Pith. sign in

REVIEW 3 major objections 4 minor 43 references

VoxStruct3D: Structure-Leading Flow Matching for Voxel-Space 3D MRI Synthesis

T0 review · 3 major / 4 minor · reviewed 2026-08-06 · deepseek-v4-flash

Pith's one-line read VoxStruct3D claims direct voxel-space flow matching with overlapping volumetric decoding and a structure-first guidance stream outperforms latent and wavelet 3D MRI generation on distribution alignment, diversity, and perceptual quality.

desk verdict VoxStruct3D has a genuinely interesting voxel-space architecture and strong generation results, but the downstream-classification claim is undermined by a likely train/test leak. read the letter →

arxiv 2608.04557 v1 pith:RC65XKHO submitted 2026-08-05 cs.CV

classification cs.CV
keywords 3DMRIsynthesisflowmatchingvoxel-spacegenerationdiffusiontransformeranatomicalpriorbrainstructureguidancemedicalimage
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

VoxStruct3D sets out to show that high-fidelity 3D brain MRI can be generated directly in the voxel domain, without the image-autoencoder bottleneck that limits latent diffusion. The paper argues that two obstacles—seams between independently decoded voxel blocks, and the loss of compact anatomical guidance when no latent exists—can both be fixed inside one flow-matching model. Its Volumetric Voxel Generator lets neighboring tokens reconstruct shared voxel regions, while its Structure-First, Image-Follows scheme runs a compressed anatomical state ahead of the image stream and lets the image read it one-way. On pathological and healthy T1 brain MRI, the proposed configuration reports the best feature-distribution alignment, sample diversity, and perceptual scores among the tested 3D generators, and its synthetic volumes keep enough condition information that classifiers trained on them remain close to classifiers trained on real data. If this holds, voxel-space generation with an explicit anatomical prior becomes a practical alternative to latent and wavelet-based 3D synthesis.

What carries the argument

The load-bearing object is the paired double-stream trajectory with structure-leading clocks. The image stream interpolates $x_t = t x_1 + (1-t)\epsilon_x$ and the structure stream interpolates $s_t = t_s s_1 + (1-t_s)\epsilon_s$, with $t_s = \min(t+\delta, 1)$ and $\delta = 0.3$; the model predicts the clean endpoints $x_1$ and $s_1$ rather than the noise. This is carried by two mechanisms: the Volumetric Voxel Generator, whose overlapping unpatchify couples neighboring tokens through transposed convolutions with kernel larger than stride; and SFIF, whose StructVAE bottleneck turns frozen 3DINO features into 64-dimensional per-token codes on a $7\times7\times7$ grid. Patch-Aligned RoPE aligns the unequal image and structure grids on least-common-multiple coordinates, and the asymmetric attention mask enforces one-way structure-to-image information flow while both streams share context tokens.

What would settle it

Train or fine-tune VoxStruct3D on a tumor cohort where lesions are small relative to the $7\times7\times7$ structure-token grid, and compare lesion-region FID or lesion segmentation overlap between the structure-leading schedule ($\delta = 0.3$) and a synchronized baseline ($\delta = 0$); a drop in lesion-region quality would falsify the claim that the compressed structure prior preserves lesion-defining anatomy.

Watch

Extended reading notes

Core claim

The paper claims that direct voxel-space generation can match or exceed latent-space 3D generation if the two failure modes of voxel-space diffusion transformers are addressed explicitly. Its central discovery is that both failure modes have local fixes that compose. Grid-aligned block artifacts disappear when independent token-to-block reconstruction is replaced by overlapping volumetric upsampling: transposed convolutions whose kernels are larger than their strides let neighboring tokens contribute to shared voxel regions, and a time-modulated residual path with U-Net-style skip connections refines the result. Loss of global anatomy is repaired by the Structure-First, Image-Follows strategy, in which a frozen 3DINO encoder's patch features are compressed by a StructVAE into compact structure tokens, and these tokens run ahead on a clock $t_s = \min(t+\delta, 1)$ with $\delta = 0.3$, so the image stream always reads a structure state closer to clean than its own state. Patch-Aligned RoPE places the $10\times10\times7$ image grid and the $7\times7\times7$ structure grid on a common lattice, and an asymmetric attention mask lets image tokens attend to structure keys while blocking the reverse path. With these components, VoxStruct3D reports the best feature-distribution alignment, inter-sample diversity, and perceptual scores on both pathological and healthy T1 brain MRI, and downstream classifiers trained on its synthetic volumes retain condition-discriminative anatomy.

Load-bearing premise

The whole approach depends on the assumption that the compact anatomical summary squeezed out of a pretrained 3D brain encoder still contains the structures that matter for the images being generated; if the compression throws away lesion-defining detail, the structure-first guidance would lead the image stream astray.

Editorial extensions

If this is right

  • Direct voxel-space flow matching becomes a practical regime for 3D MRI: the paper reports better FID and MS-SSIM than latent and wavelet baselines using only 100 network evaluations, versus 1,000 for several diffusion baselines.
  • Synthetic volumes can substitute for real training data to a meaningful degree: classifiers trained only on generated volumes stay within roughly two accuracy points of classifiers trained on real volumes, and combining real and synthetic data improves accuracy and balanced accuracy.
  • Overlapping volumetric decoding removes the need for an image autoencoder to smooth patch boundaries, showing that the fine-detail bottleneck of latent diffusion is avoidable in 3D medical imaging.
  • A frozen medical encoder can serve as an internal anatomical guide at inference without a separate sampling pass, provided its features are compressed and clocked ahead of the image stream.
  • A moderate structure lead is better than synchronized clocks: in the sensitivity sweep all nonzero leads in $\{0.15, 0.30, 0.45\}$ improve both FID and MS-SSIM over $\delta = 0$, with $\delta = 0.30$ best.

Reading between the lines

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

  • The paper does not test rare or small pathology, but its logic suggests the structure-first prior should be checked on lesion-focused cohorts; small lesions may be lost in the StructVAE 64-dimensional bottleneck, and a lesion-region metric would settle whether the healthy-cohort gains transfer.
  • The structure-lead schedule could be made adaptive: since moderate leads all beat synchronized clocks, a per-sample or per-region $\delta$, or a lead that shrinks as generation progresses, is a natural extension.
  • The same design pattern—a compressed semantic stream running ahead of a high-detail stream with one-way attention—should transfer to other 3D modalities such as CT or PET, or to conditioning on segmentations, because nothing in the mechanism is MRI-specific.
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 paper proposes VoxStruct3D, a voxel-space flow-matching framework for 3D T1-weighted brain MRI synthesis. The method combines a Volumetric Voxel Generator (VVG), built on a shared dual-stream DiT with overlapping volumetric decoding, time-modulated residual refinement, and skip fusion, with a Structure-First, Image-Follows (SFIF) guidance stream that distills compact structure tokens from a frozen 3DINO encoder via StructVAE and steers them ahead of the image stream using structure-leading clocks, Patch-Aligned RoPE, and asymmetric attention. Experiments on a pathological cohort (BraTS 2021) and a healthy cohort compare VoxStruct3D with five open-source 3D generators across distributional, diversity, and perceptual metrics, and include ablations, efficiency measurements, and downstream classification on synthetic data. The central claim is that VoxStruct3D achieves the strongest overall performance and produces anatomically coherent volumes.

Significance. If the claims are validated, the paper would demonstrate a meaningful advance: direct voxel-space generation with an explicit anatomical prior can outperform latent- and wavelet-based 3D generators on distribution matching, diversity, and perceptual quality, while also providing synthetic volumes useful for downstream classifier training. The experimental breadth is a strength: two datasets, five baselines retrained from official implementations, multiple evaluation metrics including the independent MedicalNet FID, a controlled sensitivity sweep over the structure lead, and a downstream classification study. The detailed appendix on preprocessing and metric implementation improves reproducibility. The independent MedicalNet FID results do support the main ranking, but several load-bearing issues, described below, require attention before the claims can be accepted.

major comments (3)
  1. [Appendix, 'Data partition' and Table 7; §'Downstream classification'] The downstream classification results in Table 5 may be contaminated by data leakage. The paper does not state that VoxStruct3D's Stage II training was restricted to the classifier training split. The datasets section says the pathological cohort comprises all 1,251 BraTS 2021 volumes and the healthy cohort all 958 volumes, and the implementation details say the generator is trained for 500 epochs on these cohorts. The classifier test set, however, consists of 188 pathological and 144 healthy subjects drawn from exactly these cohorts. If the generator observed those test subjects during training, the synthetic volumes used as classifier training data can contain near-duplicates of test subjects, inflating the synthetic-only and combined results. The authors must either explicitly confirm that Stage II training used only the classifier training split (e.g., the 876 + 670 volumes from Table 7) or retrain the generator on that split and regenerate Table 5.
  2. [Tables 3, 4, and 8; §'SFIF ablation'] The ablations and the structure-lead sweep report FID computed in the 3DINO feature space, but the SFIF structure tokens are distilled from the same frozen 3DINO encoder and the generator is deliberately trained to follow those structure tokens. Improvements in 3DINO FID can therefore partly reflect the model's alignment to its own teacher rather than general image quality. The main comparison in Table 1 includes the independent MedicalNet FID, but the component analyses in Tables 3, 4, and 8 rely solely on 3DINO FID. The authors should add an independent evaluation metric (for example, MedicalNet FID or a non-learned metric such as MS-SSIM) to the ablation tables to support the individual VVG and SFIF contributions.
  3. [Table 1] All quantitative results in Table 1 are reported as point estimates without confidence intervals, repeated seeds, or significance tests. Several differences between the proposed method and the closest baselines are small, for example the pathological MedicalNet FID values (0.3365 vs 0.3621 for WDM) and the healthy MedicalNet FID values (1.572 vs 1.746 for WDM). Without an estimate of variance, the claim of 'strongest overall performance' is not statistically supported. The authors should report results across multiple seeds with means and standard deviations, or provide a statistical test over independent runs, at least for the headline FID metrics.
minor comments (4)
  1. [§'Implementation details'] The sentence 'The weights of the 3D Sobel loss are set to λ_sobel = 0.5, respectively' contains a stray 'respectively'; the value could simply be listed without the adverb.
  2. [Figure 4 caption] The caption 'W/OVVGW/OSFIFOurs' is missing spaces and is hard to parse; it should be written as 'W/o VVG, w/o SFIF, Ours' or with a similar unambiguous format.
  3. [Appendix, '3DINO feature extraction'] The distinction between the class-token descriptor used for FID/MMD and the patch tokens used as the structural teacher is important and clearly stated; it would help to reference this distinction in the main text when the 3DINO FID is first introduced, to preempt concerns about circularity.
  4. [§'Computational efficiency'] In Table 2, the column header 'NFE s' is awkward; consider 'NFE' as the header with the count below, and keep 'Inference (s/vol.)' as a separate column.

Circularity Check

0 steps flagged · score 0.0 of 10

No by-construction circularity: 3DINO teacher/evaluator overlap and downstream split ambiguity are validity concerns, not circular reductions.

full rationale

The claimed derivation chain is self-contained. VoxStruct3D's generator (Eqs. 1, 2, 4, 5, 7) is trained to predict clean voxel endpoints and structure endpoints from noised inputs; the image branch is supervised in voxel space and the structure branch in token space. No evaluation metric is an algebraic transform of a fitted parameter: the 3DINO FID/MMD are computed on the class token, while the SFIF teacher supervises patch-token-derived codes (Appendix, '3DINO feature extraction'), so the overlap is a correlation/validity concern, not a by-construction reduction. Independent MedicalNet FID, MS-SSIM, NIQE, MUSIQ, Tenengrad, and the component ablations provide external corroboration that does not depend on the 3DINO structure stream. The most serious concern in the manuscript is the downstream-split ambiguity: Table 7 defines the 70/15/15 split only for the classifiers, and no sentence restricts Stage-II generator training to the classifier training split, so synthetic-only classifier accuracy in Table 5 may be inflated by the generator having seen the test subjects. That is data leakage / missing support, not circularity under the rubric. No load-bearing step reduces to its own input.

Assumptions & free parameters 6 free parameters · 5 assumptions · 0 invented entities

The central experimental claim rests on several hand-set hyperparameters (delta, p_warm, lambda_sobel, CFG scale, Stage-I loss weights) and on the domain assumption that 3DINO-derived structure tokens are a sufficient anatomical prior. The paper provides ablations for most architectural components, but not for the loss weights or CFG scale, and the structure-lead sweep only tunes delta.

free parameters (6)
  • Structure lead delta = 0.30
    Selected from a sweep on BraTS 2021 (delta in {0, 0.15, 0.30, 0.45, 0.60}) to minimize 3DINO FID and MS-SSIM; used as default for both datasets.
  • Structure-only warm-up probability p_warm = 0.15
    Hand-set hyperparameter controlling the fraction of structure-only samples in Eq. 4; no sensitivity study reported.
  • Sobel loss weight lambda_sobel = 0.5
    Hand-set weight in Eq. 7 for the time-weighted 3D Sobel gradient loss; no sensitivity study reported.
  • Sobel time-weight parameters = w(t_x) = [((t_x - 0.3) / 0.7)^1.5]_+, clipped to [0, 1]
    Hand-chosen offset 0.3 and exponent 1.5 in the Stage-II loss; no sensitivity study reported.
  • Classifier-free guidance scale = 4.0
    Hand-set inference scale with label dropout 0.1; no sweep reported.
  • Stage-I loss weights = lambda_rec=1.0, lambda_cos=1.0, lambda_KL=1e-7, lambda_stat mean=0.01, std=0.01
    Hand-set weights in Eq. 6 for StructVAE training; not swept.
assumptions (5)
  • standard math Conditional flow-matching interpolation x_t = t x_1 + (1 - t) epsilon with independent Gaussian noise yields a valid generative objective (Eq. 1).
    Standard flow-matching framework; the paper assumes the velocity and clean-data prediction objective is well-posed.
  • domain assumption Frozen 3DINO ViT-L/16 patch features on a 7x7x7 grid provide a valid and sufficient anatomical structure prior for T1 brain MRI, and StructVAE compression to 64 channels retains the dominant anatomy.
    The whole SFIF design depends on this; the paper validates it indirectly via ablations and downstream classification, not by direct anatomical evaluation.
  • domain assumption Binary pathological and healthy class labels are sufficient conditioning to generate both cohorts, and pooling IXI, NIMH, and NFBS as healthy after intensity preprocessing does not introduce harmful domain shift.
    The generator uses one conditioning variable y; no per-site or per-disease subtype conditioning is explored.
  • ad hoc to paper The structure-leading clock schedule (Eq. 4) with delta = 0.3 and warm-up probability 0.15 improves image generation without introducing misalignment between the unequal token grids.
    This schedule is a design innovation validated empirically; there is no theoretical guarantee, and the delta choice is tuned on BraTS.
  • domain assumption Preprocessing volumes by zero-padding or cropping to 240x240x168 without resizing preserves anatomical fidelity and enables fair cross-method comparison.
    All methods use the same pipeline, but zero-padding can introduce empty regions; the paper assumes this does not bias the comparisons.

how reviews work

0 comments
Cite this review

Pith. "Pith review of VoxStruct3D: Structure-Leading Flow Matching for Voxel-Space 3D MRI Synthesis." pith.science (2026). https://pith.science/paper/RC65XKHO

@misc{pith2026260804557,
  author       = {Pith},
  title        = {Pith review of: VoxStruct3D: Structure-Leading Flow Matching for Voxel-Space 3D MRI Synthesis},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/RC65XKHO}},
  note         = {Machine review of arXiv:2608.04557}
}
read the original abstract

High-fidelity 3D MRI synthesis requires both globally coherent anatomy and fine-grained voxel-level detail. Although latent diffusion makes volumetric generation tractable, its image autoencoder introduces a reconstruction bottleneck that can limit the fine detail recoverable in the final volume. We present VoxStruct3D, a voxel-space flow-matching framework that directly models full-resolution MRI volumes using a clean-data prediction objective. Its Volumetric Voxel Generator (VVG) combines factorized 3D patch embedding with overlapping upsampling, time-modulated residual refinement, and skip fusion, enabling neighboring tokens to jointly reconstruct shared voxel regions and suppress patch-boundary artifacts. To complement direct voxel-space modeling with an explicit anatomical prior, we further introduce a Structure-First, Image-Follows (SFIF) strategy. A frozen pretrained 3D medical encoder and a StructVAE extract compact structure tokens that preserve dominant anatomy, while a structure-leading schedule keeps their trajectory ahead of the image trajectory. Patch-Aligned RoPE spatially aligns the unequal token grids, and asymmetric attention enforces one-way guidance from structure to image. Experiments on pathological and healthy T1-weighted brain MRI datasets show that VoxStruct3D achieves the strongest overall performance across feature-distribution alignment, sample diversity, and perceptual quality, producing anatomically coherent and visually realistic volumes.

Figures

Figures reproduced from arXiv: 2608.04557 by the authors.

Figure 1
Figure 1. Given a clean volume x1 ∈ R C×D×H×W and its class label y, where D, H, and W denote the spatial depth, height, and width, respectively, and C denotes the number of channels, we inject voxel-level noise at image time tx to obtain xt. The noised volume is then patchified into image tokens zt ∈ R Nimg×d , where Nimg is the number of image tokens and d is the token dimension. In parallel, the frozen 3DINO encoder and St… view at source ↗
Figure 1
Figure 1. Overview of Stage II. Given a clean volume x1 and class label y, independent noise is injected into the image and structure streams at times tx and ts, respectively. The noised volume is patchified into image tokens zt, while the frozen 3DINO encoder and StructVAE produce noised structure tokens st. Both token streams are augmented with Patch-Aligned RoPE and jointly processed by a shared dual-stream DiT under an as… view at source ↗
Figure 2
Figure 2. PA-RoPE and asymmetric attention mask. Left: image and structure tokens are mapped to a common spatial lattice. Right: image tokens can read structure tokens while the reverse path is blocked. Here, g q a ∈ {0, . . . , Gq a − 1} is the native token index. The common coordinates ge q a are then used to compute the rotary vectors for both streams. Each attention head is split into depth, height, and width groups, with… view at source ↗
Figures from the paper (4 more)
Figure 3
Figure 3. Figure 3: Qualitative comparison on pathological brain MRI. Each column shows the central axial, coronal, and sagittal slices of one volume. Red and green regions are enlarged for detailed comparison. and spurious high-frequency responses rather than sharper details. 3D MedDiffu…
Figure 4
Figure 4. Figure 4 [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: Additional pathological T1 MRI volumes produced by VoxStruct3D. [PITH_FULL_IMAGE:figures/full_fig_p015_5.png]
Figure 6
Figure 6. Figure 6: Additional healthy T1 MRI volumes produced by VoxStruct3D. [PITH_FULL_IMAGE:figures/full_fig_p016_6.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

43 extracted references · 21 canonical work pages

  1. [1]

    Transactions on Machine Learning Research Journal , year=

    Dinov2: Learning robust visual features without supervision , author=. Transactions on Machine Learning Research Journal , year=

  2. [2]

    Scientific reports , volume=

    Denoising diffusion probabilistic models for 3D medical image generation , author=. Scientific reports , volume=. 2023 , publisher=

  3. [3]

    IEEE journal of biomedical and health informatics , volume=

    Hierarchical amortized GAN for 3D high resolution medical image synthesis , author=. IEEE journal of biomedical and health informatics , volume=. 2022 , publisher=

  4. [4]

    IEEE Transactions on Medical Imaging , year=

    3D MedDiffusion: A 3D medical latent diffusion model for controllable and high-quality medical image generation , author=. IEEE Transactions on Medical Imaging , year=

  5. [5]

    2025 IEEE/CVF Winter Conference on Applications of Computer Vision , pages=

    Maisi: Medical ai for synthetic imaging , author=. 2025 IEEE/CVF Winter Conference on Applications of Computer Vision , pages=. 2025 , organization=

  6. [6]

    International Conference on Medical Image Computing and Computer-Assisted Intervention , pages=

    Flow matching for medical image synthesis: Bridging the gap between speed and quality , author=. International Conference on Medical Image Computing and Computer-Assisted Intervention , pages=. 2025 , organization=

  7. [7]

    arXiv preprint arXiv:2211.07804 , year=

    Diffusion models for medical image analysis: A comprehensive survey , author=. arXiv preprint arXiv:2211.07804 , year=

  8. [8]

    MICCAI workshop on deep generative models , pages=

    Brain imaging generation with latent diffusion models , author=. MICCAI workshop on deep generative models , pages=. 2022 , organization=

Show all 43 references
  1. [9]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    High-resolution image synthesis with latent diffusion models , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  2. [10]

    The eleventh international conference on learning representations , year=

    Flow matching for generative modeling , author=. The eleventh international conference on learning representations , year=

  3. [11]

    Forty-first international conference on machine learning , year=

    Scaling rectified flow transformers for high-resolution image synthesis , author=. Forty-first international conference on machine learning , year=

  4. [12]

    arXiv preprint arXiv:2504.07963 , year=

    Pixelflow: Pixel-space generative models with flow , author=. arXiv preprint arXiv:2504.07963 , year=

  5. [13]

    International Conference on Medical Image Computing and Computer-Assisted Intervention , pages=

    Make-a-volume: Leveraging latent diffusion models for cross-modality 3d brain mri synthesis , author=. International Conference on Medical Image Computing and Computer-Assisted Intervention , pages=. 2023 , organization=

  6. [14]

    arXiv preprint arXiv:2410.06940 , year=

    Representation alignment for generation: Training diffusion transformers is easier than you think , author=. arXiv preprint arXiv:2410.06940 , year=

  7. [15]

    Advances in Neural Information Processing Systems , volume=

    U-repa: Aligning diffusion u-nets to vits , author=. Advances in Neural Information Processing Systems , volume=

  8. [16]

    IEEE Journal of Biomedical and Health Informatics , volume=

    Conditional diffusion models for semantic 3D brain MRI synthesis , author=. IEEE Journal of Biomedical and Health Informatics , volume=. 2024 , publisher=

  9. [17]

    arXiv preprint arXiv:2511.13720 , year=

    Back to basics: Let denoising generative models denoise , author=. arXiv preprint arXiv:2511.13720 , year=

  10. [18]

    International Conference on Learning Representations , volume=

    Efficient streaming language models with attention sinks , author=. International Conference on Learning Representations , volume=

  11. [19]

    arXiv preprint arXiv:2511.20645 , year=

    Pixeldit: Pixel diffusion transformers for image generation , author=. arXiv preprint arXiv:2511.20645 , year=

  12. [20]

    arXiv preprint arXiv:2605.23902 , year=

    PiD: Fast and High-Resolution Latent Decoding with Pixel Diffusion , author=. arXiv preprint arXiv:2605.23902 , year=

  13. [21]

    npj Digital Medicine , volume=

    A generalizable 3D framework and model for self-supervised learning in medical imaging , author=. npj Digital Medicine , volume=. 2025 , publisher=

  14. [22]

    MICCAI workshop on deep generative models , pages=

    Wdm: 3d wavelet diffusion models for high-resolution medical image synthesis , author=. MICCAI workshop on deep generative models , pages=. 2024 , organization=

  15. [23]

    Advances in Neural Information Processing Systems , volume=

    Scaling the codebook size of vq-gan to 100,000 with a utilization rate of 99\ author=. Advances in Neural Information Processing Systems , volume=

  16. [24]

    arXiv preprint arXiv:1904.00625 , year=

    Med3d: Transfer learning for 3d medical image analysis , author=. arXiv preprint arXiv:1904.00625 , year=

  17. [25]

    arXiv preprint arXiv:2107.02314 , year=

    The rsna-asnr-miccai brats 2021 benchmark on brain tumor segmentation and radiogenomic classification , author=. arXiv preprint arXiv:2107.02314 , year=

  18. [26]

    arXiv preprint arXiv:2506.14432 , year=

    A large-scale heterogeneous 3D magnetic resonance brain imaging dataset for self-supervised learning , author=. arXiv preprint arXiv:2506.14432 , year=

  19. [27]

    Gigascience , volume=

    The preprocessed connectomes project repository of manually corrected skull-stripped T1-weighted anatomical MRI data , author=. Gigascience , volume=. 2016 , publisher=

  20. [28]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Maisi-v2: Accelerated 3d high-resolution medical image synthesis with rectified flow and region-specific contrastive loss , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

  21. [29]

    Medical physics , volume=

    SynthRAD2023 Grand Challenge dataset: Generating synthetic CT for radiotherapy , author=. Medical physics , volume=. 2023 , publisher=

  22. [30]

    Advances in neural information processing systems , volume=

    Gans trained by a two time-scale update rule converge to a local nash equilibrium , author=. Advances in neural information processing systems , volume=

  23. [31]

    The journal of machine learning research , volume=

    A kernel two-sample test , author=. The journal of machine learning research , volume=. 2012 , publisher=

  24. [32]

    The thrity-seventh asilomar conference on signals, systems & computers, 2003 , volume=

    Multiscale structural similarity for image quality assessment , author=. The thrity-seventh asilomar conference on signals, systems & computers, 2003 , volume=. 2003 , organization=

  25. [33]

    Proceedings of the IEEE/CVF international conference on computer vision , pages=

    Musiq: Multi-scale image quality transformer , author=. Proceedings of the IEEE/CVF international conference on computer vision , pages=

  26. [34]

    completely blind

    Making a “completely blind” image quality analyzer , author=. IEEE Signal processing letters , volume=. 2012 , publisher=

  27. [35]

    IEEE Transactions on Image Processing , volume =

    No-Reference Image Quality Assessment in the Spatial Domain , author =. IEEE Transactions on Image Processing , volume =. 2012 , doi =

  28. [36]

    Medical Image Analysis , year=

    TransMorph: Transformer for unsupervised medical image registration , author=. Medical Image Analysis , year=

  29. [37]

    Scientific Data , volume=

    The NIMH intramural healthy volunteer dataset: A comprehensive MEG, MRI, and behavioral resource , author=. Scientific Data , volume=. 2022 , publisher=

  30. [38]

    Medical image analysis , volume=

    Convolutional neural networks for classification of Alzheimer's disease: Overview and reproducible evaluation , author=. Medical image analysis , volume=. 2020 , publisher=

  31. [39]

    2021 IEEE International Conference on Image Processing , pages=

    Enhancing Alzheimer’s Disease Diagnosis via Hierarchical 3D-FCN with Multi-Modal Features , author=. 2021 IEEE International Conference on Image Processing , pages=. 2021 , organization=

  32. [40]

    arXiv preprint arXiv:2601.05212 , year=

    FlowLet: Conditional 3D Brain MRI Synthesis using Wavelet Flow Matching , author=. arXiv preprint arXiv:2601.05212 , year=

  33. [41]

    2026 IEEE 23rd International Symposium on Biomedical Imaging , pages=

    Super-Resolution MRI Using Latent Fusion and Flow Matching , author=. 2026 IEEE 23rd International Symposium on Biomedical Imaging , pages=. 2026 , organization=

  34. [42]

    Medical Imaging with Deep Learning , year=

    WFM: 3D Wavelet Flow Matching for Ultrafast Multi-Modal MRI Synthesis , author=. Medical Imaging with Deep Learning , year=

  35. [43]

    arXiv preprint arXiv:2508.10104 , year=

    Dinov3 , author=. arXiv preprint arXiv:2508.10104 , year=

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.