pith. sign in

arxiv: 2606.17675 · v1 · pith:FHWH3Z4Anew · submitted 2026-06-16 · 💻 cs.CV

Do We Really Need Diffusion? A Fast U-Net for Paired Medical Image Translation

Alicia Pirwass , Birte Glimm , Michael Munz , Hans-Joachim Wilke This is my paper

Pith reviewed 2026-06-27 01:45 UTC · model grok-4.3

classification 💻 cs.CV
keywords image-to-image translationU-Netdiffusion modelsMRI signal fat fractionpaired medical imagesT2-weighted MRIlightweight networkNAKO cohort
0
0 comments X

The pith

Lightweight U-Net outperforms DDPM for paired MRI translation with higher accuracy and 208 times faster inference.

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

The paper tests whether signal fat fraction can be estimated from routine T2-weighted MRI using image-to-image translation on a large paired dataset of over 230,000 images. Both a 4-level U-Net and a DDPM learn a non-trivial mapping that beats a simple identity baseline. The U-Net reaches higher correlation and lower error than the DDPM while cutting inference time by a factor of 208. This outcome questions whether diffusion models are needed for paired medical translation tasks. The resulting speed supports real-time clinical estimation of the fat fraction biomarker.

Core claim

The lightweight 4-level U-Net outperforms the DDPM on Pearson correlation (r = 0.975 vs. 0.962) and mean absolute error (MAE = 0.014 +/- 0.015 vs. 0.019 +/- 0.019) while reducing inference time by a factor of 208 (25.2 ms vs. 5227.2 ms per image with 50 DDIM steps). Both models exceed the identity baseline (r = 0.769, MAE = 0.070), confirming they learn a meaningful cross-modal mapping from T2-weighted images to signal fat fraction maps on the NAKO cohort data.

What carries the argument

The 4-level lightweight U-Net that performs direct supervised paired image-to-image translation from T2-weighted MRI to signal fat fraction maps.

If this is right

  • Real-time clinical estimation of signal fat fraction becomes feasible on standard hardware.
  • Diffusion models are not required for this paired medical image translation task.
  • Similar lightweight U-Nets may replace diffusion approaches in other paired medical imaging problems.
  • Large paired datasets enable high-accuracy models without generative model complexity.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The result suggests direct regression with convolutional networks can outperform generative diffusion for paired data tasks.
  • The same lightweight architecture could be tested on unpaired settings or different MRI contrasts to check generalization.
  • Integration into scanner software might allow immediate fat fraction output during routine T2-weighted acquisitions.

Load-bearing premise

The DDPM implementation and training protocol represent a fair state-of-the-art baseline without undisclosed hyper-parameter disadvantages.

What would settle it

Retraining the DDPM with extensive hyperparameter search, alternative sampling schedules, or more than 50 DDIM steps and checking whether accuracy and effective speed can match or exceed the U-Net results.

Figures

Figures reproduced from arXiv: 2606.17675 by Alicia Pirwass, Birte Glimm, Hans-Joachim Wilke, Michael Munz.

Figure 1
Figure 1. Figure 1: Qualitative comparison of T2w-to-SFF translation. The columns show the T2w input image, the U-Net prediction, the DDPM prediction, and the SFF ground truth, respectively. The rows display the full image and a cropped view of the muscle compart￾ments (autochthonous and iliopsoas muscles, left and right) segmented using VIBESeg￾mentator; both models reproduce the overall fat distribution, while the U-Net pre… view at source ↗
read the original abstract

Magnetic resonance imaging-signal fat fraction (MRI-SFF) quantifies tissue fat and serves as an established biomarker for metabolic and musculoskeletal disorders. The acquisition requires, however, specialized MRI sequences, which are not available routinely. We investigate whether SFF can be estimated from widely available T2-weighted (T2w) MRI via image-to-image translation (I2I). We further compare a lightweight 4-level U-Net to a state-of-the-art Denoising Diffusion Probabilistic Model (DDPM) using a dataset of 230 048 paired 2D images (183 517 train, 23 621 val, 22 910 test) from the German National Cohort (NAKO). Both models clearly outperform the identity baseline (Pearson correlation r = 0.769, mean absolute error MAE = 0.070 +/- 0.054), which confirms that the models learn a non-trivial cross-modal mapping. Interestingly, the lightweight U-Net outperforms the DDPM in both correlation (r = 0.975 vs. 0.962) and error (MAE = 0.014 +/- 0.015 vs. 0.019 +/- 0.019), while reducing inference time by a factor of 208 (25.2 ms vs. 5 227.2 ms per image using 50 Denoising Diffusion Implicit Model (DDIM) steps). The strong clinical performance at substantially reduced computational cost enables real-time clinical use.

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

2 major / 1 minor

Summary. The paper claims that a lightweight 4-level U-Net outperforms a DDPM for paired T2w-to-MRI-SFF image translation on 230k NAKO images (r=0.975 vs 0.962; MAE=0.014 vs 0.019), while being 208x faster at inference (25.2 ms vs 5227 ms with 50 DDIM steps), and that both models substantially beat the identity baseline.

Significance. If the DDPM baseline is shown to be properly optimized, the result would indicate that diffusion models are not required for this paired medical I2I task and that a simple U-Net suffices for high-accuracy, real-time clinical use. The large paired dataset and direct numerical comparison are strengths.

major comments (2)
  1. [Abstract / Methods] Abstract and Methods: No training schedule, learning-rate schedule, noise schedule, number of diffusion steps during training, or ablation on DDIM sampling steps is reported for the DDPM. This prevents verification that the 0.013 r and 0.005 MAE gaps reflect inherent model differences rather than unequal optimization, which is load-bearing for the headline claim of U-Net superiority.
  2. [Abstract] Abstract: The DDPM is described only as 'state-of-the-art' with 50 DDIM steps at inference; without hyper-parameter search details or confirmation of classifier-free guidance tuning, the fairness of the baseline cannot be assessed from the given information.
minor comments (1)
  1. [Abstract] Abstract: The reported MAE values include standard deviations; it is unclear whether these are computed over images or over pixels and whether paired statistical tests (e.g., Wilcoxon) were performed to support the claimed superiority.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback highlighting the need for greater transparency in the DDPM baseline implementation. We agree that these details are important for verifying the fairness of the comparison and will revise the manuscript accordingly.

read point-by-point responses

  1. Referee: [Abstract / Methods] Abstract and Methods: No training schedule, learning-rate schedule, noise schedule, number of diffusion steps during training, or ablation on DDIM sampling steps is reported for the DDPM. This prevents verification that the 0.013 r and 0.005 MAE gaps reflect inherent model differences rather than unequal optimization, which is load-bearing for the headline claim of U-Net superiority.

    Authors: We acknowledge the omission of these implementation details. In the revised manuscript we will add a dedicated subsection in Methods describing the DDPM training schedule, learning-rate schedule, noise schedule, number of training diffusion steps, and an ablation study on the number of DDIM inference steps. These additions will allow readers to confirm that the reported performance gap is not attributable to unequal optimization effort. revision: yes

  2. Referee: [Abstract] Abstract: The DDPM is described only as 'state-of-the-art' with 50 DDIM steps at inference; without hyper-parameter search details or confirmation of classifier-free guidance tuning, the fairness of the baseline cannot be assessed from the given information.

    Authors: We will expand the Methods section to document the hyper-parameter choices used for the DDPM, including any grid or random search performed and the classifier-free guidance scale that was selected. This will substantiate the claim that the baseline reflects standard state-of-the-art practice for this task. revision: yes

Circularity Check

0 steps flagged

No circularity: direct empirical comparison on held-out data

full rationale

The paper reports measured Pearson correlations and MAE values on a fixed held-out test set of 22 910 images after training both models. No equations, fitted parameters, or self-citations are used to derive the reported performance numbers; the results are obtained by standard supervised training and evaluation. The comparison therefore does not reduce to any input by construction.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

This is a purely empirical machine-learning comparison paper. No mathematical axioms, free parameters fitted inside a derivation, or invented physical entities are introduced; the only learned quantities are standard neural-network weights optimized on the training split.

pith-pipeline@v0.9.1-grok · 5807 in / 1226 out tokens · 42429 ms · 2026-06-27T01:45:28.334589+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

37 extracted references · 32 canonical work pages · 3 internal anchors

  1. [1]

    In: Proceedings of the IEEE Inter- national Symposium on Biomedical Imaging (ISBI) (2025), arXiv:2410.11186

    Anand, S., et al.: Synthesizing proton-density fat fraction andR∗ 2 from 2-point Dixon MRI with generative machine learning. In: Proceedings of the IEEE Inter- national Symposium on Biomedical Imaging (ISBI) (2025), arXiv:2410.11186

    arXiv 2025

  2. [2]

    Computerized Medical Imaging and Graphics79, 101684 (Jan 2020)

    Armanious, K., Jiang, C., Fischer, M., Küstner, T., Hepp, T., Niko- laou, K., Gatidis, S., Yang, B.: MedGAN: Medical image transla- tion using GANs. Computerized Medical Imaging and Graphics79, 101684 (Jan 2020). https://doi.org/10.1016/j.compmedimag.2019.101684, https://linkinghub.elsevier.com/retrieve/pii/S0895611119300990

    work page doi:10.1016/j.compmedimag.2019.101684 2020

  3. [3]

    Radiology277(1), 206–220 (2015)

    Bamberg, F., Kauczor, H.U., Weckbach, S., Schlett, C.L., et al.: Whole-body mr imaging in the german national cohort: Rationale, de- sign, and technical background. Radiology277(1), 206–220 (2015). https://doi.org/10.1148/radiol.2015142272

    work page doi:10.1148/radiol.2015142272 2015

  4. [4]

    Cross-Modality Synthesis from CT to PET using FCN and GAN Networks for Improved Automated Lesion Detection

    Ben-Cohen, A., Klang, E., Raskin, S.P., Soffer, S., Ben-Haim, S., Konen, E., Amitai, M.M., Greenspan, H.: Cross-Modality Synthesis from CT to PET using FCN and GAN Networks for Improved Automated Lesion Detection (Jul 2018). https://doi.org/10.48550/arXiv.1802.07846, http://arxiv.org/abs/1802.07846, arXiv:1802.07846 [cs]

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1802.07846 2018

  5. [5]

    In: 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Blau, Y., Michaeli, T.: The Perception-Distortion Tradeoff. In: 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 6228–6237 (Jun 2018). https://doi.org/10.1109/CVPR.2018.00652, http://arxiv.org/abs/1711.06077, arXiv:1711.06077 [cs]

    work page doi:10.1109/cvpr.2018.00652 2018

  6. [6]

    Frontiers in Radiol- ogy1, 664444 (Jul 2021)

    Gadermayr, M., Heckmann, L., Li, K., Bähr, F., Müller, M., Truhn, D., Merhof, D., Gess, B.: Image-to-Image Translation for Simplified MRI Muscle Segmentation. Frontiers in Radiol- ogy1, 664444 (Jul 2021). https://doi.org/10.3389/fradi.2021.664444, https://www.frontiersin.org/articles/10.3389/fradi.2021.664444/full

    work page doi:10.3389/fradi.2021.664444 2021

  7. [7]

    European Radiology (2025)

    Graf, R., Platzek, P., Riedel, E.O., et al.: Vibesegmentator: Full body mri segmentation for the nako and uk biobank. European Radiology (2025). https://doi.org/10.1007/s00330-025-12035-9

    work page doi:10.1007/s00330-025-12035-9 2025

  8. [8]

    European Radiology Experimen- tal7(1), 70 (Nov 2023)

    Graf, R., Schmitt, J., Schlaeger, S., Möller, H.K., Sideri-Lampretsa, V., Sekuboyina, A., Krieg, S.M., Wiestler, B., Menze, B., Rueckert, D., Kirschke, J.S.: Denoising diffusion-based MRI to CT image translation enables automated spinal segmentation. European Radiology Experimen- tal7(1), 70 (Nov 2023). https://doi.org/10.1186/s41747-023-00385-2, https://...

    work page doi:10.1186/s41747-023-00385-2 2023

  9. [9]

    Medical Physics 44(4), 1408–1419 (2017)

    Han, X.: MR-based synthetic CT generation using a deep convolutional neural network method. Medical Physics 44(4), 1408–1419 (2017). https://doi.org/10.1002/mp.12155, https://onlinelibrary.wiley.com/doi/abs/10.1002/mp.12155

    work page doi:10.1002/mp.12155 2017

  10. [10]

    Scientific Reports (2026)

    Hess, H., Oswald, A., Daneshvar, K., Gerber, N., Schär, M., Zumstein, M.A., Ger- ber, K.: Quantitative fat-fraction analysis of the rotator cuff muscles on clinical sagittal and coronal T1-weighted MRI using deep learning algorithms. Scientific Reports (2026). https://doi.org/10.1038/s41598-026-38108-3

    work page doi:10.1038/s41598-026-38108-3 2026

  11. [11]

    Cross-modality image synthesis from unpaired data using CycleGAN: Effects of gradient consistency loss and training data size

    Hiasa, Y., Otake, Y., Takao, M., Matsuoka, T., Takashima, K., Prince, J.L., Sugano, N., Sato, Y.: Cross-modality image synthesis from un- paired data using CycleGAN: Effects of gradient consistency loss and training data size (Jul 2018). https://doi.org/10.48550/arXiv.1803.06629, http://arxiv.org/abs/1803.06629, arXiv:1803.06629 [cs] A Fast U-Net for Pair...

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1803.06629 2018

  12. [12]

    In: Advances in Neural Information Processing Systems (NeurIPS)

    Ho, J., Jain, A., Abbeel, P.: Denoising diffusion probabilistic models. In: Advances in Neural Information Processing Systems (NeurIPS). vol. 33, pp. 6840–6851. Cur- ran Associates, Inc. (2020)

    2020

  13. [13]

    https://doi.org/10.48550/ARXIV.2509.22049, https://arxiv.org/abs/2509.22049, version Number: 1

    Honey, E., Helbo, A., Petersen, J.: Comparative Analysis of GAN and Diffusion for MRI-to-CT translation (2025). https://doi.org/10.48550/ARXIV.2509.22049, https://arxiv.org/abs/2509.22049, version Number: 1

    work page doi:10.48550/arxiv.2509.22049 2025

  14. [14]

    In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR)

    Isola, P., Zhu, J.Y., Zhou, T., Efros, A.A.: Image-to-image translation with conditional adversarial networks. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR). pp. 5967–5976 (2017). https://doi.org/10.1109/CVPR.2017.632

    work page doi:10.1109/cvpr.2017.632 2017

  15. [15]

    Medical Image Analysis88, 102846 (2023)

    Kazerouni, A., Aghdam, E.K., Heidari, M., Azad, R., Fayyaz, M., Hacihaliloglu, I., Merhof, D.: Diffusion models in medical imaging: A comprehensive survey. Medical Image Analysis88, 102846 (2023). https://doi.org/10.1016/j.media.2023.102846

    work page doi:10.1016/j.media.2023.102846 2023

  16. [16]

    Supply chain logistics with quantum and classical annealing algorithms

    Khosravi, P., et al.: Denoising diffusion probabilistic models for 3D medical im- age generation. Scientific Reports13(2023). https://doi.org/10.1038/s41598-023- 34341-2

    work page doi:10.1038/s41598-023- 2023

  17. [17]

    Journal of Cachexia, Sarcopenia and Muscle13(2), 781–794 (2022)

    Li, C.W., Yu, K., Shyh-Chang, N., Jiang, Z., Liu, T., Ma, S., Luo, L., Guang, L., Liang, K., Ma, W., Miao, H., Cao, W., Liu, R., Jiang, L.J., Yu, S.L., Li, C., Liu, H.J., Xu, L.Y., Liu, R.J., Zhang, X.Y., Liu, G.S.: Pathogenesis of sarcopenia and the relationship with fat mass: Descriptive review. Journal of Cachexia, Sarcopenia and Muscle13(2), 781–794 (...

    work page doi:10.1002/jcsm.12901 2022

  18. [18]

    Quantitative Imaging in Medicine and Surgery10(6), 1223–1236 (Jun 2020)

    Li, W., Li, Y., Qin, W., Liang, X., Xu, J., Xiong, J., Xie, Y.: Mag- netic resonance image (MRI) synthesis from brain computed tomogra- phy (CT) images based on deep learning methods for magnetic reso- nance (MR)-guided radiotherapy. Quantitative Imaging in Medicine and Surgery10(6), 1223–1236 (Jun 2020). https://doi.org/10.21037/qims-19-885, http://qims....

    work page doi:10.21037/qims-19-885 2020

  19. [19]

    Computers in Biology and Medicine157, 106738 (May 2023)

    Li, Y., Xu, S., Chen, H., Sun, Y., Bian, J., Guo, S., Lu, Y., Qi, Z.: CT synthesis from multi-sequence MRI using adaptive fu- sion network. Computers in Biology and Medicine157, 106738 (May 2023). https://doi.org/10.1016/j.compbiomed.2023.106738, https://linkinghub.elsevier.com/retrieve/pii/S0010482523002032

    work page doi:10.1016/j.compbiomed.2023.106738 2023

  20. [20]

    BMC Medical Informatics and Decision Mak- ing25, 390 (Oct 2025)

    Luo, J., Yang, L., Liu, Y., Hu, C., Wang, G., Yang, Y., Yang, T.L., Zhou, X.: Review of diffusion models and its applications in biomedical informatics. BMC Medical Informatics and Decision Mak- ing25, 390 (Oct 2025). https://doi.org/10.1186/s12911-025-03210-5, https://pmc.ncbi.nlm.nih.gov/articles/PMC12541957/

    work page doi:10.1186/s12911-025-03210-5 2025

  21. [21]

    https://doi.org/10.48550/arXiv.2209.12104, http://arxiv.org/abs/2209.12104, arXiv:2209.12104 [eess]

    Lyu, Q., Wang, G.: Conversion Between CT and MRI Images Using Diffusion and Score-Matching Models (Sep 2022). https://doi.org/10.48550/arXiv.2209.12104, http://arxiv.org/abs/2209.12104, arXiv:2209.12104 [eess]

    work page doi:10.48550/arxiv.2209.12104 2022

  22. [22]

    BMC Medi- cal Imaging23(1), 48 (2023)

    Masi, S., Rye, M., Roussac, A., Naghdi, N., Rosenstein, B., Bailey, J.F., Fortin, M.: Comparison of paraspinal muscle composition measure- ments using IDEAL fat-water and T2-weighted MR images. BMC Medi- cal Imaging23(1), 48 (2023). https://doi.org/10.1186/S12880-023-00992-W, https://doi.org/10.1186/s12880-023-00992-w

    work page doi:10.1186/s12880-023-00992-w 2023

  23. [23]

    In: Rodolà, E., Galasso, F., Masi, I

    Moschetto, A., Puglisi, L., Sargood, A., Dell’Acqua, P., Guarnera, F., Battiato, S., Ravì, D.: Benchmarking gans, diffusion models, and flow matching for t1w-to-t2w mri translation. In: Rodolà, E., Galasso, F., Masi, I. (eds.) Image Analysis and Processing - ICIAP 2025 Workshops. pp. 429–440. Springer Nature Switzerland (2026) 16 A. Pirwass et al

    2025

  24. [24]

    Computer Methods and Programs in Biomedicine 210, 106371 (Oct 2021)

    Moya-Sáez, E., Peña-Nogales, Ó., Luis-García, R.D., Alberola-López, C.: A deep learning approach for synthetic MRI based on two routine sequences and train- ing with synthetic data. Computer Methods and Programs in Biomedicine 210, 106371 (Oct 2021). https://doi.org/10.1016/j.cmpb.2021.106371, https://linkinghub.elsevier.com/retrieve/pii/S0169260721004454

    work page doi:10.1016/j.cmpb.2021.106371 2021

  25. [25]

    PLOS One20(8), e0328867 (Aug 2025)

    Nasir, M., Xu, Y., Hasenstab, K., Yechoor, A., Dodhia, R., Weeks, W.B., Ferres, J.L., Cunha, G.M.: Liver MRI proton density fat fraction inference from con- trast enhanced CT images using deep learning: A proof-of-concept study. PLOS One20(8), e0328867 (Aug 2025). https://doi.org/10.1371/journal.pone.0328867, https://dx.plos.org/10.1371/journal.pone.0328867

    work page doi:10.1371/journal.pone.0328867 2025

  26. [26]

    IEEE Transactions on Medical Imaging 42(12), 3524–3539 (Dec 2023)

    Özbey, M., Dalmaz, O., Dar, S.U.H., Bedel, H.A., Özturk, Ş., Güngör, A., Çukur, T.: Unsupervised Medical Image Translation With Ad- versarial Diffusion Models. IEEE Transactions on Medical Imaging 42(12), 3524–3539 (Dec 2023). https://doi.org/10.1109/TMI.2023.3290149, https://ieeexplore.ieee.org/document/10167641/

    work page doi:10.1109/tmi.2023.3290149 2023

  27. [27]

    European Journal of Epidemiology37(10), 1107–1124 (2022)

    Peters, A., Consortium, G.N.C.N., et al.: Framework and baseline examination of the german national cohort (nako). European Journal of Epidemiology37(10), 1107–1124 (2022). https://doi.org/10.1007/s10654-022-00890-5

    work page doi:10.1007/s10654-022-00890-5 2022

  28. [28]

    BMC Medical Imaging25(1), 499 (2025)

    Pirwass, A., Glimm, B., Munz, M., Wilke, H.J.: Automatability and validity of methods for the quantification of intra-/intermuscular adipose tissue in con- ventional mri: A systematic review. BMC Medical Imaging25(1), 499 (2025). https://doi.org/10.1186/s12880-025-02037-w

    work page doi:10.1186/s12880-025-02037-w 2025

  29. [29]

    IEEE Transactions on Medical Imaging45(5), 2156– 2172 (2026)

    Rassmann, S., Kügler, D., Ewert, C., Reuter, M.: Regression is all you need for medical image translation. IEEE Transactions on Medical Imaging45(5), 2156– 2172 (2026). https://doi.org/10.1109/TMI.2025.3650412

    work page doi:10.1109/tmi.2025.3650412 2026

  30. [30]

    Journal of Magnetic Resonance Imaging36(5), 1011–1014 (2012)

    Reeder, S.B., Hu, H.H., Sirlin, C.B.: Proton density fat-fraction: A standardized MR-based biomarker of tissue fat concentration. Journal of Magnetic Resonance Imaging36(5), 1011–1014 (2012). https://doi.org/10.1002/jmri.23741

    work page doi:10.1002/jmri.23741 2012

  31. [31]

    High-Resolution Image Synthesis with Latent Diffusion Models

    Rombach, R., Blattmann, A., Lorenz, D., Esser, P., Ommer, B.: High- Resolution Image Synthesis with Latent Diffusion Models (Apr 2022). https://doi.org/10.48550/arXiv.2112.10752, http://arxiv.org/abs/2112.10752, arXiv:2112.10752 [cs]

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2112.10752 2022

  32. [32]

    Ronneberger, P

    Ronneberger, O., Fischer, P., Brox, T.: U-net: Convolutional networks for biomed- ical image segmentation. In: Navab, N., Hornegger, J., Wells, W.M., Frangi, A.F. (eds.) Medical Image Computing and Computer-Assisted Intervention – MICCAI 2015, Lecture Notes in Computer Science, vol. 9351, pp. 234–241. Springer (2015). https://doi.org/10.1007/978-3-319-24574-4_28

    work page doi:10.1007/978-3-319-24574-4_28 2015

  33. [33]

    In: International Conference on Learning Representations (ICLR) (2021), https://openreview.net/forum?id=St1giarCHLP

    Song, J., Meng, C., Ermon, S.: Denoising diffusion implicit models. In: International Conference on Learning Representations (ICLR) (2021), https://openreview.net/forum?id=St1giarCHLP

    2021

  34. [34]

    American Journal of Roentgenology221(5), 620–631 (2023)

    Wang, K., Cunha, G.M., Hasenstab, K., Henderson, W.C., Middleton, M.S., Cole, S.A., Umans, J.G., Ali, T., Hsiao, A., Sirlin, C.B.: Deep learning for inference of hepatic proton density fat fraction from T1-weighted in-phase and opposed-phase MRI: Retrospective analysis of population-based trial data. American Journal of Roentgenology221(5), 620–631 (2023)...

    work page doi:10.2214/ajr.23.29607 2023

  35. [35]

    Weißer, Linda: NAKO Gesundheitsstudie (Sep 2025), https://drks.de:443/search/de/trial/DRKS00037328

    2025

  36. [36]

    North American Spine Society journal17, 100313 (Mar 2024)

    Wesselink, E.O., Elliott, J.M., Pool-Goudzwaard, A., Coppieters, M.W., Pevenage, P.P., Di Ieva, A., Weber Ii, K.A.: Quantifying lumbar paraspinal intramuscular fat: Accuracy and reliability of automated thresholding A Fast U-Net for Paired Medical Image Translation 17 models. North American Spine Society journal17, 100313 (Mar 2024). https://doi.org/10.10...

    work page doi:10.1016/j.xnsj.2024.100313 2024

  37. [37]

    Scientific Reports 10(1), 3753 (Feb 2020)

    Yang, Q., Li, N., Zhao, Z., Fan, X., Chang, E.I.C., Xu, Y.: MRI Cross-Modality Image-to-Image Translation. Scientific Reports 10(1), 3753 (Feb 2020). https://doi.org/10.1038/s41598-020-60520-6, https://www.nature.com/articles/s41598-020-60520-6

    work page doi:10.1038/s41598-020-60520-6 2020