pith. sign in

arxiv: 2605.20267 · v1 · pith:FE7FCNE6new · submitted 2026-05-18 · 💻 cs.CV · cs.AI

Generation of Heterogeneous PET Images from Uniform Organ Activity Maps Using a Pretrained Domain-Adapted Diffusion Model

Suya Li , Kaushik Dutta , Debojyoti Pal , Jingqin Luo , Kooresh I. Shoghi This is my paper

Pith reviewed 2026-05-21 07:27 UTC · model grok-4.3

classification 💻 cs.CV cs.AI
keywords PET synthesisdiffusion modelsdomain adaptationimage generationmedical imagingquantitative PETvirtual imaging trials
0
0 comments X

The pith

A pretrained domain-adapted diffusion model generates realistic heterogeneous PET images from uniform organ activity maps.

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

The paper establishes that a diffusion model pretrained on natural images and then adapted for PET can turn basic uniform maps of organ activity into full heterogeneous PET scans. These maps come from CT organ outlines with each organ assigned its average uptake value from a real scan. The resulting images match real PET data in organ-level SUV accuracy and produce noise and texture patterns close enough that human readers cannot tell them apart in forced-choice tests. This approach avoids the heavy computation of physics-based simulators while still supporting tasks like tumor segmentation at levels comparable to actual images. The work matters for anyone needing large, varied PET datasets without collecting more patient scans.

Core claim

The PAD model adopts a natural-image pretrained text-to-image decoder, adds an upstream conditioning encoder for anatomy and an downstream PET-domain adapter, and uses a two-phase training process to first learn coarse uptake distributions then refine local details, allowing it to synthesize heterogeneous PET images from uniform organ activity maps that achieve concordance correlation coefficients above 0.92 for organ mean SUVs and visual indistinguishability in observer studies.

What carries the argument

The pretrained domain-adapted diffusion (PAD) model, which conditions a natural-image decoder on uniform activity maps via an encoder and adapter and trains in two phases to capture coarse then fine uptake patterns.

Load-bearing premise

Uniform organ activity maps created by assigning mean uptake values from paired PET images to CT-based segmentations contain enough anatomical and functional information for the diffusion model to generate clinically relevant heterogeneous uptake patterns.

What would settle it

A larger observer study or quantitative evaluation across multiple scanners and patient groups in which readers distinguish synthesized images from real ones at rates well above chance or organ SUV correlations drop below 0.9.

read the original abstract

Synthetic PET images are valuable for quantitative imaging workflow development, scalable virtual imaging trials, and deep learning model training, but conventional physics-based simulation approaches are computationally intensive, limited in anatomical variability, and often fail to capture heterogeneous PET uptake. This study developed a pretrained domain-adapted diffusion (PAD) model for anatomy-conditioned PET synthesis from uniform organ activity maps. PAD adopts a natural-image pretrained text-to-image decoder with an upstream conditioning encoder and a downstream PET-domain adapter. A two-phase training strategy was used, with the first phase learning coarse uptake distributions and the second refining local image details. Uniform organ activity maps were generated from CT-based segmentations by assigning each organ its mean uptake from the paired PET image. Evaluation included quantitative accuracy, noise assessment, radiomic analysis, tumor segmentation performance, and a human observer study. PAD-generated images achieved high quantitative accuracy, with concordance correlation coefficients above 0.92 between organ mean SUVs and assigned activity values. The synthesized images showed noise levels and texture characteristics similar to target PET images and produced comparable tumor segmentation performance. In a two-alternative forced-choice observer study, four readers achieved approximately 50% accuracy, indicating visual indistinguishability between synthesized and target images. PAD also generated realistic PET images from XCAT-derived activity maps, demonstrating compatibility with phantom-based anatomical priors. Overall, PAD provides a diffusion-based framework for generating clinically relevant heterogeneous PET images from uniform organ activity maps derived from clinical segmentations or digital phantoms, supporting data augmentation and downstream imaging studies.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 1 minor

Summary. The manuscript introduces a pretrained domain-adapted diffusion (PAD) model for synthesizing heterogeneous PET images from uniform organ activity maps obtained by assigning mean SUVs to CT-segmented organs. It employs a two-phase training strategy on a domain-adapted diffusion model starting from natural image pretraining. The evaluation includes quantitative metrics like concordance correlation coefficient (CCC > 0.92) for mean SUVs, noise and texture similarity, radiomic analysis, tumor segmentation performance, and a human 2AFC observer study showing ~50% accuracy indicating visual similarity to real PET images. It also demonstrates generation from XCAT phantom maps.

Significance. This approach, if the generated heterogeneity is indeed conditioned on the input rather than solely from pretraining priors, offers a computationally efficient alternative to physics-based simulations for creating diverse synthetic PET data. This could be significant for augmenting datasets in deep learning for PET analysis, virtual clinical trials, and workflow development in quantitative imaging. The two-phase adaptation and compatibility with phantom priors are noted strengths.

major comments (3)
  1. [Abstract] Abstract: The reported concordance correlation coefficients above 0.92 for organ mean SUVs are largely by construction, as the uniform activity maps are generated by assigning the mean uptake from the paired PET images to CT-based segmentations. This metric does not test the central claim that the model synthesizes clinically relevant heterogeneous intra-organ uptake patterns from the minimal functional conditioning in the uniform map.
  2. [Evaluation] Evaluation (quantitative accuracy, radiomic analysis, tumor segmentation): Aggregate checks such as texture similarity, radiomic feature overlap, and tumor segmentation parity are compatible with the model applying generic heterogeneity priors learned from natural-image pretraining and domain adaptation, rather than respecting patient-specific spatial or intensity patterns present in the held-out real PET. Direct per-voxel or per-organ distribution comparisons to ground truth would be required to substantiate the heterogeneity claim.
  3. [Methods] Methods (two-phase training strategy): The description of the two-phase training lacks sufficient detail on validation against overfitting or ablation results demonstrating that the second phase refines details based on the uniform-map conditioning input rather than hallucinating plausible patterns independent of the input.
minor comments (1)
  1. [Abstract] Abstract: The 2AFC observer study reports approximately 50% accuracy without specifying the number of image pairs evaluated, reader expertise levels, or statistical measures such as confidence intervals or p-values, limiting assessment of the robustness of the visual indistinguishability conclusion.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for their constructive and insightful comments, which have helped us identify areas to strengthen the presentation and evaluation of our work on the pretrained domain-adapted diffusion model for PET synthesis. We address each major comment point by point below, with plans for revisions where appropriate to better substantiate the claims regarding heterogeneity synthesis.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The reported concordance correlation coefficients above 0.92 for organ mean SUVs are largely by construction, as the uniform activity maps are generated by assigning the mean uptake from the paired PET images to CT-based segmentations. This metric does not test the central claim that the model synthesizes clinically relevant heterogeneous intra-organ uptake patterns from the minimal functional conditioning in the uniform map.

    Authors: We agree that the CCC metric for organ mean SUVs is largely by construction given how the uniform activity maps are derived from the real PET means, and it primarily verifies preservation of the input conditioning at the organ level rather than directly proving intra-organ heterogeneity. The manuscript's central claim for clinically relevant heterogeneity is instead supported by the combination of texture similarity metrics, radiomic feature distributions that capture intra-organ variability, tumor segmentation parity on synthesized vs. real images, and the 2AFC observer study showing ~50% accuracy (indistinguishability). These evaluations collectively indicate that the generated patterns go beyond generic priors. We will revise the abstract to explicitly distinguish the role of the CCC from the supporting heterogeneity validations. revision: partial

  2. Referee: [Evaluation] Evaluation (quantitative accuracy, radiomic analysis, tumor segmentation): Aggregate checks such as texture similarity, radiomic feature overlap, and tumor segmentation parity are compatible with the model applying generic heterogeneity priors learned from natural-image pretraining and domain adaptation, rather than respecting patient-specific spatial or intensity patterns present in the held-out real PET. Direct per-voxel or per-organ distribution comparisons to ground truth would be required to substantiate the heterogeneity claim.

    Authors: This is a fair point regarding the limitations of aggregate metrics in isolating patient-specific conditioning effects from pretraining priors. While the radiomic features include first- and higher-order statistics sensitive to spatial heterogeneity and intensity distributions, and the tumor segmentation results imply preservation of relevant patterns, we acknowledge that direct per-voxel or per-organ distribution comparisons (e.g., voxel-wise histograms or correlation analyses) would provide stronger evidence. We will add such direct comparisons in the revised evaluation section to better demonstrate that the heterogeneity respects the held-out real PET patterns rather than solely relying on learned priors. The XCAT phantom experiments further support conditioning fidelity since they lack real PET priors. revision: yes

  3. Referee: [Methods] Methods (two-phase training strategy): The description of the two-phase training lacks sufficient detail on validation against overfitting or ablation results demonstrating that the second phase refines details based on the uniform-map conditioning input rather than hallucinating plausible patterns independent of the input.

    Authors: We accept that the methods section would benefit from expanded details on the two-phase strategy. Phase 1 trains the model to capture coarse organ-level uptake from the uniform maps, while Phase 2 refines local textures and details via the domain-adapted diffusion process. Validation against overfitting was performed using a held-out test set with monitoring of CCC, noise, and texture metrics to ensure generalization. To directly address the concern, we will revise the methods to include more specifics on hyperparameters, training schedules, and new ablation experiments comparing one-phase vs. two-phase training, showing improved fidelity to the conditioning input in the refinement phase. revision: yes

Circularity Check

0 steps flagged

No significant circularity; evaluation metrics are independent of model inputs

full rationale

The paper describes an empirical machine-learning pipeline for conditional diffusion-based PET synthesis. Uniform organ activity maps serve as explicit conditioning inputs derived from CT segmentations and mean SUVs; the model is pretrained on natural images then adapted in two phases. Reported results rest on external quantitative checks (CCC > 0.92 for organ means, radiomic feature overlap, tumor segmentation parity, and a 50 % 2AFC observer study) that compare synthesized outputs against held-out real PET images. These checks do not reduce by construction to the fitted parameters, the uniform-map inputs, or any self-citation chain. No equations or uniqueness theorems are invoked that would make the heterogeneity generation tautological with the conditioning. The derivation chain is therefore self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central claim rests on standard assumptions about diffusion model adaptability and the sufficiency of mean-activity conditioning; the primary free parameter is the construction of input activity maps from clinical data.

free parameters (1)
  • mean uptake values per organ
    Assigned from paired PET images to uniform maps; these data-derived values serve as the conditioning input and directly influence output accuracy.
axioms (1)
  • domain assumption A natural-image pretrained text-to-image decoder can be effectively adapted to the PET domain using an upstream conditioning encoder and downstream adapter.
    Invoked in the two-phase training strategy described in the abstract.

pith-pipeline@v0.9.0 · 5824 in / 1501 out tokens · 33855 ms · 2026-05-21T07:27:51.746924+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

72 extracted references · 72 canonical work pages · 3 internal anchors

  1. [1]

    Introduction Synthetic positron emission tomography (PET) imaging data are of substantial value in the field of PET imaging because they serve as a practical ground truth for developing , optimizing, and validat ing imaging methods 1,2. Conventionally, the imaging subject is represented using digital anthropomorphic phantoms, for example the 4D extended c...

    work page

  2. [2]

    Dataset preparation and preprocessing 2.1.1

    Methods 2.1. Dataset preparation and preprocessing 2.1.1. Uniform organ activity map generation A total of 513 FDG-PET/CT images without PET-positive lesions from a publicly available clinical dataset23 were used as source data to extract patient anatomy and then generate uniform organ activity maps. For each case, the CT image was automatically segmented...

    work page

  3. [3]

    8 Figure 2

    Results To provide a visual comparison , Figure 2 displays representative slices of target images, uniform organ activity maps, and synthesized images from various networks (in SUV units). 8 Figure 2. Visual comparison of target images, uniform organ activity maps, and generated images from various networks (at columns) in SUV units across three transaxia...

    work page

  4. [4]

    In this work, we developed PAD, a diffusion–based framework for generating heterogeneous PET images from uniform organ activity maps

    Discussion Synthetic PET images play an important role in task -based image applications, as they enable imaging pipelines and analysis methods to be evaluated under conditions that approximate clinical practice. In this work, we developed PAD, a diffusion–based framework for generating heterogeneous PET images from uniform organ activity maps. By learnin...

    work page

  5. [5]

    PAD generates heterogeneous PET images from uniform organ activity maps derived from either clinical segmentations or human phantoms, such as XCAT

    Conclusion We proposed PAD, a diffusion–based framework for anatomy -conditioned PET image synthesis. PAD generates heterogeneous PET images from uniform organ activity maps derived from either clinical segmentations or human phantoms, such as XCAT . The generated images were rigorously evaluated across five dimensions, including quantitative accuracy, no...

    work page

  6. [6]

    Investigation of novel preclinical Total Body PET designed with J -PET technology: a simulation study

    Dadgar M, Parzych S, Ardebili FT, et al. Investigation of novel preclinical Total Body PET designed with J -PET technology: a simulation study. IEEE Transactions on Radiation and Plasma Medical Sciences. 2022;7(2):124-131

    work page 2022

  7. [7]

    Performance comparison of a dedicated total breast PET system with a clinical whole -body PET system: a simulation study

    Samanta S, Jiang J, Hamdi M, et al. Performance comparison of a dedicated total breast PET system with a clinical whole -body PET system: a simulation study. Physics in Medicine & Biology. 2021;66(11):115004

    work page 2021

  8. [8]

    4D XCAT phantom for multimodality imaging research

    Segars WP , Sturgeon G, Mendonca S, Grimes J, Tsui BM. 4D XCAT phantom for multimodality imaging research. Medical physics. 2010;37(9):4902-4915

    work page 2010

  9. [9]

    GATE: a simulation toolkit for PET and SPECT

    Jan S, Santin G, Strul D, et al. GATE: a simulation toolkit for PET and SPECT. Physics in Medicine & Biology. 2004;49(19):4543-4561

    work page 2004

  10. [10]

    Designing a compact high performance brain PET scanner—simulation study

    Gong K, Majewski S, Kinahan PE, et al. Designing a compact high performance brain PET scanner—simulation study. Physics in Medicine & Biology. 2016;61(10):3681-3697

    work page 2016

  11. [11]

    Comparing evaluation of quantitative imaging methods using reference standards versus regression‑without‑truth‑based technique

    Liu Y , Jha AK. Comparing evaluation of quantitative imaging methods using reference standards versus regression‑without‑truth‑based technique. SPIE; 2026:157-162

    work page 2026

  12. [12]

    An ultra-high sensitivity sensing collimation imager: design, modelling, and assessment

    Cai L, Aggarwal S, Enninful K, et al. An ultra-high sensitivity sensing collimation imager: design, modelling, and assessment. SPIE; 2025:47-53

    work page 2025

  13. [13]

    GATE simulations on multicomputer architectures

    Vega-Acevedo N, Lanas I, Rannou FR. GATE simulations on multicomputer architectures. IEEE; 2009:4193-4196

    work page 2009

  14. [14]

    Using deep learning techniques in medical imaging: a systematic review of applications on CT and PET

    Domingues I, Pereira G, Martins P , Duarte H, Santos J, Abreu PH. Using deep learning techniques in medical imaging: a systematic review of applications on CT and PET. Artificial Intelligence Review. 2020;53(6):4093-4160

    work page 2020

  15. [15]

    Deep learning-based PET image denoising and reconstruction: a review

    Hashimoto F, Onishi Y , Ote K, Tashima H, Reader AJ, Yamaya T. Deep learning-based PET image denoising and reconstruction: a review. Radiological physics and technology . 2024;17(1):24-46

    work page 2024

  16. [16]

    Auto-encoding variational bayes

    Kingma DP , Welling M. Auto-encoding variational bayes. arXiv preprint arXiv:13126114. 2013

    work page 2013

  17. [17]

    Generative adversarial nets

    Goodfellow IJ, Pouget-Abadie J, Mirza M, et al. Generative adversarial nets. Advances in neural information processing systems. 2014;27

    work page 2014

  18. [18]

    Deep learning–based dynamic PET parametric K i image generation from lung static PET

    Wang H, Wu Y , Huang Z, et al. Deep learning–based dynamic PET parametric K i image generation from lung static PET. European Radiology. 2023;33(4):2676-2685

    work page 2023

  19. [19]

    Bidirectional mapping generative adversarial networks for brain MR to PET synthesis

    Hu S, Lei B, Wang S, Wang Y , Feng Z, Shen Y . Bidirectional mapping generative adversarial networks for brain MR to PET synthesis. IEEE Transactions on Medical Imaging . 2021;41(1):145-157

    work page 2021

  20. [20]

    Supervised learning with cyclegan for low-dose FDG PET image denoising

    Zhou L, Schaefferkoetter JD, Tham IW , Huang G, Yan J. Supervised learning with cyclegan for low-dose FDG PET image denoising. Medical image analysis. 2020;65:101770

    work page 2020

  21. [21]

    A review on generative adversarial networks: Algorithms, theory, and applications

    Gui J, Sun Z, Wen Y , Tao D, Ye J. A review on generative adversarial networks: Algorithms, theory, and applications. IEEE transactions on knowledge and data engineering . 2021;35(4):3313-3332

    work page 2021

  22. [22]

    Denoising diffusion probabilistic models

    Ho J, Jain A, Abbeel P . Denoising diffusion probabilistic models. Advances in neural information processing systems. 2020;33:6840-6851. 16

    work page 2020

  23. [23]

    Diffusion models beat gans on image synthesis

    Dhariwal P , Nichol A. Diffusion models beat gans on image synthesis. Advances in neural information processing systems. 2021;34:8780-8794

    work page 2021

  24. [24]

    2D medical image synthesis using transformer -based denoising diffusion probabilistic model

    Pan S, Wang T, Qiu RL, et al. 2D medical image synthesis using transformer -based denoising diffusion probabilistic model. Physics in Medicine & Biology. 2023;68(10):105004

    work page 2023

  25. [25]

    CBCT ‐Based synthetic CT image generation using conditional denoising diffusion probabilistic model

    Peng J, Qiu RL, Wynne JF, et al. CBCT ‐Based synthetic CT image generation using conditional denoising diffusion probabilistic model. Medical physics. 2024;51(3):1847-1859

    work page 2024

  26. [26]

    Full‐dose whole‐body PET synthesis from low‐dose PET using high‐efficiency denoising diffusion probabilistic model: PET consistency model

    Pan S, Abouei E, Peng J, et al. Full‐dose whole‐body PET synthesis from low‐dose PET using high‐efficiency denoising diffusion probabilistic model: PET consistency model. Medical Physics. 2024;51(8):5468-5478

    work page 2024

  27. [27]

    Multi-sequence mri to multi- tracer pet generation via diffusion model

    Zhong W, Cong C, Azemi G, Tabassum M, Di Ieva A, Liu S. Multi-sequence mri to multi- tracer pet generation via diffusion model. IEEE; 2025:1-4

    work page 2025

  28. [28]

    A whole -body FDG -PET/CT Dataset with manually annotated Tumor Lesions

    Gatidis S, Hepp T, Früh M, et al. A whole -body FDG -PET/CT Dataset with manually annotated Tumor Lesions. Scientific Data. 2022;9(1)doi:10.1038/s41597-022-01718-3

    work page doi:10.1038/s41597-022-01718-3 2022

  29. [29]

    TotalSegmentator: Robust Segmentation of 104 Anatomic Structures in CT Images

    Wasserthal J, Breit H-C, Meyer MT, et al. TotalSegmentator: Robust Segmentation of 104 Anatomic Structures in CT Images. Radiology: Artificial Intelligence . 2023;5(5)doi:10.1148/ryai.230024

    work page doi:10.1148/ryai.230024 2023

  30. [30]

    GLIDE: Towards Photorealistic Image Generation and Editing with Text-Guided Diffusion Models

    Wang TaZ, Ting and Zhang, Bo and Ouyang, Hao and Chen, Dong and Chen, Qifeng and Wen, Fang. Pretraining is All You Need for Image -to-Image Translation. arXiv:220512952. 2022;doi:10.48550/arxiv.2112.10741

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2112.10741 2022

  31. [31]

    The opencv library

    Bradski G. The opencv library. Dr Dobb's Journal: Software T ools for the Professional Programmer. 2000;25(11):120-123

    work page 2000

  32. [32]

    GLIDE: Towards Photorealistic Image Generation and Editing with Text- Guided Diffusion Models

    Alex Nichol PD, Aditya Ramesh, Pranav Shyam, Pamela Mishkin, Bob McGrew, Ilya Sutskever, Mark Chen. GLIDE: Towards Photorealistic Image Generation and Editing with Text- Guided Diffusion Models. arXiv:211210741. 2022;doi:10.48550/arxiv.2205.12952

    work page doi:10.48550/arxiv.2205.12952 2022

  33. [33]

    Deep learning,

    Lecun Y , Bengio Y , Hinton G. Deep learning. Nature. 2015;521(7553):436 -444. doi:10.1038/nature14539

    work page doi:10.1038/nature14539 2015

  34. [34]

    Attention is all you need

    Vaswani A, Shazeer N, Parmar N, et al. Attention is all you need. Advances in neural information processing systems. 2017;30

    work page 2017

  35. [35]

    Decoupled Weight Decay Regularization

    Ilya Loshchilov FH. Decoupled Weight Decay Regularization. arXiv:171105101. 2019;doi:10.48550/arxiv.1711.05101

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1711.05101 2019

  36. [36]

    Improved Denoising Diffusion Probabilistic Models

    Alex Nichol PD. Improved Denoising Diffusion Probabilistic Models. arXiv:210209672. 2021;doi:10.48550/arxiv.2102.09672

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2102.09672 2021

  37. [37]

    Exponential Moving Average of Weights in Deep Learning: Dynamics and Benefits

    Daniel Morales-Brotons TV , Hadrien Hendrikx. Exponential Moving Average of Weights in Deep Learning: Dynamics and Benefits. arXiv:241118704. 2024;doi:10.48550/arxiv.2411.18704

    work page doi:10.48550/arxiv.2411.18704 2024

  38. [38]

    Image -to-image translation with conditional adversarial networks

    Isola P , Zhu J -Y , Zhou T, Efros AA. Image -to-image translation with conditional adversarial networks. 2017:1125-1134

    work page 2017

  39. [39]

    ResViT: Residual vision transformers for multimodal medical image synthesis

    Dalmaz O, Yurt M, Çukur T. ResViT: Residual vision transformers for multimodal medical image synthesis. IEEE Transactions on Medical Imaging. 2022;41(10):2598-2614

    work page 2022

  40. [40]

    A concordance correlation coefficient to evaluate reproducibility

    Lawrence I, Lin K. A concordance correlation coefficient to evaluate reproducibility. Biometrics. 1989:255-268. 17

    work page 1989

  41. [41]

    The bootstrap: a technique for data -driven statistics

    Henderson AR. The bootstrap: a technique for data -driven statistics. Using computer - intensive analyses to explore experimental data. Clinica chimica acta. 2005;359(1-2):1-26

    work page 2005

  42. [42]

    Potential clinical impact of LAFOV PET/CT: a systematic evaluation of image quality and lesion detection

    Honoré d’Este S, Andersen FL, Andersen JB, et al. Potential clinical impact of LAFOV PET/CT: a systematic evaluation of image quality and lesion detection. Diagnostics. 2023;13(21):3295

    work page 2023

  43. [43]

    FAST (fast analytical simulator of tracer)-PET: an accurate and efficient PET analytical simulation tool

    Li S, Hamdi M, Dutta K, et al. FAST (fast analytical simulator of tracer)-PET: an accurate and efficient PET analytical simulation tool. Physics in Medicine & Biology. 2024;69(16):165020

    work page 2024

  44. [44]

    On the histogram as a density estimator: L 2 theory

    Freedman D, Diaconis P . On the histogram as a density estimator: L 2 theory. Zeitschrift für Wahrscheinlichkeitstheorie und verwandte Gebiete. 1981;57(4):453-476

    work page 1981

  45. [45]

    PET radiomics in lung cancer: advances and translational challenges

    Zhang Y , Huang W, Jiao H, Kang L. PET radiomics in lung cancer: advances and translational challenges. EJNMMI physics. 2024;11(1):81

    work page 2024

  46. [46]

    Value of pre -treatment 18F-FDG PET/CT radiomics in predicting the prognosis of stage III-IV colorectal cancer

    Wang N, Dai M, Zhao Y , et al. Value of pre -treatment 18F-FDG PET/CT radiomics in predicting the prognosis of stage III-IV colorectal cancer. European Journal of Radiology Open. 2023;10:100480

    work page 2023

  47. [47]

    Divergence measures based on the Shannon entropy

    Lin J. Divergence measures based on the Shannon entropy. IEEE Transactions on Information theory. 2002;37(1):145-151

    work page 2002

  48. [48]

    Objective task -based evaluation of quantitative medical imaging methods: emerging frameworks and future directions

    Liu Y , Xia H, Obuchowski NA, et al. Objective task -based evaluation of quantitative medical imaging methods: emerging frameworks and future directions. PET clinics . 2025;20(4):475-488

    work page 2025

  49. [49]

    Simultaneous tumor segmentation, image restoration, and blur kernel estimation in PET using multiple regularizations

    Li L, Wang J, Lu W, Tan S. Simultaneous tumor segmentation, image restoration, and blur kernel estimation in PET using multiple regularizations. Computer Vision and Image Understanding. 2017;155:173-194

    work page 2017

  50. [50]

    Whole-body PET image denoising for reduced acquisition time

    Kruzhilov I, Kudin S, Vetoshkin L, Sokolova E, Kokh V . Whole-body PET image denoising for reduced acquisition time. Frontiers in Medicine. 2024;11:1415058

    work page 2024

  51. [51]

    Observer study -based evaluation of a stochastic and physics-based method to generate oncological PET images

    Liu Z, Laforest R, Mhlanga J, et al. Observer study -based evaluation of a stochastic and physics-based method to generate oncological PET images. SPIE; 2021:2

    work page 2021

  52. [52]

    Effect of random background inhomogeneity on observer detection performance

    Rolland JP , Barrett HH. Effect of random background inhomogeneity on observer detection performance. Journal of the Optical Society of America A. 1992;9(5):649-658

    work page 1992

  53. [53]

    PETSTEP: generation of synthetic PET lesions for fast evaluation of segmentation methods

    Berthon B, Häggström I, Apte A, et al. PETSTEP: generation of synthetic PET lesions for fast evaluation of segmentation methods. Physica Medica. 2015;31(8):969-980

    work page 2015

  54. [54]

    From FDG to PSMA: A Hitchhiker's Guide to Multitracer, Multicenter Lesion Segmentation in PET/CT Imaging

    Rokuss M, Kovacs B, Kirchhoff Y , et al. From FDG to PSMA: A Hitchhiker's Guide to Multitracer, Multicenter Lesion Segmentation in PET/CT Imaging. arXiv preprint arXiv:240909478. 2024

    work page 2024

  55. [55]

    Measures of the amount of ecologic association between species

    Dice LR. Measures of the amount of ecologic association between species. Ecology. 1945;26(3):297-302

    work page 1945

  56. [56]

    Comparison and evaluation of methods for liver segmentation from CT datasets

    Heimann T, Van Ginneken B, Styner MA, et al. Comparison and evaluation of methods for liver segmentation from CT datasets. IEEE transactions on medical imaging. 2009;28(8):1251- 1265

    work page 2009

  57. [57]

    Individual comparisons by ranking methods

    Wilcoxon F. Individual comparisons by ranking methods. Breakthroughs in statistics: Methodology and distribution. Springer; 1992:196-202

    work page 1992

  58. [58]

    Signal detection theory and psychophysics

    Green DM, Swets JA. Signal detection theory and psychophysics. vol 1. Wiley New York; 1966. 18

    work page 1966

  59. [59]

    Observer-study-based approaches to quantitatively evaluate the realism of synthetic medical images

    Liu Z, Wolfe S, Y u Z, et al. Observer-study-based approaches to quantitatively evaluate the realism of synthetic medical images. Physics in Medicine & Biology. 2023;68(7):074001

    work page 2023

  60. [60]

    Foundations of image science

    Barrett HH, Myers KJ. Foundations of image science. John Wiley & Sons; 2013

    work page 2013

  61. [61]

    Teoria statistica delle classi e calcolo delle probabilita

    Bonferroni C. Teoria statistica delle classi e calcolo delle probabilita. Pubblicazioni del R istituto superiore di scienze economiche e commericiali di firenze. 1936;8:3-62

    work page 1936

  62. [62]

    Deep le arning generation of preclinical positron emission tomography (PET) images from low‐count PET with task‐based performance assessment

    Dutta K, Laforest R, Luo J, Jha AK, Shoghi KI. Deep le arning generation of preclinical positron emission tomography (PET) images from low‐count PET with task‐based performance assessment. Medical physics. 2024;51(6):4324-4339

    work page 2024

  63. [63]

    Noise ‐aware system generative model (NASGM): positron emission tomography (PET) image simulation framework with observer validation studies

    Li S, Dutta K, Pal D, Shoghi KI. Noise ‐aware system generative model (NASGM): positron emission tomography (PET) image simulation framework with observer validation studies. Medical Physics. 2025;52(7)doi:10.1002/mp.17962

    work page doi:10.1002/mp.17962 2025

  64. [64]

    Deep unsupervised learning using nonequilibrium thermodynamics

    Sohl-Dickstein J, Weiss E, Maheswaranathan N, Ganguli S. Deep unsupervised learning using nonequilibrium thermodynamics. pmlr; 2015:2256-2265

    work page 2015

  65. [65]

    Improved sample complexity bounds for diffusion model training

    Gupta S, Parulekar A, Price E, Xun Z. Improved sample complexity bounds for diffusion model training. Advances in Neural Information Processing Systems. 2024;37:40976-41012

    work page 2024

  66. [66]

    Virtual clinical trials in medical imaging: a review

    Abadi E, Segars WP , Tsui BM, et al. Virtual clinical trials in medical imaging: a review. Journal of Medical Imaging. 2020;7(4):042805-042805

    work page 2020

  67. [67]

    Development of a Prototype AWSM -PET Device for Augmented Whole-Body PET Imaging and Initial Human Study

    Chen Y , Mintzer RA, Cho S, et al. Development of a Prototype AWSM -PET Device for Augmented Whole-Body PET Imaging and Initial Human Study. Journal of Nuclear Medicine. 2026;67(2):313-321

    work page 2026

  68. [68]

    Automated whole‐body bone lesion detection for multiple myeloma on 68Ga‐Pentixafor PET/CT imaging using deep learning methods

    Xu L, Tetteh G, Lipkova J, et al. Automated whole‐body bone lesion detection for multiple myeloma on 68Ga‐Pentixafor PET/CT imaging using deep learning methods. Contrast media & molecular imaging. 2018;2018(1):2391925

    work page 2018

  69. [69]

    Automatic PET cervical tumor segmentation by combining deep learning and anatomic prior

    Chen L, Shen C, Zhou Z, et al. Automatic PET cervical tumor segmentation by combining deep learning and anatomic prior. Physics in Medicine & Biology. 2019;64(8):085019

    work page 2019

  70. [70]

    Aw -net: A novel fully connected attention -based medical image segmentation model

    Pal D, Meena T, Mahapatra D, Roy S. Aw -net: A novel fully connected attention -based medical image segmentation model. 2023:2532-2541

    work page 2023

  71. [71]

    A deep learning anthropomorphic model observer for a detection task in PET

    Shao M, Byrd DW, Mitra J, et al. A deep learning anthropomorphic model observer for a detection task in PET. Medical physics. 2024;51(10):7093-7107

    work page 2024

  72. [72]

    A comprehensive survey on transfer learning

    Zhuang F, Qi Z, Duan K, et al. A comprehensive survey on transfer learning. Proceedings of the IEEE. 2020;109(1):43-76

    work page 2020