Pith. sign in

REVIEW 4 major objections 5 minor 1 cited by

TauGenNet: Plasma-Driven Tau PET Image Synthesis via Text-Guided 3D Diffusion Models

T0 review · 4 major / 5 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read A text-guided 3D diffusion model generates tau PET brain images from an MRI and a plasma p-tau217 measurement, reproducing the known spatial trajectory of Alzheimer's tau pathology.

desk verdict A credible but under-evidenced application of 3D latent diffusion to tau PET synthesis; the evaluation only supports group-level templates, not individualized clinical utility. read the letter →

arxiv 2509.04269 v1 pith:6ZN3SKQS submitted 2025-09-04 cs.CV

Yuxin Gong , Se-in Jang , Wei Shao , Yi Su , Kuang Gong (for the Alzheimer's Disease Neuroimaging Initiative (ADNI)) This is my paper
classification cs.CV
keywords tauPETsynthesis3Dlatentdiffusiontext-guidedgenerationplasmap-tau217Alzheimer'sdiseaseMRI-guideddataaugmentationprogressionsimulation
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

This paper tries to show that a 3D diffusion model can synthesize tau PET brain images from two cheaper, non-invasive inputs: a structural MRI and a single blood biomarker value, plasma p-tau217. The motivation is practical, because tau PET is scarce and expensive; if synthesis works, it can fill missing data, augment machine-learning datasets, and simulate how tau pathology spreads as the biomarker rises. The central claim is that the generated volumes reproduce the known spatial trajectory of Alzheimer's tau—early signal in medial temporal regions, later spread to cortex—and that regional uptake at the group level tracks real PET across plasma ranges. The paper explicitly stops short of claiming the images are suitable for clinical diagnosis.

What carries the argument

The engine is a three-stage 3D latent diffusion model. A shared 3D autoencoder maps both tau PET and registered MRI into a compact latent space; the MRI latent stays clean and is concatenated channel-wise with the noised tau PET latent; a U-Net denoiser with layer-wise cross-attention takes a CLIP text embedding of the p-tau217 value as a biomarker prompt. The cross-attention is what lets a one-dimensional plasma value modulate brain-wide generation, while the concatenated MRI latent supplies anatomy. This mechanism makes the generative prior a learned mapping from 'anatomy + scalar biomarker' to a tau distribution.

What would settle it

Take a held-out subject with low p-tau217 but focal tau in an atypical region (or different amyloid status) and generate from their MRI; if the synthetic volume misses that focal pathology while a per-subject SUVR error metric stays high, the claim of individualized synthesis fails. More simply, recompute regional error without averaging within plasma bins: group-mean regression predicts near-zero MSE, but per-subject errors would expose how much individual signal is actually preserved.

Watch

Extended reading notes

Core claim

On its own terms, the paper reports that text-guided 3D latent diffusion, conditioned by concatenating the MRI latent with the noisy tau PET latent and by cross-attending to a CLIP embedding of the text prompt 'Plasma is [p-tau217 value]', produces tau PET volumes whose group-averaged regional tau values match real PET across six AD-relevant regions. Adding the MRI as an anatomical prior sharply reduces regional error compared to plasma-only conditioning. Visual and surface-map analyses show the generated images reproduce the canonical entorhinal-to-temporal-to-parietal spread as p-tau217 increases. The paper frames this as a first use of p-tau217 for tau PET synthesis, with implications for

Load-bearing premise

The model assumes that an individual's tau pattern is fully determined by their brain shape and one blood value, learned from 260 training pairs; if real tau patterns vary across people in ways those two inputs don't capture, the generated images will reproduce only the group average.

Editorial extensions

If this is right

  • If correct, tau PET data can be augmented in multi-cohort studies where PET is missing, improving machine-learning model training.
  • Varying the p-tau217 prompt while keeping a fixed MRI yields a simulated disease-trajectory sequence showing where tau would appear at later stages.
  • The plasma-only vs MRI+plasma comparison indicates that anatomical structure is essential; blood values alone cannot localize tau.
  • Generated group statistics are close enough to reproduce biological associations like Braak-like spread, supporting in-silico biomarker-association studies.
  • The framework can in principle accept other text-encoded clinical variables, such as cognitive scores, through the same CLIP prompt channel.

Reading between the lines

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

  • The paper's evaluation averages individual regional values within plasma bins before computing error, so the reported near-zero MSE partly reflects regression to the group mean; per-subject regional error remains untested and is likely much larger.
  • Because training uses one scalar blood value and a rigidly registered MRI per subject, the model may collapse individual heterogeneity—amyloid status, atypical variants, deformable anatomy—into a mean template; this should be probed with per-subject fidelity metrics.
  • The same text-conditioning mechanism could be inverted: instead of synthesizing PET from blood, one could search for the plasma prompt that best reproduces a given individual's PET pattern, turning the generator into a biomarker-to-topography probe.
  • If larger multi-site cohorts become available, group-level MSE improvements may not translate to clinical utility without explicit calibration; a reader study comparing synthetic and real images would be the decisive next test.
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

4 major / 5 minor

Summary. The paper proposes TauGenNet, a latent 3D diffusion model that synthesizes tau PET volumes conditioned on rigidly registered T1-weighted MRI and on plasma p-tau217 values encoded into a CLIP text prompt. The method is trained on 260 ADNI pairs and tested on 100 held-out pairs. Quantitative evaluation computes region-wise mean SUVR within manually defined p-tau217 bins and reports the squared difference between real and generated group means (Eq. 16-18), supplemented by boxplots and averaged cortical surface maps. The paper claims that the generated tau PET images are realistic, capture AD progression patterns consistent with the known trajectory of tau pathology, and can support data augmentation and in-silico progression simulation.

Significance. If the claims were fully substantiated, the contribution would be valuable: tau PET is expensive and scarce, and a non-invasive synthesis route combining anatomical MRI and plasma biomarkers could support missing-data imputation, dataset augmentation, and hypothesis-driven simulation. The use of a 3D latent diffusion model with cross-attention on a CLIP-embedded clinical scalar is a reasonable design, and building on p-tau217 rather than MMSE is a useful step beyond prior 2D work. However, the current quantitative evidence does not establish per-subject fidelity or clinical utility. The central evaluation is group-level and can be satisfied by bin-wise mean templates; no trivial baseline, confidence interval, or external comparator is provided, and the generated variance is visibly lower than the real variance. The paper is therefore potentially significant but currently under-validated. I also note that no code, checkpoints, or preprocessing scripts are provided, which limits reproducibility.

major comments (4)
  1. [Section III-B, Eq. (16)-(18)] The primary quantitative metric averages individual regional means within p-tau217 bins before computing the squared error between group means. A model that outputs the bin-level conditional mean (or a bin template) can achieve near-zero MSE with no per-subject accuracy, so Table II's very small values in the 2-4 bin do not demonstrate subject-specific fidelity. The claim that the model generates individualized tau PET requires per-subject metrics such as voxel-wise MAE/SSIM, per-region correlation or error per test subject, or Bland-Altman analysis. The reduced generated variance visible in Fig. 3 is consistent with regression toward group means and should be addressed quantitatively, not only by a qualitative discussion of missing demographic covariates.
  2. [Section IV-A / Tables I-II] No baseline comparison is provided against a trivial predictor, such as the bin-wise mean of real tau PET in the training set, or a simple conditional-mean regression from p-tau217 and MRI. Without such a baseline, the absolute MSE values in Tables I and II are uncalibrated, and the claim that adding MRI reduces error (Table I) could be driven by the inherent anatomical correspondence between MRI and PET rather than by a learned biological mapping. I recommend reporting per-subject error and comparing against at least one non-generative or template-based baseline with confidence intervals over test subjects.
  3. [Section IV-C / overall evaluation] The evaluation stratifies test subjects by the same p-tau217 value used as the model input, and the training objective explicitly fits the plasma-to-PET association. Consequently, observing different group-mean tau distributions across bins partly reflects the known input-output correlation that the model is trained to reproduce; it does not independently establish that the model captures disease progression beyond that correlation. A stronger test would assess per-subject agreement between generated and real tau after accounting for p-tau217, or evaluate on subjects with discordant plasma-PET relationships (e.g., by amyloid status or cognitive diagnosis). Otherwise, the 'captures disease progression' claim is circular with respect to the conditioning variable.
  4. [Abstract / Section V Discussion] The abstract and introduction describe the output as 'clinically meaningful' and as a non-invasive alternative for visualizing tau pathology, while the Discussion states that the generated images are not suitable for clinical diagnosis. This is not necessarily a contradiction, but the claims need to be reconciled. More importantly, the proposed downstream uses - data augmentation and in-silico progression simulation - require per-subject realism, not just group-level concordance. Since the current evaluation cannot distinguish a subject-specific generator from a group-template generator, the practical utility claims are not empirically supported and should be either softened or backed by task-based evaluation (e.g., whether augmented training data improve a downstream classifier).
minor comments (5)
  1. [Table II] The plasma bins are listed as 0-2, 2-4, 4-6, 6-8, and 10+, with the 8-10 interval missing. If no subjects fall in that range, this should be stated; if the final bin is intended to mean 'greater than 8', the label should be changed. Group sizes N_g per bin should also be reported.
  2. [Section IV-E / Figure 4] The surface maps appear to be averaged across subjects within each plasma bin. Please clarify whether they are averages of individual generated surfaces or a single generation per bin, and whether the real maps are computed in the identical manner. This matters for interpreting the visual concordance.
  3. [Equation (18)] The MSE is defined on group means, but the values in Tables I and II are not accompanied by confidence intervals or standard errors. Given the small test set and uneven bin sizes, reporting bootstrap or subject-level variability would improve interpretability.
  4. [Section IV-E] There is an apparent citation error: '[1]' appears at the end of the paragraph 'This establishes a solid foundation for future investigations...', where no citation seems intended. Please remove or correct.
  5. [Section III-A] The prompt template is given as 'Plasma is [p-tau217 value].' Please specify whether the numeric value is inserted in the same normalized units used for binning, and whether any prompt augmentation or variation was used during training. The choice of the CLIP text encoder and its tokenization of numerical values should also be described, as this can affect sensitivity to the conditioning variable.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the synthesis pipeline and evaluation are self-contained; group-mean MSE weakness is a validation limitation, not a circular step.

full rationale

The paper's claimed derivation—synthesizing tau PET from MRI anatomy and p-tau217 via a conditional latent diffusion model—is self-contained. The training objective (Eq. 14) is a standard denoising loss on latent PET, conditioned on the MRI latent and CLIP text embedding of p-tau217; evaluation (Eqs. 16–18) measures group-level region-wise MSE on held-out subjects stratified by plasma bins. Although this evaluation is vulnerable to regression-to-the-mean and does not establish per-subject fidelity (a limitation the authors acknowledge in the Discussion, noting lower generated variance and missing demographic/clinical inputs), it does not constitute circularity: the model is not fitted to the bin-level group means, and the observed progression pattern is an empirical outcome of the learned mapping, not an identity with the input. Self-citations ([15], [21], [28]) are background/feasibility references only and do not carry a load-bearing argument; there is no imported uniqueness theorem or ansatz-by-citation. The central claim therefore has independent content, and the quantitative weakness should be scored as a correctness/validation risk rather than circularity.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

The paper's central claim rests on the assumption that MRI anatomy plus one plasma scalar is sufficient to determine individual tau distribution (learned from 260 training pairs), and on evaluation metrics computed at the group-mean level over hand-chosen plasma bins. No new physical entities are introduced; the main risk is under-determination of the mapping and dependence on coarse, unvalidated metrics.

free parameters (3)
  • Plasma interval bins for evaluation = 0-2, 2-4, 4-6, 6-8, 10+
    Post-hoc thresholds chosen for Table II and Figs. 3-4; the 2-4 bin yields near-zero MSE, likely because group-mean averaging of a small bin cancels individual errors.
  • Image normalization and resampling = [0,1] intensity; 160x160x96 voxels
    SUVR values are scaled to [0,1] and volumes resampled; this hand-chosen preprocessing determines the scale and spatial resolution of all generated images.
  • Prompt template = "Plasma is [value]."
    Hand-chosen text representation of a scalar; CLIP embeddings of numbers are known to be poorly calibrated, so this format affects conditioning quality.
assumptions (5)
  • domain assumption Tau PET spatial distribution is determined by rigidly registered MRI anatomy plus the global p-tau217 scalar.
    The whole architecture (MRI latent concatenation, single CLIP prompt) presupposes that individual tau topography is a function of these two inputs; if individual variability (e.g., Braak stage variation, amyloid status) matters, the generated images are only population templates.
  • domain assumption A single shared autoencoder can encode both PET and MRI without modality-specific loss.
    Section II.A states PET and MRI share the same encoder/decoder 'to minimize parameters.' No evidence is given that this shared latent space preserves the information each modality needs.
  • domain assumption Rigid registration (ANTs) aligns MRI to PET sufficiently for voxel-level latent concatenation.
    Section III.A describes rigid registration only; residual misalignment or individual anatomical variation could corrupt the concatenated MRI latent.
  • domain assumption CLIP text embeddings of numeric p-tau217 values maintain ordinal and magnitude information.
    Section II.C encodes the prompt through CLIP; CLIP was trained on natural language and images, not on calibrated numerical quantities, so the embedding may not reflect the quantitative relationship between p-tau217 and tau burden.
  • domain assumption 260 training pairs are sufficient to learn the plasma-to-PET mapping across the disease spectrum.
    Section III.A uses 260/360 for training; the authors themselves flag small dataset size in Section V. The sparse 10+ plasma bin is particularly underdetermined.

how reviews work

0 comments
Cite this review

Pith. "Pith review of TauGenNet: Plasma-Driven Tau PET Image Synthesis via Text-Guided 3D Diffusion Models." pith.science (2026). https://pith.science/paper/6ZN3SKQS

@misc{pith2026250904269,
  author       = {Pith},
  title        = {Pith review of: TauGenNet: Plasma-Driven Tau PET Image Synthesis via Text-Guided 3D Diffusion Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/6ZN3SKQS}},
  note         = {Machine review of arXiv:2509.04269}
}
read the original abstract

Accurate quantification of tau pathology via tau positron emission tomography (PET) scan is crucial for diagnosing and monitoring Alzheimer's disease (AD). However, the high cost and limited availability of tau PET restrict its widespread use. In contrast, structural magnetic resonance imaging (MRI) and plasma-based biomarkers provide non-invasive and widely available complementary information related to brain anatomy and disease progression. In this work, we propose a text-guided 3D diffusion model for 3D tau PET image synthesis, leveraging multimodal conditions from both structural MRI and plasma measurement. Specifically, the textual prompt is from the plasma p-tau217 measurement, which is a key indicator of AD progression, while MRI provides anatomical structure constraints. The proposed framework is trained and evaluated using clinical AV1451 tau PET data from the Alzheimer's Disease Neuroimaging Initiative (ADNI) database. Experimental results demonstrate that our approach can generate realistic, clinically meaningful 3D tau PET across a range of disease stages. The proposed framework can help perform tau PET data augmentation under different settings, provide a non-invasive, cost-effective alternative for visualizing tau pathology, and support the simulation of disease progression under varying plasma biomarker levels and cognitive conditions.

Figures

Figures reproduced from arXiv: 2509.04269 by the authors.

Figure 1
Figure 1. Overview of the proposed 3D TauGenNet framework: (a) Pretraining a 3D autoencoder for tau PET and MRI to obtain latent representations. (b) [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. One example of generated tau PET images across varying p-tau 217 concentrations (0.65, 3.65, 6.65, and 10.65) shown at axial, sagittal, and coronal [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. The box plot of the average values of different brain regions for the generated and the real measured tau images at different p-tau 217 ranges. [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: The brain surface plots under different p-tau 217 ranges for (a) real tau PET images and (b) generated tau PET images. [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]

Discussion (0). Sign in to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. LNODE: latent dynamics reveal the shared spatiotemporal structure of amyloid-$\beta$ progression

    q-bio.QM 2026-04 unverdicted novelty 6.0 of 10

    LNODE is a regional neural ODE model with latent states that fits amyloid-β PET data at R² > 0.99, predicts future scans over multi-year intervals, and clusters subjects into progression subtypes.

Reference graph

Works this paper leans on

38 extracted references · 33 canonical work pages · cited by 1 Pith paper

  1. [1]

    Tau positron emission tomographic imaging in aging and early a lzheimer disease,

    K. A. Johnson, A. Schultz, R. A. Betensky, J. A. Becker, J. Sepulcre, D. Rentz, E. Mormino, J. Chhatwal, R. Amariglio, K. Papp, et al., “Tau positron emission tomographic imaging in aging and early a lzheimer disease,” Annals of neurology , vol. 79, no. 1, pp. 110–119, 2016

    work page 2016

  2. [2]

    Neuropathological stageing of alzheimer-related changes,

    H. Braak and E. Braak, “Neuropathological stageing of alzheimer-related changes,” Acta neuropathologica, vol. 82, no. 4, pp. 239–259, 1991

    work page 1991

  3. [3]

    Early clinical pet imaging results with the novel phf-tau radioligand [f-18]-t807,

    D. T. Chien, S. Bahri, A. K. Szardenings, J. C. Walsh, F. Mu, M.-Y . Su, W. R. Shankle, A. Elizarov, and H. C. Kolb, “Early clinical pet imaging results with the novel phf-tau radioligand [f-18]-t807,” Journal of Alzheimer’s Disease , vol. 34, no. 2, pp. 457–468, 2013

    work page 2013

  4. [4]

    Preclinical characterization of 18f-mk-6240, a promising pet tracer for in vivo quantification of human neurofibrillary tangles,

    E. D. Hostetler, A. M. Walji, Z. Zeng, P. Miller, I. Bennacef, C. Salinas, B. Connolly, L. Gantert, H. Haley, M. Holahan, et al. , “Preclinical characterization of 18f-mk-6240, a promising pet tracer for in vivo quantification of human neurofibrillary tangles,” Journal of nuclear medicine, vol. 57, no. 10, pp. 1599–1606, 2016

    work page 2016

  5. [5]

    Discovery and preclinical characterization of [18f] pi-2620, a next- generation tau pet tracer for the assessment of tau pathology in alzheimer’s disease and other tauopathies,

    H. Kroth, F. Oden, J. Molette, H. Schieferstein, F. Capotosti, A. Mueller, M. Berndt, H. Schmitt-Willich, V . Darmency, E. Gabellieri, et al. , “Discovery and preclinical characterization of [18f] pi-2620, a next- generation tau pet tracer for the assessment of tau pathology in alzheimer’s disease and other tauopathies,” European journal of nuclear medici...

    work page 2019

  6. [6]

    Regional profiles of the candidate tau pet ligand 18 f-av-1451 recapitulate key features of braak histopathological stages,

    A. J. Schwarz, P. Yu, B. B. Miller, S. Shcherbinin, J. Dickson, M. Navit- sky, A. D. Joshi, M. D. Devous Sr, and M. S. Mintun, “Regional profiles of the candidate tau pet ligand 18 f-av-1451 recapitulate key features of braak histopathological stages,” Brain, vol. 139, no. 5, pp. 1539–1550, 2016

    work page 2016

  7. [7]

    Relationships between flortaucipir pet tau binding and amyloid burden, clinical diagnosis, age and cognition,

    M. J. Pontecorvo, M. D. Devous Sr, M. Navitsky, M. Lu, S. Salloway, F. W. Schaerf, D. Jennings, A. K. Arora, A. McGeehan, N. C. Lim, et al., “Relationships between flortaucipir pet tau binding and amyloid burden, clinical diagnosis, age and cognition,” Brain, vol. 140, no. 3, pp. 748–763, 2017

    work page 2017

  8. [8]

    Tau pet patterns mirror clinical and neuroanatomical variability in alzheimer’s disease,

    R. Ossenkoppele, D. R. Schonhaut, M. Sch ¨oll, S. N. Lockhart, N. Ayakta, S. L. Baker, J. P. O’Neil, M. Janabi, A. Lazaris, A. Cantwell, et al., “Tau pet patterns mirror clinical and neuroanatomical variability in alzheimer’s disease,” Brain, vol. 139, no. 5, pp. 1551–1567, 2016

    work page 2016

Show all 38 references
  1. [9]

    The alzheimer’s disease neuroimaging initiative,

    S. G. Mueller, M. W. Weiner, L. J. Thal, R. C. Petersen, C. Jack, W. Ja- gust, J. Q. Trojanowski, A. W. Toga, and L. Beckett, “The alzheimer’s disease neuroimaging initiative,” Neuroimaging Clinics, vol. 15, no. 4, pp. 869–877, 2005

    2005

  2. [10]

    Oasis-3: longitudinal neuroimaging, clinical, and cognitive dataset for normal aging and alzheimer disease,

    P. J. LaMontagne, T. L. Benzinger, J. C. Morris, S. Keefe, R. Hornbeck, C. Xiong, E. Grant, J. Hassenstab, K. Moulder, A. G. Vlassenko, et al., “Oasis-3: longitudinal neuroimaging, clinical, and cognitive dataset for normal aging and alzheimer disease,” medrxiv, pp. 2019–12, 2019

    2019

  3. [11]

    Harvard aging brain study: dataset and accessibility,

    A. Dagley, M. LaPoint, W. Huijbers, T. Hedden, D. G. McLaren, J. P. Chatwal, K. V . Papp, R. E. Amariglio, D. Blacker, D. M. Rentz, et al., “Harvard aging brain study: dataset and accessibility,” Neuroimage, vol. 144, pp. 255–258, 2017

    2017

  4. [12]

    Multimodal subtypes identified in alzheimer’s disease neuroimaging initiative participants by missing-data-enabled subtype and stage inference,

    M. Estarellas, N. P. Oxtoby, J. M. Schott, D. C. Alexander, and A. L. Young, “Multimodal subtypes identified in alzheimer’s disease neuroimaging initiative participants by missing-data-enabled subtype and stage inference,” Brain Communications, vol. 6, no. 4, p. fcae219, 2024

    2024

  5. [13]

    Machine learning prediction of tau-pet in alzheimer’s disease using plasma, mri, and clinical data,

    L. Karlsson, J. V ogel, I. Arvidsson, K. ˚Astr¨om, O. Strandberg, J. Seidlitz, R. A. Bethlehem, E. Stomrud, R. Ossenkoppele, N. J. Ashton, et al. , “Machine learning prediction of tau-pet in alzheimer’s disease using plasma, mri, and clinical data,” Alzheimer’s & Dementia, vol...

    2025

  6. [14]

    Effect of data augmentation of f-18-florbetaben positron-emission tomography images by using deep learning convolutional neural network architecture for amyloid positive patients,

    H. J. Yoon, Y . J. Jeong, D.-Y . Kang, H. Kang, K. K. Yeo, J. E. Jeong, K. W. Park, G. E. Choi, and S.-W. Ha, “Effect of data augmentation of f-18-florbetaben positron-emission tomography images by using deep learning convolutional neural network architecture for amyloid posit...

    2019

  7. [15]

    Taupetgen: Text- conditional tau pet image synthesis based on latent diffusion models,

    S.-I. Jang, C. Lois, E. Thibault, J. A. Becker, Y . Dong, M. D. Normandin, J. C. Price, K. A. Johnson, G. E. Fakhri, and K. Gong, “Taupetgen: Text- conditional tau pet image synthesis based on latent diffusion models,” arXiv preprint arXiv:2306.11984 , 2023

    2023 arXiv

  8. [16]

    Image-and-label condition- ing latent diffusion model: Synthesizing a β-pet from mri for detecting amyloid status,

    Z. Ou, Y . Pan, F. Xie, Q. Guo, and D. Shen, “Image-and-label condition- ing latent diffusion model: Synthesizing a β-pet from mri for detecting amyloid status,” IEEE Journal of Biomedical and Health Informatics , 2024

    2024

  9. [17]

    Plasma-cyclegan: Plasma biomarker-guided mri to pet cross-modality translation using conditional cyclegan,

    Y . Chen, Y . Su, C. Dumitrascu, K. Chen, D. Weidman, R. J. Caselli, N. Ashton, E. M. Reiman, and Y . Wang, “Plasma-cyclegan: Plasma biomarker-guided mri to pet cross-modality translation using conditional cyclegan,” in 2025 IEEE 22nd International Symposium on Biomedical Imag...

    2025

  10. [18]

    Denoising diffusion probabilistic models,

    J. Ho, A. Jain, and P. Abbeel, “Denoising diffusion probabilistic models,” Advances in neural information processing systems , vol. 33, pp. 6840– 6851, 2020

    2020

  11. [19]

    Denoising diffusion implicit models,

    J. Song, C. Meng, and S. Ermon, “Denoising diffusion implicit models,” arXiv preprint arXiv:2010.02502 , 2020

    2010 arXiv

  12. [20]

    High- resolution image synthesis with latent diffusion models,

    R. Rombach, A. Blattmann, D. Lorenz, P. Esser, and B. Ommer, “High- resolution image synthesis with latent diffusion models,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp. 10684–10695, 2022

    2022

  13. [21]

    Pet image denoising based on denoising diffusion probabilistic model,

    K. Gong, K. Johnson, G. El Fakhri, Q. Li, and T. Pan, “Pet image denoising based on denoising diffusion probabilistic model,” European Journal of Nuclear Medicine and Molecular Imaging , vol. 51, no. 2, pp. 358–368, 2024

    2024

  14. [22]

    Pet-diffusion: Unsupervised pet enhancement based on the latent diffusion model,

    C. Jiang, Y . Pan, M. Liu, L. Ma, X. Zhang, J. Liu, X. Xiong, and D. Shen, “Pet-diffusion: Unsupervised pet enhancement based on the latent diffusion model,” in International Conference on Medical Image Computing and Computer-Assisted Intervention , pp. 3–12, Springer, 2023

    2023

  15. [23]

    Dose-aware diffusion model for 3d low-dose pet: multi-institutional validation with reader study and real low-dose data,

    H. Xie, W. Gan, B. Zhou, M.-K. Chen, M. Kulon, A. Boustani, B. A. Spencer, R. Bayerlein, X. Chen, Q. Liu, et al. , “Dose-aware diffusion model for 3d low-dose pet: multi-institutional validation with reader study and real low-dose data,” arXiv preprint arXiv:2405.12996 , 2024

    2024 arXiv

  16. [24]

    Bidirectional condition diffusion probabilistic models for pet image denoising,

    C. Shen, C. Tie, Z. Yang, N. Zhang, and Y . Zhang, “Bidirectional condition diffusion probabilistic models for pet image denoising,” IEEE Transactions on Radiation and Plasma Medical Sciences , vol. 8, no. 4, pp. 402–415, 2024. JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUG...

    2024

  17. [25]

    Score-based generative models for pet image reconstruction,

    I. R. Singh, A. Denker, R. Barbano, ˇZ. Kereta, B. Jin, K. Thielemans, P. Maass, and S. Arridge, “Score-based generative models for pet image reconstruction,” arXiv preprint arXiv:2308.14190 , 2023

    2023 arXiv

  18. [26]

    Super- vised diffusion-model-based pet image reconstruction,

    G. Webber, A. Hammers, A. P. King, and A. J. Reader, “Super- vised diffusion-model-based pet image reconstruction,” arXiv preprint arXiv:2506.24034, 2025

    2025

  19. [27]

    Joint reconstruction of activity and attenuation in pet by diffusion posterior sampling in wavelet coefficient space,

    C. Phung-Ngoc, A. Bousse, A. De Paepe, H.-P. Dang, O. Saut, and D. Visvikis, “Joint reconstruction of activity and attenuation in pet by diffusion posterior sampling in wavelet coefficient space,” arXiv preprint arXiv:2505.18782, 2025

    2025 arXiv

  20. [28]

    Pet image reconstruction using deep diffusion image prior,

    F. Hashimoto and K. Gong, “Pet image reconstruction using deep diffusion image prior,” arXiv preprint arXiv:2507.15078 , 2025

    2025

  21. [29]

    Measurement guidance in diffusion models: Insight from medical image synthesis,

    Y . Luo, Q. Yang, Y . Fan, H. Qi, and M. Xia, “Measurement guidance in diffusion models: Insight from medical image synthesis,” IEEE Transactions on Pattern Analysis and Machine Intelligence , vol. 46, no. 12, pp. 7983–7997, 2024

    2024

  22. [30]

    Medsyn: text- guided anatomy-aware synthesis of high-fidelity 3-d ct images,

    Y . Xu, L. Sun, W. Peng, S. Jia, K. Morrison, A. Perer, A. Zandifar, S. Visweswaran, M. Eslami, and K. Batmanghelich, “Medsyn: text- guided anatomy-aware synthesis of high-fidelity 3-d ct images,” IEEE Transactions on Medical Imaging, vol. 43, no. 10, pp. 3648–3660, 2024

    2024

  23. [31]

    Rl4med-ddpo: Reinforcement learning for controlled guidance towards diverse medical image generation using vision-language foundation models,

    P. Saremi, A. Kumar, M. Mohamed, Z. TehraniNasab, and T. Arbel, “Rl4med-ddpo: Reinforcement learning for controlled guidance towards diverse medical image generation using vision-language foundation models,” arXiv preprint arXiv:2503.15784 , 2025

    2025 arXiv

  24. [32]

    Dis- criminative accuracy of plasma phospho-tau217 for alzheimer disease vs other neurodegenerative disorders,

    S. Palmqvist, S. Janelidze, Y . T. Quiroz, H. Zetterberg, F. Lopera, E. Stomrud, Y . Su, Y . Chen, G. E. Serrano, A. Leuzy, et al. , “Dis- criminative accuracy of plasma phospho-tau217 for alzheimer disease vs other neurodegenerative disorders,” Jama, vol. 324, no. 8, pp. 772–...

    2020

  25. [33]

    Plasma phosphorylated tau 217 and phosphorylated tau 181 as biomarkers in alzheimer’s disease and frontotemporal lobar degeneration: a retrospective diagnostic performance study,

    E. H. Thijssen, R. La Joie, A. Strom, C. Fonseca, L. Iaccarino, A. Wolf, S. Spina, I. E. Allen, Y . Cobigo, H. Heuer,et al., “Plasma phosphorylated tau 217 and phosphorylated tau 181 as biomarkers in alzheimer’s disease and frontotemporal lobar degeneration: a retrospective di...

    2021

  26. [34]

    Performance of plasma phosphorylated tau 181 and 217 in the community,

    M. M. Mielke, J. L. Dage, R. D. Frank, A. Algeciras-Schimnich, D. S. Knopman, V . J. Lowe, G. Bu, P. Vemuri, J. Graff-Radford, C. R. Jack Jr, et al. , “Performance of plasma phosphorylated tau 181 and 217 in the community,” Nature medicine, vol. 28, no. 7, pp. 1398–1405, 2022

    2022

  27. [35]

    Plasma phospho-tau217 for alzheimer’s disease diagnosis in primary and secondary care using a fully automated platform,

    S. Palmqvist, N. Warmenhoven, F. Anastasi, A. Pilotto, S. Janelidze, P. Tideman, E. Stomrud, N. Mattsson-Carlgren, R. Smith, R. Ossenkop- pele, et al. , “Plasma phospho-tau217 for alzheimer’s disease diagnosis in primary and secondary care using a fully automated platform,” Na...

    2025

  28. [36]

    Alzheimer’s disease neuroimaging initiative,

    S. G. Mueller, M. W. Weiner, L. J. Thal, R. C. Petersen, C. Jack, W. Jagust, J. Q. Trojanowski, A. W. Toga, and L. Beckett, “Alzheimer’s disease neuroimaging initiative,” in Advances in Alzheimer’s and Parkin- son’s Disease: Insights, Progress, and Perspectives , pp. 183–189, ...

    2008

  29. [37]

    A reproducible evaluation of ants similarity metric performance in brain image registration,

    B. B. Avants, N. J. Tustison, G. Song, P. A. Cook, A. Klein, and J. C. Gee, “A reproducible evaluation of ants similarity metric performance in brain image registration,” Neuroimage, vol. 54, no. 3, pp. 2033–2044, 2011

  30. [38]

    An automated labeling system for subdividing the human cerebral cortex on mri scans into gyral based regions of interest,

    R. S. Desikan, F. S ´egonne, B. Fischl, B. T. Quinn, B. C. Dickerson, D. Blacker, R. L. Buckner, A. M. Dale, R. P. Maguire, B. T. Hyman, et al., “An automated labeling system for subdividing the human cerebral cortex on mri scans into gyral based regions of interest,” Neuroima...

    2006

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.