Pith. sign in

REVIEW 3 major objections 4 minor 67 references

Towards Virtual Clinical Trials of Radiology AI with Conditional Generative Modeling

T0 review · 3 major / 4 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read Synthetic CT patient cohorts generated by a conditional latent diffusion model can reproduce the performance degradation and bias that a radiology AI model exhibits on real out-of-distribution data.

desk verdict A real generative model and a convincing degradation-detection demo, but the quantitative 'replication' of real error distributions depends on an oracle culling step that the deployment scenario won't have. read the letter →

arxiv 2502.09688 v1 pith:TC3OEK4G submitted 2025-02-13 cs.CV cs.AIcs.LG

Benjamin D. Killeen , Bohua Wan , Aditya V. Kulkarni , Nathan Drenkow , Michael Oberst , Paul H. Yi , Mathias Unberath This is my paper · ORCID
classification cs.CVcs.AIcs.LG
keywords virtualclinicaltrialsconditionalgenerativemodelinglatentdiffusionfull-bodyCTsynthesisradiologyAIrobustnessalgorithmicauditingbodycompositionmeasurementout-of-distributiongeneralization
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper aims to replace costly real-image clinical trials for radiology AI with virtual trials run on synthetic CT scans. It claims that a conditional generative model trained on full-body CTs can produce synthetic patient cohorts whose attribute distributions match a target population, and that running a downstream AI model on those cohorts reproduces the degradation and bias the model would show on real out-of-distribution patients. In tests on body-fat and muscle-mass regression, synthetic cohorts matched the real mean absolute error boundaries (under 2% in-distribution, over 3% out-of-distribution) and identified the same attributes as driving error. If correct, this offers a scalable way to audit AI models for bias and degradation before deployment.

What carries the argument

The central object is a conditional latent diffusion model over the joint latent space of a stacked VQ-VAE image encoder and a segmentation encoder: the diffusion model samples $Z=[Z_{\mathrm{img}},Z_{\mathrm{seg}}]$ conditioned on categorical attribute embeddings for sex, age, height, and weight, then decoders produce a synthetic CT and its segmentation. The stacked autoencoder compresses 3D volumes patch-wise to keep memory feasible, a multi-window L1 loss reweights soft versus hard tissue contributions, classifier-free guidance drops attributes with 20% probability during training, and a post-processing 3D U-Net restores high-frequency detail. Re-biasing the synthetic cohort by the same in-distribution/out-of-distribution attribute boundary is the step that makes the error distributions align with real images.

What would settle it

Train the generator and downstream models on one hospital population, then run the VCT for a second hospital population without allowing any of the second population's real images or its in-distribution/out-of-distribution boundary into the pipeline; if the synthetic MAE no longer matches the real out-of-distribution MAE, or if the synthetic bias ranking diverges from the real one, the central replication claim is refuted.

Watch

Extended reading notes

Core claim

The paper's central claim is that learning the joint distribution $p(X,Y|a)$ of full-body CT images $X$ and organ segmentations $Y$, conditioned on patient attributes $a$, yields a generative model whose synthetic cohorts are realistic and controllable enough to stand in for real patient populations in virtual clinical trials. Concretely, when the downstream body-fat-percentage and muscle-mass-percentage regressors are trained with a body-volume shortcut, VCTs on synthetic images detect the same in-distribution to out-of-distribution degradation as real images, with MAE below 2% on the in-distribution population and above 3% on the out-of-distribution population after re-biasing. The paper further shows that the attribute-level structure of the error is replicated: feature-importance rankings of error drivers correlate at 0.993 and 0.919 with real images for the two tasks, positioning the model as the first CT generative model with sufficient realism, anatomical consistency, and conditioning fidelity for full-body virtual clinical trials.

Load-bearing premise

The replication result assumes the generative model has been trained on data that overlaps the target population and that the synthetic cohort is filtered with the same in-distribution/out-of-distribution boundary that defines the real test split; neither condition is available when running a VCT for a genuinely new deployment population.

Editorial extensions

If this is right

  • Synthetic cohorts can probe a radiology AI model before deployment: the VCT predicted acceptable in-distribution performance (MAE under 2%) and clinically significant out-of-distribution degradation (MAE over 3%) for both body-fat and muscle-mass regression.
  • Conventional importance weighting on real in-distribution data failed to reveal the degradation (estimated 1.31% versus true 3.66% for body-fat percentage), while the synthetic VCT did, so VCTs provide information that attribute reweighting cannot.
  • The same attribute-level error analysis on synthetic images ranks the patient attributes driving model error in nearly the same order as real images, supporting the use of VCTs for algorithmic auditing of bias-inducing attributes.
  • If the generator has seen the relevant population, a vendor, hospital, or regulator can test a model on arbitrary attribute distributions without collecting or annotating new real scans, shifting the burden of data collection to a centralized entity.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The re-biasing step used to match real error distributions depends on knowing the real in-distribution/out-of-distribution decision boundary; in a true deployment VCT that boundary is unknown, so a natural next step is to infer or marginalize over plausible boundaries rather than assume one.
  • The assumption that the generator was trained on data overlapping the target population implies the method currently validates models for populations similar to the training data; whether it extends to genuinely unseen populations hinges on the emergent generalization the authors flag as an open question.
  • The conditioning machinery could be extended beyond demographics to text-based conditioning on diagnoses or history, though the paper notes the required scale is much larger and natural-language conditioning may overstate the true diversity of supported attributes.
  • A practical sensitivity test emerging from the paper is to compare VCT-predicted error with and without the re-biasing filter; the reported unmatched out-of-distribution muscle-mass comparison (p = 0.026) indicates the replication claim is sensitive to this filtering step.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 4 minor

Summary. The manuscript proposes a conditional latent diffusion model for full-body CT synthesis, conditioned on sex, age, height, and weight, and jointly modeling image and segmentation latents. It evaluates realism with FID (5.97), organ-level Dice (0.727), and organ volume/centroid correlations; it evaluates conditioning fidelity; and it presents virtual clinical trial (VCT) experiments in which two downstream models (body fat percentage and muscle mass percentage regression) are trained with a body-volume shortcut and then tested on in-distribution (ID) and out-of-distribution (OOD) real images and on synthetic cohorts with matched attributes. The central claim is that the synthetic cohorts can detect model degradation and replicate the real error distributions, with quantitative alignment reported after a re-biasing step.

Significance. The technical contribution is substantial: full-body conditional CT generation with joint anatomical modeling at this resolution is an important engineering advance, and the controlled demonstration that synthetic cohorts can separate ID from OOD degradation (roughly MAE below 2% vs above 3%) is a valuable proof of concept. The realism metrics, including organ-level Dice and volume/centroid correlations, are concrete and useful. However, the strongest claims—precise replication of real error distributions and the deployment-oriented VCT protocol—are not established, because the replication relies on an oracle re-biasing step that uses the real ID/OOD decision boundary, and because the generator is trained on the same image pool from which the real OOD test images are drawn. The current value of the paper is primarily in the generative model and the degradation-detection experiment, not in quantitative distributional replication.

major comments (3)
  1. [1.5.1 and Table 1] The claim that 'the MAE on synthetic images aligns with that of real images' holds only after the re-biasing step, which culls synthetic images using the real ID/OOD decision boundary shown in Fig. 6 and described in Section 4.3.1. Before re-biasing, the MMP OOD synthetic MAE (4.34 [3.86, 4.95]) differs significantly from the real OOD MAE (5.54 [5.02, 6.16]) with Z = -3.44 and p = 0.026; the re-biased result (n = 80 after removing 70 of 150 samples) is the one reported as aligning (p = 0.324). This procedure uses the exact real test-set information that a VCT is supposed to predict, so the quantitative replication claim is not a prediction from conditioning attributes alone; it is a post hoc selection. The sentence in Section 1.5.1 that the method works 'as long as the conditioning attributes are sufficient to reproduce the biasing attributes' identifies the unverified condition, and the re-biasing step bypasses rather than tests it. The abstract and conclusion should be revised to separate the demonstrated degradation-detection result from the unsupported distribution-replication claim.
  2. [4.2.1 and Discussion] The generator's training data are derived from the same 798-scan NMDID pool that is used to create the downstream ID and OOD test sets, so the real OOD images are within the generator's training distribution. The Discussion acknowledges this, but the abstract and conclusion overstate the generality of the results. As presented, the VCT demonstration is in-distribution for the generator, and the experiments provide no evidence that synthetic cohorts generated for an unseen target population would replicate real error distributions. This limitation should be reflected in the main claims, not only in the Discussion.
  3. [1.2, 1.5.1, and Table 1] The statistical evidence for distributional replication is reported as p > 0.05 or p > 0.3 for a two-sample Z-test, but failure to reject the null is not evidence that distributions are the same, and with n between 75 and 150 these tests have limited power. For example, the BFP ID synthetic comparison has p = 0.103 before re-biasing and the MMP ID comparison has p = 0.063; these are not strong confirmations of alignment, yet the text describes them as showing alignment. The manuscript should use an equivalence testing framework or report effect sizes and confidence bounds on distributional differences, and it should avoid stating that a p-value above 0.05 indicates a high probability that the errors come from the same distribution.
minor comments (4)
  1. [Abstract and Section 4.1.2] There are typos that should be corrected, including 'being being' in the introduction, 'Frech´ et' for Fréchet, and 'dicussed' in Section 4.1.2.
  2. [Figure 2 caption] The caption writes that the autoencoders compress the image and segmentation to latent embeddings 'Zimg and Zimg, respectively'; the second symbol should be Zseg.
  3. [Table numbering] The anatomical consistency table in Section 1.3 and the VCT performance table in Section 1.5.1 are both labeled Table 1; renumber the tables consistently.
  4. [Section 4.3.1] The sentence 'The backbone image encoder processes a sagittal and coronal loss' should read 'sagittal and coronal slice' (or 'image'); the current phrasing is confusing.

Circularity Check

1 steps flagged · score 6.0 of 10

Quantitative replication of real error distributions is achieved only after re-biasing synthetic images with the real ID/OOD decision boundary; uncompensated MMP OOD synthetic errors do not match real (p=0.026).

  1. fitted input called prediction [Section 1.5.1 and Table 1 (MMP OOD rows)]
    "To test this hypothesis, we re-bias the synthetic data in the same manner as the real images, by culling synthetic images outside the specified distribution. This results in a close match between the distribution of absolute errors on synthetic and real images, across both the ID and OOD data in both tasks."

    The claimed alignment of synthetic and real error distributions is obtained only after culling synthetic images with the real ID/OOD decision boundary, which is exactly the population-shift information a VCT is supposed to provide from attributes alone. Before re-biasing, the MMP OOD synthetic MAE (4.34 [3.86, 4.95]) differs significantly from the real MAE (5.54 [5.02, 6.16]), with p=0.026; after excluding 70 of 150 samples using the real boundary, the MAE becomes 5.10 [4.64, 5.65] with p=0.324. The re-biased result is therefore selected to match the real labels, not predicted from the conditioning attributes.

full rationale

The paper contains substantial independent content: the FID score, anatomical consistency checks (Dice 0.727, organ volume and centroid correlations), attribute fidelity measurements, and the independently trained downstream model are genuine external evaluations, and no load-bearing argument rests on a self-citation chain. The main circularity is confined to the quantitative replication claim in Section 1.5.1 and Table 1. There, the statement that synthetic-image MAE distributions 'align' with real-image distributions is obtained only after 're-biasing' synthetic images by culling with the real ID/OOD decision boundary. For MMP OOD, the synthetic MAE before re-biasing is 4.34 [3.86, 4.95] versus the real 5.54 [5.02, 6.16], with p=0.026; after culling 70 of 150 samples using the real boundary it becomes 5.10 [4.64, 5.65], p=0.324. This is selection using the very real-world labels a VCT is supposed to supply, rather than a prediction from the conditioning attributes; the paper's caveat that conditioning attributes must be sufficient to reproduce the biasing attributes is bypassed, not tested. The degradation-detection claim (ID error below 2% versus OOD error above 3%) does have independent content, since it holds even before re-biasing, which is why the circularity is partial rather than total. The Discussion honestly scopes the assumption that the generative model was exposed to data with attributes similar to the target population, but the abstract and conclusion overstate the generality of the replication claim.

Assumptions & free parameters 4 free parameters · 4 assumptions · 0 invented entities

The central claim rests on two domain assumptions (TotalSegmentator and HU-to-density accuracy) and one ad hoc assumption specific to this protocol (the generator must have seen data similar to the target population). The free parameters are architectural and conditioning choices that affect image realism and conditioning fidelity, which are necessary for the VCT demonstration. No new physical entities or unobserved mediators are introduced.

free parameters (4)
  • Multi-window loss weights and HU thresholds = lambda_soft, lambda_hard, and lambda_other not reported; HU bounds described only as 'chosen'
    Equation (3) reweights the reconstruction loss for soft and hard tissue; these hand-chosen values affect image realism and therefore the downstream error replication that the central claim depends on.
  • VQ-VAE codebook size = 4096
    Selected by hyperparameter search over codebook sizes (Table A3); it affects reconstruction quality, which is a prerequisite for realistic synthetic cohorts.
  • Stacked autoencoder number of levels Lae = 2
    Selected by ablation (Table A2); the two-level stack improves PSNR and SSIM over a vanilla VQ-VAE, and the FID comparison depends on this architectural choice.
  • Conditioning attribute bin boundaries = e.g., male, 50-60 years, 170-180 cm, 60-70 kg (Fig. 3c)
    The demographic attributes are discretized into categorical bins by the authors; conditioning fidelity (Section 1.4) is measured relative to these bins, so the bin boundaries directly influence the evaluation of how well the model replicates patient populations.
assumptions (4)
  • domain assumption TotalSegmentator provides accurate segmentations of organs and tissue types for both real and synthetic CT images.
    Segmentations from TotalSegmentator are used to compute ground-truth body fat and muscle mass percentages (Section 1.5), to evaluate organ-volume realism (Section 1.3), and to measure conditioned attributes (Section 1.4). If this tool is biased, all derived ground truths share the bias and the comparison between real and synthetic is not externally anchored.
  • domain assumption The HU-to-density conversion (Eq. B1) maps CT intensities to physical mass accurately.
    All body composition measurements in Appendix B rely on this conversion; a systematic error would shift absolute BFP and MMP values, though it may partially cancel when comparing real and synthetic images.
  • ad hoc to paper A generator trained on data with attributes similar to the target population can synthesize valid test cohorts for that population.
    Explicitly assumed in the Discussion ('we make the assumption that the generative AI model is capable of faithfully representing samples from the relevant patient attributes...'). The experimental setup satisfies this by training the generator on the same NMDID pool from which the test images are drawn, which is a load-bearing condition for the central claim.
  • standard math Standard diffusion, VQ-VAE, and classifier-free guidance formulations are valid.
    The generative model relies on established theory from Esser et al., Rombach et al., and Ho and Salimans; these are background results not re-derived in this paper.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Towards Virtual Clinical Trials of Radiology AI with Conditional Generative Modeling." pith.science (2026). https://pith.science/paper/TC3OEK4G

@misc{pith2026250209688,
  author       = {Pith},
  title        = {Pith review of: Towards Virtual Clinical Trials of Radiology AI with Conditional Generative Modeling},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/TC3OEK4G}},
  note         = {Machine review of arXiv:2502.09688}
}
read the original abstract

Artificial intelligence (AI) is poised to transform healthcare by enabling personalized and efficient care through data-driven insights. Although radiology is at the forefront of AI adoption, in practice, the potential of AI models is often overshadowed by severe failures to generalize: AI models can have performance degradation of up to 20% when transitioning from controlled test environments to clinical use by radiologists. This mismatch raises concerns that radiologists will be misled by incorrect AI predictions in practice and/or grow to distrust AI, rendering these promising technologies practically ineffectual. Exhaustive clinical trials of AI models on abundant and diverse data is thus critical to anticipate AI model degradation when encountering varied data samples. Achieving these goals, however, is challenging due to the high costs of collecting diverse data samples and corresponding annotations. To overcome these limitations, we introduce a novel conditional generative AI model designed for virtual clinical trials (VCTs) of radiology AI, capable of realistically synthesizing full-body CT images of patients with specified attributes. By learning the joint distribution of images and anatomical structures, our model enables precise replication of real-world patient populations with unprecedented detail at this scale. We demonstrate meaningful evaluation of radiology AI models through VCTs powered by our synthetic CT study populations, revealing model degradation and facilitating algorithmic auditing for bias-inducing data attributes. Our generative AI approach to VCTs is a promising avenue towards a scalable solution to assess model robustness, mitigate biases, and safeguard patient care by enabling simpler testing and evaluation of AI models in any desired range of diverse patient populations.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

67 extracted references · 63 canonical work pages

  1. [1]

    Titano, J. J. et al. Automated deep-neural-network surveillance of cranial images for acute neurologic events. Nature medicine 24, 1337–1341 (2018)

    work page 2018

  2. [2]

    Levin, S. et al. Machine-learning-based electronic triage more accurately differ- entiates patients with respect to clinical outcomes compared with the emergency severity index. Annals of emergency medicine 71, 565–574 (2018)

    work page 2018

  3. [3]

    Liu, X. et al. A comparison of deep learning performance against health-care professionals in detecting diseases from medical imaging: a systematic review and meta-analysis. The lancet digital health 1, e271–e297 (2019)

    work page 2019

  4. [4]

    Pickhardt, P. J. et al. Opportunistic screening for osteoporosis using abdominal computed tomography scans obtained for other indications. Annals of internal medicine 158, 588–595 (2013)

    work page 2013

  5. [5]

    Jang, S. et al. Opportunistic osteoporosis screening at routine abdominal and tho- racic ct: normative l1 trabecular attenuation values in more than 20 000 adults. Radiology 291, 360–367 (2019)

    work page 2019

  6. [6]

    Eng, D. et al. Automated coronary calcium scoring using deep learning with multicenter external validation. NPJ digital medicine 4, 88 (2021)

    work page 2021

  7. [7]

    Oh, S. et al. Evaluation of deep learning-based quantitative computed tomogra- phy for opportunistic osteoporosis screening. Sci. Rep. 14, 1–9 (2024)

    work page 2024

  8. [8]

    Zopfs, D. et al. Evaluating body composition by combining quantitative spectral detector computed tomography and deep learning-based image segmentation. Eur. J. Radiol. 130, 109153 (2020)

    work page 2020

Show all 67 references
  1. [9]

    S., Mustapha, M

    Ozsahin, I., Sekeroglu, B., Musa, M. S., Mustapha, M. T. & Ozsahin, D. U. Review on Diagnosis of COVID-19 from Chest CT Images Using Artificial Intelligence. Comput. Math. Methods Med. 2020, 9756518 (2020). 26

    2020

  2. [10]

    Murugesan, M. et al. A Hybrid deep learning model for effective segmentation and classification of lung nodules from CT images. J. Intell. Fuzzy Syst. 42, 2667–2679 (2022)

    2022

  3. [11]

    Mohammadi, S. et al. Deep learning-based detection of coronary artery calcifica- tion in non-contrast and contrast-enhanced CT scans (2024). [Online; accessed

    2024

  4. [12]

    & Mesk´ o, B

    Benjamens, S., Dhunnoo, P. & Mesk´ o, B. The state of artificial intelligence-based fda-approved medical devices and algorithms: an online database. NPJ digital medicine 3, 118 (2020)

    2020

  5. [13]

    URL https://www.fda.gov/medical-devices/software-medical-device-samd/ artificial-intelligence-and-machine-learning-aiml-enabled-medical-devices

    Artificial Intelligence and Machine Learning (AI/ML)-Enabled Medical D (2024). URL https://www.fda.gov/medical-devices/software-medical-device-samd/ artificial-intelligence-and-machine-learning-aiml-enabled-medical-devices. [Online; accessed 27. Jan. 2025]

    2024

  6. [14]

    & Unberath, M

    Drenkow, N., Sani, N., Shpitser, I. & Unberath, M. A systematic review of robustness in deep learning for computer vision: Mind the gap? arXiv preprint arXiv:2112.00639 (2021)

    2021 arXiv

  7. [15]

    Ong Ly, C. et al. Shortcut learning in medical AI hinders generalization: method for estimating AI model generalization without external data. npj Digital Medicine 7, 1–10 (2024)

    2024

  8. [16]

    Drenkow, N. G. & Unberath, M. RobustCLEVR: A benchmark and framework for evaluating robustness in object-centric learning. Proc. IEEE Workshop Appl. Comput. Vis. 4506–4515 (2023)

    2023

  9. [17]

    Rodrigues, G. et al. Automated large artery occlusion detection in stroke: a single- center validation study of an artificial intelligence algorithm. Cerebrovascular Diseases 51, 259–264 (2022)

    2022

  10. [18]

    Rava, R. A. et al. Validation of an artificial intelligence-driven large vessel occlu- sion detection algorithm for acute ischemic stroke patients. The Neuroradiology Journal 34, 408–417 (2021)

    2021

  11. [19]

    Matsoukas, S. et al. Ai software detection of large vessel occlusion stroke on ct angiography: a real-world prospective diagnostic test accuracy study. Journal of Neurointerventional Surgery 15, 52–56 (2023)

    2023

  12. [20]

    Wong, A. et al. External validation of a widely implemented proprietary sepsis prediction model in hospitalized patients. JAMA internal medicine 181, 1065– 1070 (2021)

    2021

  13. [21]

    F., Meram, E., Garrett, J

    Voter, A. F., Meram, E., Garrett, J. W. & John-Paul, J. Y. Diagnostic accu- racy and failure mode analysis of a deep learning algorithm for the detection 27 of intracranial hemorrhage. Journal of the American College of Radiology 18, 1143–1152 (2021)

    2021

  14. [22]

    & Kunst, M

    Small, J., Osler, P., Paul, A. & Kunst, M. Ct cervical spine fracture detection using a convolutional neural network. American Journal of Neuroradiology 42, 1341–1347 (2021)

    2021

  15. [23]

    Kunst, M. et al. Real-world performance of large vessel occlusion cadt ai algorithms-what the stroke team needs to know. Journal of the American College of Radiology: JACR S1546–1440 (2023)

    2023

  16. [24]

    Wu, E. et al. How medical AI devices are evaluated: limitations and rec- ommendations from an analysis of FDA approvals. Nat. Med. 27, 582–584 (2021)

    2021

  17. [25]

    Lam, T. Y. T. et al. Randomized Controlled Trials of Artificial Intelligence in Clinical Practice: Systematic Review. J. Med. Internet Res. 24, e37188 (2022)

    2022

  18. [26]

    Oakden-Rayner, L., Dunnmon, J., Carneiro, G. & Re, C. Hidden stratifica- tion causes clinically meaningful failures in machine learning for medical imaging (2020)

    2020

  19. [27]

    URL https://arpa-h.gov/news-and-events/ arpa-h-launches-program-help-ai-enabled-medical-tools-maintain-peak-performance

    ARPA-H launches program to help AI-enabled medical tools maintain peak performance (2024). URL https://arpa-h.gov/news-and-events/ arpa-h-launches-program-help-ai-enabled-medical-tools-maintain-peak-performance. [Online; accessed 3. Dec. 2024]

    2024

  20. [28]

    & Topol, E

    Rajpurkar, P., Chen, E., Banerjee, O. & Topol, E. J. AI in health and medicine. Nat. Med. 28, 31–38 (2022)

    2022

  21. [29]

    P., Mahesh, M., Beck, T

    Segars, W. P., Mahesh, M., Beck, T. J., Frey, E. C. & Tsui, B. M. W. Realistic CT simulation using the 4D XCAT phantom. Med. Phys. 35, 3800–3808 (2008)

    2008

  22. [30]

    Abadi, E. et al. Virtual clinical trials in medical imaging: a review. J. Med. Imaging 7 (2020)

    2020

  23. [31]

    Badano, A. et al. The stochastic digital human is now enrolling for in silico imaging trials—methods and tools for generating digital cohorts. Prog. Biomed. Eng. 5, 042002 (2023)

    2023

  24. [32]

    Ibrahim, M. et al. Generative AI for Synthetic Data Across Multiple Medical Modalities: A Systematic Review of Recent Developments and Challenges. arXiv (2024)

    2024

  25. [33]

    Hung, A. L. Y. et al. Med-cDiff: Conditional Medical Image Generation with Diffusion Models. Bioengineering 10, 1258 (2023). 28

    2023

  26. [34]

    Khader, F. et al. Denoising diffusion probabilistic models for 3D medical image generation. Sci. Rep. 13, 1–12 (2023)

    2023

  27. [35]

    Chen, W. et al. Medical Image Synthesis via Fine-Grained Image-Text Alignment and Anatomy-Pathology Prompting (2024)

    2024

  28. [36]

    Gao, C. et al. Synthetic data accelerates the development of generalizable learning-based algorithms for x-ray image analysis. Nature Machine Intelligence (2023)

    2023

  29. [37]

    Hennessy, D. A. et al. The Population Health Model (POHEM): an overview of rationale, methods and applications. Popul. Health Metrics 13, 1–12 (2015)

    2015

  30. [38]

    Guo, P. et al. MAISI: Medical AI for Synthetic Imaging. arXiv (2024)

    2024

  31. [39]

    & Ommer, B

    Esser, P., Rombach, R. & Ommer, B. Taming transformers for high-resolution image synthesis (2020). 2012.09841

    2020 arXiv

  32. [40]

    & Ommer, B

    Rombach, R., Blattmann, A., Lorenz, D., Esser, P. & Ommer, B. High- Resolution Image Synthesis With Latent Diffusion Models (2022). URL https: //openaccess.thecvf.com/content/CVPR2022/html/Rombach High-Resolution Image Synthesis With Latent Diffusion Models CVPR 2022 paper.htm...

    2022

  33. [41]

    Wasserthal, J. et al. TotalSegmentator: Robust Segmentation of 104 Anatomic Structures in CT Images. Radiology: Artificial Intelligence (2023). URL https: //pubs.rsna.org/doi/10.1148/ryai.230024

    2023 doi

  34. [42]

    & Hochreiter, S

    Heusel, M., Ramsauer, H., Unterthiner, T., Nessler, B. & Hochreiter, S. GANs Trained by a Two Time-Scale Update Rule Converge to a Local Nash Equilibrium. Advances in Neural Information Process- ing Systems 30 (2017). URL https://papers.nips.cc/paper/2017/hash/ 8a1d694707eb0fe...

    2017

  35. [43]

    & O’Reilly, J

    Asadi, F. & O’Reilly, J. A. Artificial computed tomography images with progressively growing generative adversarial network 1–5 (2021)

    2021

  36. [44]

    O’Reilly, J. A. & Asadi, F. Pre-trained vs. random weights for calculating fr´ echet inception distance in medical imaging 1–4 (2021)

    2021

  37. [45]

    Gatidis, S. et al. A whole-body FDG-PET/CT Dataset with manually annotated Tumor Lesions. Sci. Data 9, 1–7 (2022)

    2022

  38. [46]

    Leichter, I. et al. The effect of age and sex on bone density, bone mineral content and cortical index. Clinical Orthopaedics and Related Research ® 156, 232–239 (1981). 29

    1981

  39. [47]

    Bates, D. D. B., Pickhardt, P. J., Bates, D. D. B. & Pickhardt, P. J. CT-Derived Body Composition Assessment as a Prognostic Tool in Oncologic Patients: From Opportunistic Research to Artificial Intelligence–Based Clinical Implementation. Am. J. Roentgenol. (2022)

    2022

  40. [48]

    F., Kohl, S

    Isensee, F., Jaeger, P. F., Kohl, S. A. A., Petersen, J. & Maier-Hein, K. H. nnU-Net: a self-configuring method for deep learning-based biomedical image segmentation. Nat. Methods 18, 203–211 (2021)

    2021

  41. [49]

    Avesta, A. et al. Comparing 3D, 2.5D, and 2D Approaches to Brain Image Auto-Segmentation. Bioengineering 10, 181 (2023)

    2023

  42. [50]

    Stevens, J., Ou, F.-S., Cai, J., Heymsfield, S. B. & Truesdale, K. P. Prediction of percent body fat measurements in Americans 8 years and older. Int. J. Obes. 40, 587–594 (2016)

    2016

  43. [51]

    & Lekamwasam, S

    Rathnayake, N., Alwis, G., Lenora, J. & Lekamwasam, S. Development and Cross-Validation of Anthropometric Predictive Equations to Estimate Total Body Fat Percentage in Adult Women in Sri Lanka. Journal of Obesity 2020, 2087346 (2020)

    2020

  44. [52]

    Talma, H. et al. Bioelectrical impedance analysis to estimate body composition in children and adolescents: a systematic review and evidence appraisal of validity, responsiveness, reliability and measurement error. Obes. Rev. 14, 895–905 (2013)

    2013

  45. [53]

    URL https://www.cdc.gov/mmwr/preview/mmwrhtml/mm5751a4.htm

    QuickStats: Mean Percentage Body Fat, ∗ by Age Group and Sex — National Health and Nutrition Examination Survey, United States, 1999–2004 (2009). URL https://www.cdc.gov/mmwr/preview/mmwrhtml/mm5751a4.htm. [Online; accessed 4. Feb. 2025]

    2009

  46. [54]

    B., Wang, Z

    Janssen, I., Heymsfield, S. B., Wang, Z. & Ross, R. Skeletal muscle mass and distribution in 468 men and women aged 18–88 yr. J. Appl. Physiol. (2000)

    2000

  47. [55]

    Pavlak, M., Drenkow, N., Petrick, N., Farhangi, M. M. & Unberath, M. Data AUDIT: Identifying Attribute Utility- and Detectability-Induced Bias in Task Models (2023)

    2023

  48. [56]

    Cho, J. et al. MediSyn: Text-Guided Diffusion Models for Broad Medical 2D and 3D Image Synthesis. arXiv (2024)

    2024

  49. [57]

    Van Den Oord, A., Vinyals, O. et al. Neural discrete representation learning. Advances in neural information processing systems 30 (2017)

    2017

  50. [58]

    Monai: Medical open network for ai (2024)

    Consortium, M. Monai: Medical open network for ai (2024). URL https://doi. org/10.5281/zenodo.13942962. 30

    2024 doi

  51. [59]

    & Fei-Fei, L

    Johnson, J., Alahi, A. & Fei-Fei, L. Leibe, B., Matas, J., Sebe, N. & Welling, M. (eds) Perceptual losses for real-time style transfer and super-resolution . (eds Leibe, B., Matas, J., Sebe, N. & Welling, M.) Computer Vision – ECCV 2016 , 694–711 (Springer International Publis...

    2016

  52. [60]

    & Salimans, T

    Ho, J. & Salimans, T. Classifier-Free Diffusion Guidance. arXiv (2022)

    2022

  53. [61]

    Edgar, H. et al. New mexico decedent image database. Office of the Medical Investigator, University of New Mexico (2020)

    2020

  54. [62]

    & Hutter, F

    Loshchilov, I. & Hutter, F. Decoupled weight decay regularization (2019). URL https://openreview.net/forum?id=Bkg6RiCqY7

    2019

  55. [63]

    Paszke, A. et al. PyTorch: An Imperative Style, High-Performance Deep Learning Library. arXiv (2019)

    2019

  56. [64]

    Liu, Z. et al. Swin transformer: Hierarchical vision transformer using shifted windows (2021)

    2021

  57. [65]

    & Zelnik-Manor, L

    Ridnik, T., Ben-Baruch, E., Noy, A. & Zelnik-Manor, L. ImageNet-21K Pretraining for the Masses. arXiv (2021)

    2021

  58. [66]

    On the molecular relationship between Hounsfield Unit (HU), mass density, and electron density in computed tomography (CT)

    Sudhyadhom, A. On the molecular relationship between Hounsfield Unit (HU), mass density, and electron density in computed tomography (CT). PLoS One 15, e0244861 (2020)

    2020

  59. [67]

    & Chen, W.-Y

    Wang, C.-M. & Chen, W.-Y. The human-height measurement scheme by using image processing techniques (2012). 31 Table 1: AI model performance on biased populations. For each task, the ID test set comes from the same distribution as the training set, while the out-of-distribution...

    2012

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.