CT-Conditioned Diffusion Prior with Physics-Constrained Sampling for PET Super-Resolution

Liutao Yang; Zi Wang; Peiyuan Jing; Xiaowen Wang; Javier A. Montoya-Zegarra; Kuangyu Shi; Daoqiang Zhang; Guang Yang

cs.CVMar 14, 2026

CT-Conditioned Diffusion Prior with Physics-Constrained Sampling for PET Super-Resolution

Authors: Liutao Yang, Zi Wang, Peiyuan Jing, Xiaowen Wang, Javier A. Montoya-Zegarra, Kuangyu Shi, Daoqiang Zhang, Guang Yang

Organizations: Department of Bioengineering and Imperial-X, Imperial College London, London, U.K. · College of Artificial Intelligence, Nanjing University of Aeronautics and Astronautics, Nanjing, China. · School of Engineering, Zurich University of Applied Sciences, Winterthur, Switzerland. · Department of Nuclear Medicine, Inselspital, University of Bern, Bern, Switzerland. ·

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 hallucinated 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 reducing hallucination artifacts and improving structural fidelity.

Figures & tables

Figure 1: Conceptual motivation and method positioning for PET super-resolution. (a) PET super-resolution is ill-posed due to scanner-dependent degradation and limited paired data. (b) Supervised CNN mapping depends on paired LR–HR data and tends to over-smooth details. (c) Image-domain diffusion can recover texture but may introduce hallucinations without structural/physics constraints. (d) Our method combines CT-conditioned priors with scanner-aware, PSF-aware consistency for faithful and physically consistent reconstruction.

Figure 2: Overview of the proposed CT-guided diffusion framework. Pretraining (a): learn a CT-conditioned diffusion prior from high-quality PET/CT. Inference (b): start from noisy PET, iteratively denoise, and apply measurement-domain data-consistency updates. PPCR: progressively enforces physics constraints (late activation and coarse-to-fine PSF) to produce a scanner-consistent high-resolution PET output.

Setting	FWHM	Dose	Angular	Radial	Target Spacing	SR Factor
Standard	8 mm	10%	$\times$ 2	$\times$ 2	8 mm	$\times$ 4
OOD	12 mm	5%	$\times$ 3	$\times$ 2	12 mm	$\times$ 6

Table 1: Degradation settings for standard and out-of-distribution (OOD) evaluation.

	8 mm (SR $\times$ 4)			12 mm (SR $\times$ 6)
Method	PSNR $\uparrow$	SSIM $\uparrow$	NMSE $\downarrow$	PSNR $\uparrow$	SSIM $\uparrow$	NMSE $\downarrow$
LR	36.63 $\pm$ 4.56 ∗	0.887 $\pm$ 0.04 ∗	0.09 $\pm$ 0.05 ∗	34.98 $\pm$ 3.75 ∗	0.849 $\pm$ 0.05 ∗	0.09 $\pm$ 0.05 ∗
DeepSR [ 5 ]	42.49 $\pm$ 3.70 ∗	0.957 $\pm$ 0.02 ∗	0.03 $\pm$ 0.03 ∗	37.27 $\pm$ 3.31 ∗	0.896 $\pm$ 0.04 ∗	0.06 $\pm$ 0.05 ∗
UNet [ 20 ]	39.50 $\pm$ 4.55 ∗	0.929 $\pm$ 0.05 ∗	0.06 $\pm$ 0.07 ∗	36.35 $\pm$ 3.81 ∗	0.877 $\pm$ 0.07 ∗	0.09 $\pm$ 0.10 ∗
SRCNN [ 21 ]	42.75 $\pm$ 3.70 ∗	0.960 $\pm$ 0.02 ∗	0.03 $\pm$ 0.03 ∗	37.33 $\pm$ 3.32 ∗	0.899 $\pm$ 0.04 ∗	0.06 $\pm$ 0.05 ∗
SRDPM [ 7 ]	40.30 $\pm$ 3.29 ∗	0.923 $\pm$ 0.03 ∗	0.04 $\pm$ 0.05 ∗	36.74 $\pm$ 3.08 ∗	0.877 $\pm$ 0.05 ∗	0.08 $\pm$ 0.05 ∗

Table 2: Quantitative comparison under standard (8 mm, SR

\times

4) and OOD (12 mm, SR

\times

6) settings. Best results are bolded ; second-best are underlined .

Figure 3: Qualitative comparison under the standard setting (8 mm, SR $\times$ 4).

	Components				Metrics
Variant	CT cond.	DC	PSF	Sched.	PSNR $\uparrow$	SSIM $\uparrow$	NMSE $\downarrow$
w/o DC	CrossAttn				43.92 $\pm$ 3.45	0.962 $\pm$ 0.02	0.018 $\pm$ 0.02
w/o PSF	CrossAttn	✓		✓	45.11 $\pm$ 3.26	0.971 $\pm$ 0.02	0.014 $\pm$ 0.01
w/o PPCR	CrossAttn	✓	✓		45.78 $\pm$ 3.18	0.975 $\pm$ 0.01	0.012 $\pm$ 0.01
CT Concat	Concat	✓	✓	✓	45.94 $\pm$ 3.15	0.976 $\pm$ 0.01	0.011 $\pm$ 0.01
Ours	CrossAttn	✓	✓	✓	46.84 $\pm$ 3.08	0.980 $\pm$ 0.01	0.010 $\pm$ 0.01

Table 3: Ablation study under the standard degradation setting (8 mm, SR

\times

4). “CT cond.” indicates the conditioning strategy. Best results are bolded .

Figure 4: Lesion-level quantitative comparison on two lung cancer cases. Each lesion reports $\Delta$ SUVmax, $\Delta$ SUVmean, and lesion NMSE within the lesion mask; lower values indicate better consistency.

Medical Image Super-Resolution

Medical image super-resolution reconstructs fine spatial detail from low-resolution scans to support clearer anatomy and more useful clinical measurements.

See all 57 papers

PhyDiCT: Plug-and-Play CT Reconstruction from Sparse X-Rays via Differentiable Rendering and Strong Priors
Oct 7, 2026Weicheng Dai, Shantanu Ghosh, Kayhan BatmanghelichSparse-View CT Reconstruction 3D Medical Imaging
Department of Electrical and Computer Engineering, Boston University, Boston, MA, USA
AnF-DiffPET: Anatomy- and Frequency-Guided Diffusion for PET/CT Denoising
Jul 1, 2026Xuepeng Liu, Ruili Li, Zetong Liu +5Diffusion Models for Image Restoration Diffusion Model Guidance
College of Medicine and Biological Information Engineering, Northeastern University, Shenyang 110169, China · Tohoku University School of Medicine, Sendai 980-8575, Japan · Department of Clinical Neurosciences, University of Cambridge, UK
Domain-Adapted Latent Diffusion Models for CT-to-PET Image Translation
Oct 26, 2026Yann Bourdé, Lea Bogensperger, Arnela Hadzic +1Medical Image Synthesis Image-to-Image Translation
Department of Quantitative Biomedicine, University of Zurich · Institute for Medical Informatics, Statistics and Documentation, Medical University of Graz

Oct 7, 2026J/K move · Enter open · S save

PhyDiCT: Plug-and-Play CT Reconstruction from Sparse X-Rays via Differentiable Rendering and Strong Priors

Weicheng Dai, Shantanu Ghosh, Kayhan Batmanghelich

Department of Electrical and Computer Engineering, Boston University, Boston, MA, USA

Sparse-View CT Reconstruction 3D Medical Imaging Diffusion-Based Inverse Problems Physics-Informed Diffusion Models

Open paper AskPDF arXiv

Reconstructing 3D Computed Tomography (CT) images from a few X-ray projections is a highly ill-posed inverse problem due to the loss of volumetric information. We propose PhyDiCT, a training-free framework that integrates a differentiable Physics-based forward model, grounded in the Beer-Lambert law, with a text-conditioned Diffusion as a strong prior to reconstruct 3D lung CT images. We refer to our approach as training-free since the prior model is used without fine-tuning, and our goal is to steer the denoising procedure to generate samples consistent with X-ray observations. We guide the diffusion generation using Split Gibbs sampling to jointly optimize for projection fidelity (reward) and consistency with prior knowledge. Also, we introduce a test-time refinement step that enhances image realism and anatomical coherence. We extensively evaluate our method on publicly available 3D CT datasets using both perceptual and semantic metrics, demonstrating that it surpasses existing plug-and-play diffusion and fully trained reconstruction approaches. Our findings highlight that combining a strong generative prior with the underlying physics of image formation substantially improves reconstruction quality, e.g., 7.5% improvement on SSIM compared to full training methods. Code will be released at https://github.com/batmanlab/PhyDiCT.

Abstract

Medical Image Super-Resolution

Explore similar work

PhyDiCT: Plug-and-Play CT Reconstruction from Sparse X-Rays via Differentiable Rendering and Strong Priors

AnF-DiffPET: Anatomy- and Frequency-Guided Diffusion for PET/CT Denoising

Domain-Adapted Latent Diffusion Models for CT-to-PET Image Translation