List of Papers Browse by Subject Areas Author List
Abstract
PET super-resolution is highly under-constrained because paired multi-resolution scans from the same subject are rarely available, and effective resolution is determined by scanner-specific physics (e.g., PSF, detector geometry, and acquisition settings). This limits supervised end-to-end training and makes purely image-domain generative restoration prone to physically inconsistent structures when anatomical and physical constraints are weak. We formulate PET super-resolution as posterior inference under heterogeneous system configurations and propose a CT-conditioned diffusion framework with physics-constrained sampling. During training, a conditional diffusion prior is learned from high-quality PET/CT pairs using cross-attention for anatomical guidance, without requiring paired LR–HR PET data. During inference, measurement consistency is enforced through a scanner-aware forward model with explicit PSF effects and gradient-based data-consistency refinement. Under both standard and OOD settings, the proposed method consistently improves experimental metrics and lesion-level clinical relevance indicators over strong baselines, while improving measurement consistency and structural fidelity.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/3390_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to the Code Repository
https://github.com/yan4243/CT-Conditioned-Diffusion-PET
Link to the Dataset(s)
N/A
BibTex
@InProceedings{YanLiu_CTConditioned_MICCAI2026,
author = { Yang, Liutao AND Wang, Zi AND Jing, Peiyuan AND Wang, Xiaowen AND Montoya-Zegarra, Javier A. AND Shi, Kuangyu AND Zhang, Daoqiang AND Yang, Guang},
title = { { CT-Conditioned Diffusion Prior with Physics-Constrained Sampling for PET Super-Resolution } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 16890},
month = {September},
page = {pending}
}
Reviews
Review #1
- Please describe the contribution of the paper
This paper presents a CT-conditioned diffusion framework with physics-constrained sampling for PET super-resolution. The method formulates the task as posterior inference by combining a diffusion-based prior with a scanner-aware forward model, explicitly incorporating PSF effects and measurement consistency. The integration of CT guidance through cross-attention and physics-based data consistency provides a principled approach to improve structural fidelity and robustness.
- Please list the major strengths of the paper: you should highlight a novel formulation, an original way to use data, demonstration of clinical feasibility, a novel application, a particularly strong evaluation, or anything else that is a strong aspect of this work. Please provide details, for instance, if a method is novel, explain what aspect is novel and why this is interesting.
The problem is well-motivated and relevant, as PET super-resolution is inherently ill-posed and depends on scanner-specific physics. The use of CT as conditioning is practical and clinically accessible, providing useful anatomical guidance. The framework combines diffusion priors with physics-based constraints, which is aligned with current best practices in inverse problems. The paper includes comparisons with multiple state-of-the-art methods, covering both supervised and diffusion-based approaches. The method demonstrates consistent improvements in both standard and OOD settings, as well as lesion-level evaluation, which strengthens its empirical impact.
- Please list the major weaknesses of the paper. Please provide details: for instance, if you state that a formulation, way of using data, demonstration of clinical feasibility, or application is not novel, then you must provide specific references to prior work.
Some implementation details are not sufficiently described, which may limit reproducibility. The evaluation is based on simulated degradations and 2D slice-wise experiments, which may not fully reflect real-world clinical scenarios. Certain claims (e. g. , hallucination reduction) are not directly supported by dedicated evaluation.
- Please rate the clarity and organization of this paper
Satisfactory
- Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.
The submission does not provide sufficient information for reproducibility.
- Based on your review and your understanding of the MICCAI Scientific Code of Ethics, do you believe this submission may involve a potential ethics concern or violation?
N/A
- Optional: If you have any additional comments to share with the authors, please provide them here. Please also refer to our Reviewer’s guide on what makes a good review and pay specific attention to the different assessment criteria for the different paper categories: https://conferences.miccai.org/2026/en/REVIEWER-GUIDELINES.html
1.Reproducibility and implementation details: The paper would benefit from more detailed descriptions of key components (e. g. , training setup, parameter choices, and inference settings). In addition, it is unclear whether code will be publicly released. Providing implementation details or code availability would significantly improve reproducibility. 2.Clarification of PSF modeling and scheduling: The proposed PPCR strategy relies on progressive PSF scheduling; however, it is not entirely clear how the PSF strength is parameterized or controlled across iterations. The current ablation mainly compares presence/absence of PSF, but a more detailed analysis (e. g. , varying PSF strength or scheduling strategies) would strengthen the justification. 3.2D formulation limitation: The method is implemented in a 2D slice-wise manner, which simplifies the problem but may limit applicability in realistic 3D PET reconstruction scenarios. A discussion on how the method could extend to 3D would be valuable. 4.Hallucination claim requires further support: The abstract mentions reduction of hallucination artifacts, but this is not explicitly evaluated in the results. Including more targeted analysis (e. g. , lesion-level false positives/negatives or uncertainty estimation) would better support this claim. 5.Clarity of terminology and abbreviations: Some abbreviations (e. g. , DC) are introduced without clear definition at first occurrence. For clarity, it would be helpful to provide the full term when first introduced.
- Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making.
(5) Accept — should be accepted, independent of rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
Overall, this paper presents a technically sound and well-structured framework for PET super-resolution, combining diffusion-based priors with physics-constrained sampling. The approach is well-motivated and demonstrates strong empirical performance across multiple settings. The identified limitations mainly concern clarity, reproducibility, and evaluation completeness rather than fundamental flaws.
- Reviewer confidence
Confident but not absolutely certain (3)
- [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.
Accept
- [Post rebuttal] Please justify your final decision from above.
The rebuttal addresses most of my concerns at the clarification level. The explanation of PSF-aware data consistency and PPCR is helpful, and the response clarifies the distinction from ScorePET as well as the notation in the likelihood gradient.
Review #2
- Please describe the contribution of the paper
The paper formulates PET super-resolution as posterior inference under heterogeneous system configurations and propose a CT-conditioned diffusion framework with physics-constrained sampling. PSF modeling and Progressive Physics-Constrained Refinement are also used in the sampling process of diffusion reconstruction.
- Please list the major strengths of the paper: you should highlight a novel formulation, an original way to use data, demonstration of clinical feasibility, a novel application, a particularly strong evaluation, or anything else that is a strong aspect of this work. Please provide details, for instance, if a method is novel, explain what aspect is novel and why this is interesting.
1.The paper uses high-quality CT image as a guide for PET reconstruction, and apply cross-attention for CT and PET. This method may improve the anatomical results for PET reconstruction.
2.It incorporates comprehensive physics constraints in reconstruction, including PSF modeling and Progressive Physics-Constrained Refinement, which improves the final reconstruction quality via ablation study.
- Please list the major weaknesses of the paper. Please provide details: for instance, if you state that a formulation, way of using data, demonstration of clinical feasibility, or application is not novel, then you must provide specific references to prior work.
The paper compares its results with some previous diffusion model based reconstruction methods (such as ScorePET), but it didn’t make a clear comparison on how the proposed method is different from previous diffusion reconstruction methods.
- Please rate the clarity and organization of this paper
Good
- Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.
The submission does not mention open access to source code or data, but provides a clear and detailed description of the algorithm to ensure reproducibility.
- Based on your review and your understanding of the MICCAI Scientific Code of Ethics, do you believe this submission may involve a potential ethics concern or violation?
N/A
- Optional: If you have any additional comments to share with the authors, please provide them here. Please also refer to our Reviewer’s guide on what makes a good review and pay specific attention to the different assessment criteria for the different paper categories: https://conferences.miccai.org/2026/en/REVIEWER-GUIDELINES.html
1.I would suggest the authors to write their innovations compared to ScorePET more clearly. 2.Suggest the authors to show the reconstruction time comparison for different methods.
- Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making.
(4) Weak Accept — marginally above the acceptance threshold, but would not mind if rejected, dependent on rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
The paper formulates PET super-resolution as posterior inference under heterogeneous system configurations and propose a CT-conditioned diffusion framework with physics-constrained sampling. PSF modeling and Progressive Physics-Constrained Refinement are also used in the sampling process of diffusion reconstruction. The proposed method outperforms previous reconstruction methods in synthetic testing set.
Although the paper compares its results with some previous diffusion model based reconstruction methods (such as ScorePET), it didn’t make a clear comparison on how the proposed method is different from previous diffusion reconstruction methods. Those parts should be clearly stated to show the innovation.
- Reviewer confidence
Very confident (4)
- [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.
Accept
- [Post rebuttal] Please justify your final decision from above.
N/A
Review #3
- Please describe the contribution of the paper
The paper’s main contribution is a PET super-resolution framework that combines a CT-conditioned diffusion prior with physics-constrained sampling.
- Please list the major strengths of the paper: you should highlight a novel formulation, an original way to use data, demonstration of clinical feasibility, a novel application, a particularly strong evaluation, or anything else that is a strong aspect of this work. Please provide details, for instance, if a method is novel, explain what aspect is novel and why this is interesting.
The paper proposes a diffusion sampler for PET resolution enhancement. The authors propose a PSF scheduling for resolution enhancement.
- Please list the major weaknesses of the paper. Please provide details: for instance, if you state that a formulation, way of using data, demonstration of clinical feasibility, or application is not novel, then you must provide specific references to prior work.
1) Equation 4: The authors can explicitly use a linear operator for PET system matrix as it is strictly defined by physical meaning.
2) I believe Eq. 6 is wrong. It should be system matrix transpose at the beginning.
3) PPCR is not clearly introduced. It is unclear to me how PSF is scheduled during the inference. The mechanism of its improvement on resolution is also unclear.
4) The implementation of CT cross-attention is unclear. This is important for the proposed method to leverage structural information from CT.
5) In the experiment, the details of the simulation are unclear. It is important for this work to have a specific application scenario where the low-resolution scanner that needs enhancement is clearly stated. Also, a 8-mm image domain PSF is too large for most modern scanners.
6) The real image-domain PSF is heterogeneous. This should be discussed in a study for resolution enhancement.
- Please rate the clarity and organization of this paper
Poor
- Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.
The submission does not provide sufficient information for reproducibility.
- Based on your review and your understanding of the MICCAI Scientific Code of Ethics, do you believe this submission may involve a potential ethics concern or violation?
N/A
- Optional: If you have any additional comments to share with the authors, please provide them here. Please also refer to our Reviewer’s guide on what makes a good review and pay specific attention to the different assessment criteria for the different paper categories: https://conferences.miccai.org/2026/en/REVIEWER-GUIDELINES.html
N/A
- Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making.
(2) Reject — should be rejected, independent of rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
The method is not clearly introduced, which raises some doubts about the reproducibility of the paper. The experiment design is also too simple and unrealistic for real scenarios.
- Reviewer confidence
Very confident (4)
- [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.
Reject
- [Post rebuttal] Please justify your final decision from above.
The authors have partly addressed my concerns regarding PSF schedule, notation, and simulation. Although the method has some merits, the paper may need major revision to clarify many parts, including PSF schedule and physics-based PSF simulation, to justify the scientific merit.
Author Feedback
Our method couples a CT-conditioned diffusion prior with scanner-aware PSF/DC sampling for PET SR. R1 considered it technically sound with standard/OOD and lesion-level gains; R2 noted CT guidance and PSF/PPCR improve quality in ablations; the meta-review appreciated the physics-constrained reconstruction. Q&A: 1.PSF/PPCR mechanism, CT attention, and reproducibility (Meta, R1.1-2, R3.3-4, R3.6). In DC, a candidate HR PET z is blurred by H and projected/rebinned by G_lr; the Poisson residual gives the likelihood gradient. The forward model is simplified but physics-informed, including PSF blur, projection/rebinning, and Poisson statistics, not simple image post-processing. PSF does not create resolution alone: diffusion proposes HR candidates, while PSF-aware DC keeps those whose degraded predictions match the measurement. Strong PSF constraints too early can over-restrict reverse diffusion before stable anatomy forms; PPCR therefore applies full physical constraints near the final high-fidelity stage. Table 3 shows removing DC, PSF, or PPCR degrades PSNR/SSIM/NMSE, supporting this mechanism. Since H is inference-time only, measured scanner PSFs, including spatially varying/heterogeneous PSFs with adjoints, can be used without retraining. CT cross-attention uses PET features as queries and multi-scale CT features as keys/values, outperforming CT concatenation in Table 3.To further support reproducibility, code and configurations will be released upon acceptance.
2.Simulation realism and application scenario (R1.3, R3.5-6). Real paired LR-HR PET is rarely available, while quantitative SR evaluation needs a reference; controlled simulation therefore enables metrics and fair comparison. Our LR generation is physics-based rather than plain downsampling: it includes PSF blur, projection geometry, angular/radial rebinning, Poisson statistics, and dose reduction. Thus 8/12 mm denote effective image-domain degradation settings for SRx4/OOD SRx6, not intrinsic detector resolution. The target application is resolution enhancement for legacy PET scanners with limited effective resolution: a high-quality PET/CT prior learned without LR-HR pairs can be combined with the available scanner operator/PSF to improve resolution and diagnostic utility. Standard/OOD gains and lesion-level SUV/NMSE gains support clinical fidelity. As this is a methodology paper, validation with measured scanner PSFs and fully 3D clinical pipelines is outside the scope of the current study.
3.ScorePET difference, hallucination, and runtime (Meta, R1.4, R2.1-2). ScorePET uses a PET score prior and PET likelihood/DC for reconstruction. In contrast, our work addresses PET SR under heterogeneous scanner-dependent LR observations with a different posterior formulation using CT conditioning, PSF-aware degradation, and PPCR-based physics-constrained sampling.Table 2 directly addresses R2’s comparison request: ours outperforms ScorePET in SRx4/OOD SRx6 on all metrics (PSNR +2.87/+2.81 dB). Hallucination is reduced because generated HR PET is constrained so that physically degraded outputs match measured LR PET and CT anatomy guides plausible structure; this is supported by Fig. 4 lesion SUV/NMSE and Table 3 consistency ablations. Runtime is 2.38 s, close to ScorePET/PETDM (2.08/1.96 s) and slower than single-pass UNet (0.004 s). With current computing power rapidly increasing, this moderate diffusion-level cost has translational value when it brings accuracy, lesion-fidelity, and robustness gains.
4.Notation/terminology (R1.5, R3.1-2). Eq. 4 defines ybar_lr(z)=G_lr H z+b_lr. By the chain rule, Eq. 6 applies the adjoint of the linear measurement operator G_lr H, i.e., H^T G_lr^T, to the Poisson residual, while b_lr is constant and does not affect the gradient. We will revise the notation to make this explicit, and define DC/abbreviations at first occurrence. These revisions do not affect implementation.
Meta-Review
Meta-review #1
- Your recommendation
Invite for Rebuttal
- Please justify your decision. In case you deviate from the reviewers’ recommendations, explain in detail the reasons why. In case of an invitation for rebuttal, clarify which points are important to address in the rebuttal.
The paper proposed a diffusion model-based method for PET super-resolution, utilizing CT as an anatomical prior in pre-training, and adding in physics constraints through PSF modeling. A mixture of review scores was received. The physics constraints in reconstruction was appreciated, but across the reviews more concerns arose, including clarifying PSF implementation details, its mechanism of improving resolution, and feasibility in implementing real image-domain PSF as it can be heterogeneous. These key questions might need to be sufficiently addressed during rebuttal, along with clarifying more implementation details as requested by R1 and R3.More discussion for results including hallucination reduction and computational cost might also be needed.
- After you have reviewed the rebuttal and updated reviews, please provide your recommendation based on all reviews and the authors’ rebuttal.
Accept
- Please justify your recommendation.
The authors have sufficiently addressed concerns raised in the previous round of revision, including physics awareness in diffusion process, forward model, and in data simulation. Generally the paper is interesting and the physics constraints in diffusion model are worth discussing in the community. The authors are encouraged to include more details of PSF schedule and simulation details as asked by R3 in their camera-ready version.
Meta-review #2
- After you have reviewed the rebuttal and updated reviews, please provide your recommendation based on all reviews and the authors’ rebuttal.
Accept
- Please justify your recommendation.
The three reviewers have recognized some novelty in the proposed PET superresolution method, which is based on a diffusion-based prior and physics-based sampling. The main limitations come from the 2D simulated slice-wise evaluation (R1) and model simplifications and clarity issues (R3). Despite these drawbacks, the framework seems interesting enough in its current state to spark discussion within the PET reconstruction community.
Meta-review #3
- After you have reviewed the rebuttal and updated reviews, please provide your recommendation based on all reviews and the authors’ rebuttal.
Accept
- Please justify your recommendation.
After rebuttal, the paper has two positive reviews (accept) and one remaining reject. The rebuttal reasonably addresses the main concerns from R1 and R2, particularly by clarifying the PSF-aware data consistency formulation, PPCR mechanism, distinction from ScorePET, likelihood-gradient notation, and planned code release. However, R3 remains unconvinced and continues to list several weaknesses: insufficient clarity in PSF scheduling, simulation realism, and reproducibility. Overall, I view the rebuttal as mostly successful: several concerns appear to be presentation/clarification issues rather than fundamental flaws. I would recommend acceptance, while noting that the final paper should improve clarity including the description of the PSF/PPCR implementation, CT cross-attention and simulation assumptions.