framework that reformulates reconstruction as volume-preserving transport in a symplectic phase space. By parameterizing the posterior dynamics via a Separable Hamiltonian System, our approach guarantees a divergence-free vector field by construction, theoretically immunizing weak signals against probability mass collapse. To steer this conservative flow, we introduce conjugate boundary conditions based on the Range-Null space decomposition of the PET operator; this strictly enforces data consistency in the range space while confining stochastic uncertainty injection to the unobserved null space. We train the model via symplectic flow matching and perform inference using a symplectic leapfrog integrator. Extensive experiments on BrainWeb, clinical pediatric, and UDPET datasets demonstrate that \textbf{FlowPET} not only surpasses state-of-the-art deterministic and stochastic baselines in SSIM and PSNR but, more crucially, exhibits superior recovery of low-contrast lesions. The results confirm that imposing Hamiltonian structural constraints offers a robust geometric safeguard for medical inverse problems in high-noise regimes.

Figures

Figures reproduced from arXiv: 2607.11104 by the authors.

Figure 1
Figure 1. Conceptual comparison of generative dynamics in PET reconstruction. Standard dissipative solvers induce a contractive field (∇ · v < 0), causing weak lesion signals to be “washed out” alongside noise. In contrast, symplectic solvers lift the dynamics to a symplectic phase space, enforcing volume preservation (∇ · v = 0) to losslessly transport clinically vital signals under strict physical constraints. 1. Introducti… view at source ↗
Figure 2
Figure 2. Overview of the FlowPET framework. FlowPET formulates reconstruction as volume-preserving transport in symplectic phase space Z = X × P, driven by a Separable Hamiltonian System (Uψ, Kϕ) ensuring divergence-free dynamics. The trajectory utilizes physics-informed momentum initialization, where p0 embeds range-space data consistency and p1 injects orthogonal null-space uncertainty to recover textures strictly adhering… view at source ↗
Figure 3
Figure 3. Symplectic Leapfrog Integrator. Leveraging the sep￾arable Hamiltonian, this time-staggered Stormer-Verlet scheme ¨ interleaves half-step momentum kicks (∇Uψ) with full-step posi￾tion drifts (∇Kϕ). This discretization maintains a unit Jacobian determinant, guaranteeing exact phase-space volume preservation. Null-Space Momentum (t = 1: Orthogonal Noise Injec￾tion). At the target boundary, we isolate uncertainty. We in… view at source ↗
Figures from the paper (9 more)
Figure 4
Figure 4. Figure 4: Qualitative visualization results on the clinical pediatric dataset. The figure displays Axial cross-sections of the brain (top, indicated by the blue line) and the torso (bottom, indicated by the red line) accompanied by zoomed-in patches and error maps, alongside the…
Figure 5
Figure 5. Figure 5: Quantitative analysis of the signal wash-out phenomenon on Spherical Lesion Phantom and Synthetic Lesion BrainWeb datasets. The curves track the evolution of Mean and Total SUV Ratios during the reverse reconstruction process. sion BrainWeb (anatomically complex backgr…
Figure 6
Figure 6. Figure 6: Qualitative visualization results on the UDPET Brain (1% Count) dataset. The figure displays Axial accompanied by zoomed-in patches and error maps, alongside the central Sagittal / Coronal views. Sagittal / Coronal Axial-1 Axial-2 OSEM autoContextCNN FourierPET DGLM_u …
Figure 7
Figure 7. Figure 7: Qualitative visualization results on the clinical pediatric dataset. The figure displays Axial cross-sections of the brain (top, indicated by the blue line) and the torso (bottom, indicated by the red line) accompanied by zoomed-in patches and error maps, alongside the…
Figure 8
Figure 8. Figure 8: Qualitative visualization results on the clinical pediatric dataset. The figure displays Axial cross-sections of the brain (top, indicated by the blue line) and the torso (bottom, indicated by the red line) accompanied by zoomed-in patches and error maps, alongside the…
Figure 9
Figure 9. Figure 9: Qualitative visualization results on the UDPET Brain (1% Count) dataset. The figure displays Axial accompanied by zoomed-in patches and error maps, alongside the central Sagittal / Coronal views. D. Limitations and Future Work Limitations A critical constraint is that,…
Figure 10
Figure 10. Figure 10: Qualitative visualization results on the UDPET Brain (1% Count) dataset. The figure displays Axial accompanied by zoomed-in patches and error maps, alongside the central Sagittal / Coronal views. Sagittal / Coronal Axial-1 Axial-2 OSEM autoContextCNN FourierPET DGLM_u…
Figure 11
Figure 11. Figure 11: Qualitative visualization results on the clinical pediatric dataset. The figure displays Axial cross-sections of the brain (top, indicated by the blue line) and the torso (bottom, indicated by the red line) accompanied by zoomed-in patches and error maps, alongside th…
Figure 12
Figure 12. Figure 12: Qualitative visualization results on the clinical pediatric dataset. The figure displays Axial cross-sections of the brain (top, indicated by the blue line) and the torso (bottom, indicated by the red line) accompanied by zoomed-in patches and error maps, alongside th…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

125 extracted references · 3 linked inside Pith

  1. [1]

    Jiaqi Cui and Pinxian Zeng and Xinyi Zeng and Yuanyuan Xu and Peng Wang and Jiliu Zhou and Yan Wang and Dinggang Shen , title =

  2. [2]

    Qiyang Zhang and Yingying Hu and Yumo Zhao and Jing Cheng and Wei Fan and Debin Hu and Fuxiao Shi and Shuangliang Cao and Yun Zhou and Yongfeng Yang and Xin Liu and Hairong Zheng and Dong Liang and Zhanli Hu , title =

  3. [3]

    Xingyu Xie and Wenjie Zhao and Mu Nan and Zheng Zhang and Yaping Wu and Hairong Zheng and Dong Liang and Meiyun Wang and Zhanli Hu , title =

  4. [4]

    Contrastive Diffusion Model with Auxiliary Guidance for Coarse-to-Fine

    Zeyu Han and Yuhan Wang and Luping Zhou and Peng Wang and Binyu Yan and Jiliu Zhou and Yan Wang and Dinggang Shen , editor =. Contrastive Diffusion Model with Auxiliary Guidance for Coarse-to-Fine

  5. [5]

    Zixin Tang and Caiwen Jiang and Zhiming Cui and Dinggang Shen , title =

  6. [6]

    European Journal of Nuclear Medicine and Molecular Imaging , volume=

    A deep learning method for the recovery of standard-dose imaging quality from ultra-low-dose PET on wavelet domain , author=. European Journal of Nuclear Medicine and Molecular Imaging , volume=. 2025 , publisher=

    2025

  7. [7]

    Malcolm Hudson and Richard S

    H. Malcolm Hudson and Richard S. Larkin , title =

  8. [8]

    2007 , publisher=

    Maximum likelihood reconstruction for emission tomography , author=. 2007 , publisher=

    2007

Show all 125 references
  1. [9]

    Physics in Medicine & Biology , volume=

    Anatomically-aided PET reconstruction using the kernel method , author=. Physics in Medicine & Biology , volume=. 2016 , publisher=

    2016

  2. [10]

    Physics in Medicine & Biology , volume=

    FBP-Net for direct reconstruction of dynamic PET images , author=. Physics in Medicine & Biology , volume=. 2020 , publisher=

    2020

  3. [11]

    Image reconstruction using UNET-transformer network for fast and low-dose

    Sanaz Kaviani and Amirhossein Sanaat and Mersede Mokri and Claire Cohalan and Jean. Image reconstruction using UNET-transformer network for fast and low-dose. Comput. Medical Imaging Graph. , volume =

  4. [12]

    Jiaqi Cui and Xinyi Zeng and Pinxian Zeng and Bo Liu and Xi Wu and Jiliu Zhou and Yan Wang , title =

  5. [13]

    Rui Hu and Huafeng Liu , title =

  6. [14]

    Kim , title =

    Sanghyeok Lee and Joonmyung Choi and Hyunwoo J. Kim , title =

  7. [15]

    arXiv preprint arXiv:2407.18559 , year=

    Vssd: Vision mamba with non-causal state space duality , author=. arXiv preprint arXiv:2407.18559 , year=

    arXiv

  8. [16]

    Forty-first International Conference on Machine Learning,

    Tri Dao and Albert Gu , title =. Forty-first International Conference on Machine Learning,

  9. [17]

    Twenty New Digital Brain Phantoms for Creation of Validation Image Data Bases , journal =

    B. Twenty New Digital Brain Phantoms for Creation of Validation Image Data Bases , journal =

  10. [18]

    European journal of nuclear medicine and molecular imaging , pages=

    A cross-scanner and cross-tracer deep learning method for the recovery of standard-dose imaging quality from low-dose PET , author=. European journal of nuclear medicine and molecular imaging , pages=. 2022 , publisher=

    2022

  11. [19]

    Qiyang Zhang and Juan Gao and Yongshuai Ge and Na Zhang and Yongfeng Yang and Xin Liu and Hairong Zheng and Dong Liang and Zhanli Hu , title =

  12. [20]

    Kuang Gong and Jiahui Guan and Kyung Sang Kim and Xuezhu Zhang and Jaewon Yang and Youngho Seo and Georges El Fakhri and Jinyi Qi and Quanzheng Li , title =

  13. [21]

    CoRR , volume =

    Caiwen Jiang and Mianxin Liu and Kaicong Sun and Dinggang Shen , title =. CoRR , volume =

  14. [22]

    Howes and Alexander Hammers and Andrew P

    George Webber and Yuya Mizuno and Oliver D. Howes and Alexander Hammers and Andrew P. King and Andrew J. Reader , title =

  15. [23]

    IEEE Journal of Biomedical and Health Informatics , year=

    Multistage Diffusion Model With Phase Error Correction for Fast PET Imaging , author=. IEEE Journal of Biomedical and Health Informatics , year=

  16. [24]

    Jiaqi Cui and Yan Wang and Luping Zhou and Yuchen Fei and Jiliu Zhou and Dinggang Shen , title =

  17. [25]

    Pinxian Zeng and Xinyi Zeng and Yan Wang and Luping Zhou and Chen Zu and Xi Wu and Jiliu Zhou and Dinggang Shen , title =

  18. [26]

    Fourmer: An Efficient Global Modeling Paradigm for Image Restoration , booktitle =

    Man Zhou and Jie Huang and Chun. Fourmer: An Efficient Global Modeling Paradigm for Image Restoration , booktitle =

  19. [27]

    NeurIPS , year =

    Man Zhou and Hu Yu and Jie Huang and Feng Zhao and Jinwei Gu and Chen Change Loy and Deyu Meng and Chongyi Li , title =. NeurIPS , year =

  20. [28]

    Shihao Zhou and Jinshan Pan and Jinglei Shi and Duosheng Chen and Lishen Qu and Jufeng Yang , title =

  21. [29]

    CoRR , volume =

    Chunming He and Rihan Zhang and Fengyang Xiao and Chengyu Fang and Longxiang Tang and Yulun Zhang and Sina Farsiu , title =. CoRR , volume =

  22. [30]

    Benito and

    Daniel Feijoo and Juan C. Benito and. DarkIR: Robust Low-Light Image Restoration , booktitle =

  23. [31]

    Ning Gao and Xingyu Jiang and Xiuhui Zhang and Yue Deng , title =

  24. [32]

    Pattern Recognit

    Yuning Cui and Qiang Wang and Chaopeng Li and Wenqi Ren and Alois Knoll , title =. Pattern Recognit. , volume =

  25. [33]

    FourierMamba: Fourier Learning Integration with State Space Models for Image Deraining , booktitle =

    Dong Li and Yidi Liu and Xueyang Fu and Jie Huang and Senyan Xu and Qi Zhu and Zheng. FourierMamba: Fourier Learning Integration with State Space Models for Image Deraining , booktitle =

  26. [34]

    Johnson , title =

    Matteo Frigo and Steven G. Johnson , title =

  27. [35]

    Boyd and Neal Parikh and Eric Chu and Borja Peleato and Jonathan Eckstein , title =

    Stephen P. Boyd and Neal Parikh and Eric Chu and Borja Peleato and Jonathan Eckstein , title =. Found. Trends Mach. Learn. , volume =

  28. [36]

    Bovik and Hamid R

    Zhou Wang and Alan C. Bovik and Hamid R. Sheikh and Eero P. Simoncelli , title =

  29. [37]

    DeepPET:

    Ida H. DeepPET:. Medical Image Anal. , volume =

  30. [38]

    Neurocomputing , volume =

    Lei Xiang and Yu Qiao and Dong Nie and Le An and Weili Lin and Qian Wang and Dinggang Shen , title =. Neurocomputing , volume =

  31. [39]

    Applied Sciences , volume=

    Deep-learning-based framework for PET image reconstruction from sinogram domain , author=. Applied Sciences , volume=. 2022 , publisher=

    2022

  32. [40]

    Quantitative imaging in medicine and surgery , volume=

    LCPR-Net: low-count PET image reconstruction using the domain transform and cycle-consistent generative adversarial networks , author=. Quantitative imaging in medicine and surgery , volume=

  33. [41]

    Medical Image Analysis , volume=

    RED: Residual estimation diffusion for low-dose PET sinogram reconstruction , author=. Medical Image Analysis , volume=. 2025 , publisher=

    2025

  34. [42]

    Physica Medica , volume=

    Machine learning in quantitative PET: A review of attenuation correction and low-count image reconstruction methods , author=. Physica Medica , volume=. 2020 , publisher=

    2020

  35. [43]

    Medical Physics , volume=

    Full-count PET recovery from low-count image using a dilated convolutional neural network , author=. Medical Physics , volume=. 2020 , publisher=

    2020

  36. [44]

    Cancer Imaging , volume=

    A method to assess image quality for low-dose PET: analysis of SNR, CNR, bias and image noise , author=. Cancer Imaging , volume=. 2016 , publisher=

    2016

  37. [45]

    Magnetic Resonance Imaging Clinics , volume=

    Attenuation correction of PET/MR imaging , author=. Magnetic Resonance Imaging Clinics , volume=. 2017 , publisher=

    2017

  38. [46]

    (No Title) , year=

    Magnetic resonance imaging: physical principles and sequence design , author=. (No Title) , year=

  39. [47]

    Physics in Medicine & Biology , volume=

    Energy-selective reconstructions in x-ray computerised tomography , author=. Physics in Medicine & Biology , volume=. 1976 , publisher=

    1976

  40. [48]

    Radiology , volume=

    Susceptibility-weighted imaging: technical essentials and clinical neurologic applications , author=. Radiology , volume=. 2021 , publisher=

    2021

  41. [49]

    Information , volume=

    DIP-UP: Deep Image Prior for Unwrapping Phase , author=. Information , volume=. 2025 , publisher=

    2025

  42. [50]

    Japanese journal of radiology , volume=

    Dual-energy CT: minimal essentials for radiologists , author=. Japanese journal of radiology , volume=. 2022 , publisher=

    2022

  43. [51]

    Haipeng Chen and Sifan Wu and Zhigang Wang and Yifang Yin and Yingying Jiao and Yingda Lyu and Zhenguang Liu , title =

  44. [52]

    Hao Tang and Zechao Li and Dong Zhang and Shengfeng He and Jinhui Tang , title =

  45. [53]

    Sifan Wu and Hongzhe Zhang and Zhenguang Liu and Haipeng Chen and Yingying Jiao , title =

  46. [54]

    Tianhao Fu and Yingying Li and Xiaoqing Ye and Xiao Tan and Hao Sun and Fumin Shen and Errui Ding , title =

  47. [55]

    Xiayu Wang and Ke Ma and Ruiyun Zhong and Xinggang Wang and Yi Fang and Yang Xiao and Tian Xia , title =

  48. [56]

    NeurIPS , pages =

    Samuel Greydanus and Misko Dzamba and Jason Yosinski , title =. NeurIPS , pages =

  49. [57]

    Symplectic Recurrent Neural Networks , booktitle =

    Zhengdao Chen and Jianyu Zhang and Mart. Symplectic Recurrent Neural Networks , booktitle =

  50. [59]

    Rezende and Andrew Jaegle and S

    Peter Toth and Danilo J. Rezende and Andrew Jaegle and S. Hamiltonian Generative Networks , booktitle =

  51. [60]

    Tim Dockhorn and Arash Vahdat and Karsten Kreis , title =

  52. [61]

    Ricky T. Q. Chen and Brandon Amos and Maximilian Nickel , title =

  53. [62]

    Physics in Medicine & Biology , volume=

    Iterative reconstruction techniques in emission computed tomography , author=. Physics in Medicine & Biology , volume=. 2006 , publisher=

    2006

  54. [63]

    Ernst Hairer , title =. J. Sci. Comput. , volume =

  55. [64]

    Laurent Dinh and David Krueger and Yoshua Bengio , title =

  56. [65]

    NeurIPS , pages =

    Prafulla Dhariwal and Alexander Quinn Nichol , title =. NeurIPS , pages =

  57. [66]

    Courville , title =

    Ethan Perez and Florian Strub and Harm de Vries and Vincent Dumoulin and Aaron C. Courville , title =

  58. [67]

    Gustafsson and Zheng Zhao and Jens Sj

    Ziwei Luo and Fredrik K. Gustafsson and Zheng Zhao and Jens Sj. Image Restoration with Mean-Reverting Stochastic Differential Equations , booktitle =

  59. [69]

    Fleet and Mohammad Norouzi , title =

    Chitwan Saharia and Jonathan Ho and William Chan and Tim Salimans and David J. Fleet and Mohammad Norouzi , title =

  60. [70]

    Score-Based Generative Modeling through Stochastic Differential Equations , booktitle =

    Yang Song and Jascha Sohl. Score-Based Generative Modeling through Stochastic Differential Equations , booktitle =

  61. [71]

    Image-to-Image Translation with Conditional Adversarial Networks , booktitle =

    Phillip Isola and Jun. Image-to-Image Translation with Conditional Adversarial Networks , booktitle =

  62. [72]

    Medical Physics , volume=

    Utilizing deep learning techniques to improve image quality and noise reduction in preclinical low-dose PET images in the sinogram domain , author=. Medical Physics , volume=. 2024 , publisher=

    2024

  63. [73]

    Medical Image Anal

    Yan Wang and Yanmei Luo and Chen Zu and Bo Zhan and Zhengyang Jiao and Xi Wu and Jiliu Zhou and Dinggang Shen and Luping Zhou , title =. Medical Image Anal. , volume =

  64. [74]

    Yumin Kim and Gayoon Choi and Seong Jae Hwang , title =

  65. [75]

    IEEE Transactions on Radiation and Plasma Medical Sciences , year=

    Pseudo-MRI-guided PET image reconstruction method based on a diffusion probabilistic model , author=. IEEE Transactions on Radiation and Plasma Medical Sciences , year=

  66. [76]

    European Journal of Nuclear Medicine and Molecular Imaging , pages=

    Robust whole-body PET image denoising using 3D diffusion models: evaluation across various scanners, tracers, and dose levels , author=. European Journal of Nuclear Medicine and Molecular Imaging , pages=

  67. [77]

    IEEE Transactions on Radiation and Plasma Medical Sciences , volume=

    Bidirectional condition diffusion probabilistic models for pet image denoising , author=. IEEE Transactions on Radiation and Plasma Medical Sciences , volume=

  68. [78]

    Yaron Lipman and Ricky T. Q. Chen and Heli Ben. Flow Matching for Generative Modeling , booktitle =

  69. [79]

    Xingchao Liu and Chengyue Gong and Qiang Liu , title =

  70. [81]

    Applied Numerical Mathematics , volume=

    A history of Runge-Kutta methods , author =. Applied Numerical Mathematics , volume=

  71. [82]

    2002 , publisher=

    Classical mechanics , author=. 2002 , publisher=

    2002

  72. [83]

    Experiments

    Computer "Experiments" on Classical Fluids. II. Equilibrium Correlation Functions , author =. Phys. Rev. , volume =

  73. [84]

    Journal of mathematical imaging and vision , volume=

    Geodesic shooting for computational anatomy , author=. Journal of mathematical imaging and vision , volume=. 2006 , publisher=

    2006

  74. [85]

    International Journal of Computer Vision , volume=

    Diffeomorphic 3D image registration via geodesic shooting using an efficient adjoint calculation , author=. International Journal of Computer Vision , volume=. 2012 , publisher=

    2012

  75. [86]

    Magnetic resonance in medicine , volume=

    Symplectomorphic registration with phase space regularization by entropy spectrum pathways , author=. Magnetic resonance in medicine , volume=. 2019 , publisher=

    2019

  76. [87]

    Red: Residual estimation diffusion for low-dose pet sinogram reconstruction

    Ai, X., Huang, B., Chen, F., Shi, L., Li, B., Wang, S., and Liu, Q. Red: Residual estimation diffusion for low-dose pet sinogram reconstruction. Medical Image Analysis, 102: 0 103558, 2025

    2025

  77. [88]

    and Aich, A

    Aich, A. and Aich, A. Symplectic generative networks (sgns): A hamiltonian framework for invertible deep generative modeling. arXiv preprint arXiv:2505.22527, 2025

    2025

  78. [89]

    B., Evans, A

    Aubert - Broche, B., Griffin, M., Pike, G. B., Evans, A. C., and Collins, D. L. Twenty new digital brain phantoms for creation of validation image data bases. IEEE Trans. Medical Imaging , 25 0 (11): 0 1410--1416, 2006

    2006

  79. [90]

    A history of runge-kutta methods

    Butcher, J. A history of runge-kutta methods. Applied Numerical Mathematics, 20 0 (3): 0 247--260, 1996

    1996

  80. [91]

    Chen, R. T. Q., Amos, B., and Nickel, M. Learning neural event functions for ordinary differential equations. In ICLR . OpenReview.net, 2021

    2021

  81. [92]

    Symplectic recurrent neural networks

    Chen, Z., Zhang, J., Arjovsky, M., and Bottou, L. Symplectic recurrent neural networks. In ICLR . OpenReview.net, 2020

    2020

  82. [93]

    MCAD: multi-modal conditioned adversarial diffusion model for high-quality PET image reconstruction

    Cui, J., Zeng, X., Zeng, P., Liu, B., Wu, X., Zhou, J., and Wang, Y. MCAD: multi-modal conditioned adversarial diffusion model for high-quality PET image reconstruction. In MICCAI (7) , volume 15007 of Lecture Notes in Computer Science, pp.\ 467--477. Springer, 2024

    2024

  83. [94]

    and Nichol, A

    Dhariwal, P. and Nichol, A. Q. Diffusion models beat gans on image synthesis. In NeurIPS, pp.\ 8780--8794, 2021

    2021

  84. [95]

    NICE: non-linear independent components estimation

    Dinh, L., Krueger, D., and Bengio, Y. NICE: non-linear independent components estimation. In ICLR (Workshop) , 2015

    2015

  85. [96]

    Lifting the veil of frequency in joint segmentation and depth estimation

    Fu, T., Li, Y., Ye, X., Tan, X., Sun, H., Shen, F., and Ding, E. Lifting the veil of frequency in joint segmentation and depth estimation. In ACM Multimedia , pp.\ 944--952, 2021

    2021

  86. [97]

    T., Andrade, A., Anazodo, U

    Gan, W., Xie, H., von Gall, C., Platsch, G., Jurkiewicz, M. T., Andrade, A., Anazodo, U. C., Kamilov, U. S., An, H., and Cabello, J. Pseudo-mri-guided pet image reconstruction method based on a diffusion probabilistic model. IEEE Transactions on Radiation and Plasma Medical Sc...

    2025

  87. [98]

    a ggstr \

    H \" a ggstr \" o m, I., Schmidtlein, C. R., Campanella, G., and Fuchs, T. J. Deeppet: A deep encoder-decoder network for directly solving the PET image reconstruction inverse problem. Medical Image Anal., 54: 0 253--262, 2019

    2019

  88. [99]

    and Liu, H

    Hu, R. and Liu, H. Transem: Residual swin-transformer based regularized PET image reconstruction. In MICCAI (4) , volume 13434 of Lecture Notes in Computer Science, pp.\ 184--193. Springer, 2022

    2022

  89. [100]

    Diffusion transformer meets random masks: An advanced pet reconstruction framework

    Huang, B., He, B., Chen, Y., Liu, Z., Wang, X., Li, B., and Liu, Q. Diffusion transformer meets random masks: An advanced pet reconstruction framework. arXiv preprint arXiv:2503.08339, 2025

    2025 arXiv

  90. [101]

    Hudson, H. M. and Larkin, R. S. Accelerated image reconstruction using ordered subsets of projection data. IEEE Trans. Medical Imaging , 13 0 (4): 0 601--609, 1994

    1994

  91. [102]

    Isola, P., Zhu, J., Zhou, T., and Efros, A. A. Image-to-image translation with conditional adversarial networks. In CVPR , pp.\ 5967--5976. IEEE Computer Society, 2017

    2017

  92. [103]

    Image reconstruction using unet-transformer network for fast and low-dose PET scans

    Kaviani, S., Sanaat, A., Mokri, M., Cohalan, C., and Carrier, J. Image reconstruction using unet-transformer network for fast and low-dose PET scans. Comput. Medical Imaging Graph., 110: 0 102315, 2023

    2023

  93. [104]

    Lipman, Y., Chen, R. T. Q., Ben - Hamu, H., Nickel, M., and Le, M. Flow matching for generative modeling. In ICLR . OpenReview.net, 2023

    2023

  94. [105]

    Flow straight and fast: Learning to generate and transfer data with rectified flow

    Liu, X., Gong, C., and Liu, Q. Flow straight and fast: Learning to generate and transfer data with rectified flow. In ICLR . OpenReview.net, 2023

    2023

  95. [106]

    Deep-learning-based framework for pet image reconstruction from sinogram domain

    Liu, Z., Ye, H., and Liu, H. Deep-learning-based framework for pet image reconstruction from sinogram domain. Applied Sciences, 12 0 (16): 0 8118, 2022

    2022

  96. [107]

    o lund, J., and Sch \

    Luo, Z., Gustafsson, F. K., Zhao, Z., Sj \" o lund, J., and Sch \" o n, T. B. Image restoration with mean-reverting stochastic differential equations. In ICML , volume 202 of Proceedings of Machine Learning Research, pp.\ 23045--23066. PMLR , 2023

    2023

  97. [108]

    Manoj Doss, K. K. and Chen, J.-C. Utilizing deep learning techniques to improve image quality and noise reduction in preclinical low-dose pet images in the sinogram domain. Medical Physics, 51 0 (1): 0 209--223, 2024

    2024

  98. [109]

    Perez, E., Strub, F., de Vries, H., Dumoulin, V., and Courville, A. C. Film: Visual reasoning with a general conditioning layer. In AAAI , pp.\ 3942--3951. AAAI Press, 2018

    2018

  99. [110]

    Classical mechanics, 2002

    Safko, J., Goldstein, H., and Poole, C. Classical mechanics, 2002

    2002

  100. [111]

    J., and Norouzi, M

    Saharia, C., Ho, J., Chan, W., Salimans, T., Fleet, D. J., and Norouzi, M. Image super-resolution via iterative refinement. IEEE Trans. Pattern Anal. Mach. Intell. , 45 0 (4): 0 4713--4726, 2023

    2023

  101. [112]

    Bidirectional condition diffusion probabilistic models for pet image denoising

    Shen, C., Tie, C., Yang, Z., Zhang, N., and Zhang, Y. Bidirectional condition diffusion probabilistic models for pet image denoising. IEEE Transactions on Radiation and Plasma Medical Sciences, 8 0 (4): 0 402--415, 2024

    2024

  102. [113]

    Shepp, L. A. and Vardi, Y. Maximum likelihood reconstruction for emission tomography. IEEE Trans. Medical Imaging , 1 0 (2): 0 113--122, 2007

    2007

  103. [114]

    P., Kumar, A., Ermon, S., and Poole, B

    Song, Y., Sohl - Dickstein, J., Kingma, D. P., Kumar, A., Ermon, S., and Poole, B. Score-based generative modeling through stochastic differential equations. In ICLR . OpenReview.net, 2021

    2021

  104. [115]

    Hf-resdiff: High-frequency-guided residual diffusion for multi-dose PET reconstruction

    Tang, Z., Jiang, C., Cui, Z., and Shen, D. Hf-resdiff: High-frequency-guided residual diffusion for multi-dose PET reconstruction. In MICCAI (7) , volume 15007 of Lecture Notes in Computer Science, pp.\ 372--381. Springer, 2024

    2024

  105. [116]

    J., Jaegle, A., Racani \` e re, S., Botev, A., and Higgins, I

    Toth, P., Rezende, D. J., Jaegle, A., Racani \` e re, S., Botev, A., and Higgins, I. Hamiltonian generative networks. In ICLR . OpenReview.net, 2020

    2020

  106. [117]

    experiments

    VERLET, L. Computer "experiments" on classical fluids. ii. equilibrium correlation functions. Phys. Rev., 165: 0 201--214, 1968

    1968

  107. [118]

    and Liu, H

    Wang, B. and Liu, H. Fbp-net for direct reconstruction of dynamic pet images. Physics in Medicine & Biology, 65 0 (23): 0 235008, 2020

    2020

  108. [119]

    3d multi-modality transformer-gan for high-quality PET reconstruction

    Wang, Y., Luo, Y., Zu, C., Zhan, B., Jiao, Z., Wu, X., Zhou, J., Shen, D., and Zhou, L. 3d multi-modality transformer-gan for high-quality PET reconstruction. Medical Image Anal., 91: 0 102983, 2024

    2024

  109. [120]

    D., Hammers, A., King, A

    Webber, G., Mizuno, Y., Howes, O. D., Hammers, A., King, A. P., and Reader, A. J. Likelihood-scheduled score-based generative modeling for fully 3d PET image reconstruction. IEEE Trans. Medical Imaging , 44 0 (11): 0 4445--4456, 2025

    2025

  110. [121]

    Deep auto-context convolutional neural networks for standard-dose PET image estimation from low-dose PET/MRI

    Xiang, L., Qiao, Y., Nie, D., An, L., Lin, W., Wang, Q., and Shen, D. Deep auto-context convolutional neural networks for standard-dose PET image estimation from low-dose PET/MRI . Neurocomputing, 267: 0 406--416, 2017

    2017

  111. [122]

    Prompt-agent-driven integration of foundation model priors for low-count PET reconstruction

    Xie, X., Zhao, W., Nan, M., Zhang, Z., Wu, Y., Zheng, H., Liang, D., Wang, M., and Hu, Z. Prompt-agent-driven integration of foundation model priors for low-count PET reconstruction. IEEE Trans. Medical Imaging , 44 0 (10): 0 4073--4086, 2025

    2025

  112. [123]

    Lcpr-net: low-count pet image reconstruction using the domain transform and cycle-consistent generative adversarial networks

    Xue, H., Zhang, Q., Zou, S., Zhang, W., Zhou, C., Tie, C., Wan, Q., Teng, Y., Li, Y., Liang, D., et al. Lcpr-net: low-count pet image reconstruction using the domain transform and cycle-consistent generative adversarial networks. Quantitative imaging in medicine and surgery, 1...

    2021

  113. [124]

    P., Matzke, J., Viscione, M., Alberts, I., Meng, H., Sun, C., Zhang, M., Zhang, M., et al

    Xue, S., Guo, R., Bohn, K. P., Matzke, J., Viscione, M., Alberts, I., Meng, H., Sun, C., Zhang, M., Zhang, M., et al. A cross-scanner and cross-tracer deep learning method for the recovery of standard-dose imaging quality from low-dose pet. European journal of nuclear medicine...

    2022

  114. [125]

    Robust whole-body pet image denoising using 3d diffusion models: evaluation across various scanners, tracers, and dose levels

    Yu, B., Ozdemir, S., Dong, Y., Shao, W., Pan, T., Shi, K., and Gong, K. Robust whole-body pet image denoising using 3d diffusion models: evaluation across various scanners, tracers, and dose levels. European Journal of Nuclear Medicine and Molecular Imaging, pp.\ 1--14, 2025

    2025

  115. [126]

    PET image reconstruction using a cascading back-projection neural network

    Zhang, Q., Gao, J., Ge, Y., Zhang, N., Yang, Y., Liu, X., Zheng, H., Liang, D., and Hu, Z. PET image reconstruction using a cascading back-projection neural network. IEEE J. Sel. Top. Signal Process. , 14 0 (6): 0 1100--1111, 2020

    2020

  116. [127]

    Deep generalized learning model for PET image reconstruction

    Zhang, Q., Hu, Y., Zhao, Y., Cheng, J., Fan, W., Hu, D., Shi, F., Cao, S., Zhou, Y., Yang, Y., Liu, X., Zheng, H., Liang, D., and Hu, Z. Deep generalized learning model for PET image reconstruction. IEEE Trans. Medical Imaging , 43 0 (1): 0 122--134, 2024

    2024

  117. [128]

    Fourierpet: Deep fourier-based unrolled network for low-count pet reconstruction

    Zhang, Z., Tang, H., Hu, Y., Hu, Z., and Qin, J. Fourierpet: Deep fourier-based unrolled network for low-count pet reconstruction. arXiv preprint arXiv:2601.11680, 2026

    2026 arXiv

Pith tools

Reviewed July 14, 2026 · model on record in the stance chip above.