Pith. sign in

REVIEW 3 major objections 5 minor 36 references

This paper claims that a Brownian bridge diffusion model in motion space can refine registration-derived cardiac motion from routine cine CMR into DENSE-quality motion, improving myocardial strain estimates without specialized acquisition.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-04 23:03 UTC pith:EAPNLDPJ

load-bearing objection A competent adaptation of BBDM to cardiac motion fields, but the 'significant' gains over LaMoD are small and unsupported by significance tests; the coordinate-alignment assumption is a real unverified weakness. the 3 major comments →

arxiv 2608.01677 v1 pith:EAPNLDPJ submitted 2026-08-03 cs.CV cs.LG

Generative Brownian Bridge Diffusion In Motion Space For Enhanced Myocardial Strain Analysis

Rishov Paul , Frederick H. Epstein , Miaomiao Zhang This is my paper
classification cs.CV cs.LG
keywords myocardial strainBrownian bridge diffusioncine CMRDENSE MRIimage registrationmotion refinementgenerative modelcardiac function
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper's central claim is that a Brownian bridge diffusion model operating directly on spatiotemporal displacement fields can take the rough motion estimates produced by standard image registration on routine cine CMR scans and refine them into motion close to what DENSE, an advanced MRI technique that directly encodes tissue displacement, measures. The refinement is conditioned on the original CMR image contours so the generated motion stays anatomically consistent. On a paired multi-center dataset of cine and DENSE acquisitions, the paper reports that this reduces errors in whole-slice and segmental myocardial strain estimates below existing learning-based methods, with the largest gains at end-systole. If correct, the approach would make DENSE-quality strain analysis available from routine scans without the specialized acquisition, at post-processing cost. That matters because myocardial strain is a clinically useful measure of cardiac function, but current access is limited by human-adjusted post-processing or costly imaging.

Core claim

Using paired fields (x, u) of DENSE-quality motion and registration-derived motion registered to the same reference image, the paper defines a forward process that linearly interpolates from x toward u with variance schedule δs = 2τ(ms − ms²), then trains a conditional denoiser to reverse the bridge. The network is conditioned on cardiac contours via cross-attention and predicts a bridge gradient rather than raw noise; the output at s = 0 is the refined motion. The paper's central claim is that this stochastic bridge learns the multimodal mapping between registration output and high-fidelity DENSE motion, so that at inference the model can start from registration motion and return a refined

What carries the argument

Brownian bridge diffusion in motion space: a generative process whose forward path starts at the DENSE ground-truth motion x and ends at the registration motion u over a monotone schedule ms = s/S, with variance δs = 2τ(ms − ms²), and whose reverse path, conditioned on CMR image features, removes the bridge noise to recover a motion field close to x. The model predicts the bridge gradient bs = ms(u − x) + √δs ε, and the strain output is the Green-Lagrange strain of the resulting displacement field.

Load-bearing premise

The registration-derived motion and the DENSE reference motion lie in the same spatial coordinate system relative to the same reference image, so the Brownian bridge can interpolate between fields without correcting for misalignment or reparameterization.

What would settle it

Take a paired cine-DENSE test case, apply a known small translation (e.g. 2 mm) to the registration motion field before feeding it into the model, and measure the refined strain error against DENSE. If the error jumps by the size of the shift, the claimed refinement depends on exact coordinate pre-alignment rather than on learned motion enhancement.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

Share X Bluesky LinkedIn Reddit HN

If this is right

  • Strain estimates from routine cine CMR can approach DENSE reference accuracy without DENSE acquisition; the paper reports whole-slice strain MAE drops from 3.69% (best baseline) to 3.18%, and segmental end-systolic strain improves from 5.80% to 5.17%.
  • The framework is architecture-agnostic, so swapping the registration backbone should not require changes to the diffusion formulation, and the same bridge could refine motion from other registration methods.
  • Because the model is trained only on DENSE data and applied to cine at inference, it offers a cost-effective route to strain analysis in busy workflows without altering scan protocols.
  • Contour conditioning contributes part of the error reduction, as the ablation study shows removing it increases both displacement and strain errors.
  • Lower standard deviations of errors across all metrics suggest more stable motion trajectories, which matters for clinical repeatability.

Where Pith is reading between the lines

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

  • The same Brownian-bridge scheme could be applied to other paired cardiac-motion modalities (e.g., tagging or feature tracking as input, DENSE as target), or to other organs where a low-cost motion estimate and a high-cost reference coexist; the paper does not test these settings.
  • If the coordinate-system assumption holds only when DENSE fields are processed into the same grid as cine registration, applying the model to unaligned acquisitions would require an explicit pre-registration step; the current evaluation does not address breath-hold or through-plane motion.
  • A natural extension is to replace the fixed linear schedule ms = s/S with an anatomy-aware schedule that slows the bridge in regions of high strain variation, potentially improving segmental end-systolic estimates; the paper keeps the schedule fixed.
  • The model's stochasticity could be sampled multiple times per patient to produce a distribution of strain curves and per-segment uncertainty estimates, an option the paper does not explore.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper proposes a conditional Brownian bridge diffusion model operating in spatiotemporal motion/displacement space to map registration-derived motion fields (estimated from standard cine CMR via a temporal registration network) into high-quality DENSE motion fields, with the aim of improving myocardial strain estimation from routine cine CMR. The forward process linearly interpolates between the DENSE target x and the registration-derived source u, and the reverse process is conditioned on CMR image features/contours. Experiments on a multi-center paired cine-DENSE dataset compare against several deterministic and generative baselines, report DENSE displacement EPE and cine-derived strain MAE under multiple evaluation settings, and include an ablation study showing each component contributes. The authors claim significant improvements in motion prediction and strain analysis.

Significance. If the central claim is supported, the framework would provide a cost-effective way to obtain DENSE-quality strain estimates from standard cine CMR, which is clinically valuable given the cost and limited availability of DENSE. The paper provides code, subject-level train/test splits, and an ablation that independently evaluates the Brownian bridge and contour guidance. However, the evidence for 'significant' improvement is currently weak, and a load-bearing alignment assumption is asserted rather than validated. The methodological novelty is moderate, building on an existing Brownian bridge image translation framework (BBDM) and applying it to a new motion-space domain.

major comments (3)
  1. [§4, Table 1] The abstract and Section 5 claim 'significant improvements' over learning-based methods, but no statistical tests are reported. For example, whole-slice MAE is 3.18±1.75 vs. LaMoD's 3.69±1.71 (difference 0.51 pp), and segmental MAE is 4.14±1.49 vs. 4.22±1.45 (difference 0.08 pp). These differences are well within one standard deviation and may not be statistically significant. Please add paired significance tests (e.g., Wilcoxon signed-rank or paired t-test) with effect sizes and confidence intervals, or soften the 'significant' claim throughout.
  2. [§3, Eq. (4)] The forward process linearly interpolates between the DENSE field x and the registration-derived field u, based on the assumption that they are 'in the same spatial coordinate system using the same reference image.' This assumption is stated but never verified. DENSE displacement is encoded relative to its own reference frame, while cine registration is relative to the cine reference; differences in slice position, breath-hold, and through-plane motion could break pointwise correspondence. If x and u are misaligned, the bridge targets mismatched fields and the measured strain gains may reflect systematic coordinate correction rather than genuine motion refinement. Please validate this alignment (e.g., report residual registration errors between cine-derived fields and DENSE fields in a common frame, or show that EPE for cine-derived displacement vs. DENSE is meaningful after alignment) a
  3. [§4, Dataset] The description of training data is ambiguous. The text says the model is 'trained exclusively on the DENSE-CMR dataset' but also describes 'paired standard cine CMR and advanced strain imaging acquisitions' and subject-level splits. Please clarify how paired (u, x) training examples are constructed: is u computed from cine images or from DENSE magnitude images? If training uses DENSE-derived registration fields, the domain gap between DENSE-derived u and cine-derived u at test time is a potential confounder that should be discussed. This affects the interpretation of the core contribution.
minor comments (5)
  1. [§4, Dataset] 'All CMR videos are cropped to a 482 ×20 LV region of interest' — likely a typo for '48×20' or similar; please correct.
  2. [Fig. 2] The caption reads 'Left to right: visual examples...' but the subfigure layout is not self-explanatory; please clarify the ordering of panels.
  3. [§2, Eq. (1)] The text states SSD is adopted, but Eq. (1) uses a generic Dist(·,·); please make the connection explicit.
  4. [§4, Table 1] The standard deviations are reported but the numbers of subjects/sequences per split are not given. Please report the test-set size and the number of paired cine-DENSE cases used for each metric.
  5. [§5] The conclusion repeats the abstract's 'significant improvements' without qualification; please align with the statistical evidence.

Circularity Check

0 steps flagged

No circular derivation: the Brownian-bridge mapping is trained under held-out supervised targets; self-citations are non-load-bearing.

full rationale

The derivation chain is not circular. The Brownian bridge in Eq. (4) is a standard conditional translation construction: the endpoints are the DENSE-quality motion x and the registration-derived motion u, and the forward process is a variance-preserving interpolation between them. The reverse model is trained with the supervised loss in Eq. (6) against paired targets, and evaluation is on subject-disjoint splits for both DENSE displacement EPE and cine-derived strain MAE. The registration motion u is produced by an independent registration network, not fitted from the DENSE target x in a way that forces the reported strain values. The bridge framework is explicitly borrowed from external prior work [16], not from the authors' own results. Self-citations to TLRN [33] and LaMoD [34] are used as a registration backbone and a comparison baseline, respectively; neither is invoked as a uniqueness theorem or as the source of the load-bearing claim, and the ablation provides internal checks that the learned module contributes beyond the registration-only baseline. The explicit assumption that x and u are in the same spatial coordinate system with the same reference image is a stated premise and an empirical risk, not a circular reduction: if the fields are misaligned the learned mapping may be impaired, but the method still does not assume the conclusion it claims to demonstrate. No specific circular step—self-definitional, fitted-input-as-prediction, self-citation load-bearing, uniqueness-imported, ansatz-smuggled, or renaming—could be identified in the paper's equations or evaluation protocol.

Axiom & Free-Parameter Ledger

4 free parameters · 5 axioms · 0 invented entities

The central claim rests on the standard BBDM machinery, an alignment assumption between two motion domains, an unstated domain-transfer assumption for the registration network, and the availability of contours at inference. No new physical entities are introduced.

free parameters (4)
  • tau (variance scaling factor in Brownian bridge forward process) = 1
    Set by grid search; controls the amount of noise in the diffusion process (Eq. 4).
  • lambda_reg (spatial regularization weight in diffusion loss) = 0.01
    Set by grid search; penalizes non-smooth motion updates in Eq. (6).
  • S (number of diffusion steps) = not reported
    The Brownian bridge process uses an integer number of steps S, but the value is never disclosed; only a subsampled non-Markovian schedule is mentioned.
  • registration loss weights lambda and beta (Eq. 3) = not reported
    The registration network in Eq. (3) has a spatial regularization weight lambda and temporal regularization weight beta, but their values are not disclosed.
axioms (5)
  • domain assumption BBDM forward and reverse processes as defined in [16] are valid for motion fields
    The paper adopts the Brownian bridge diffusion formulation from Li et al. [16] without modification, treating motion fields as images in R^{d×T×H×W}.
  • domain assumption Paired DENSE and registration-derived motion fields are in the same spatial coordinate system with respect to the same reference image, so linear interpolation in Eq. (4) introduces no misalignment
    Stated explicitly in Section 3. This is the load-bearing alignment assumption for the bridge to be meaningful.
  • standard math Green-Lagrange strain computed from the deformation gradient via Eq. (2) is the correct strain measure
    Standard continuum mechanics; used to compute strain from predicted displacement fields.
  • domain assumption The registration network trained on DENSE-CMR magnitude images generalizes to routine cine CMR at inference
    The paper states the model is 'trained exclusively on the DENSE-CMR dataset' yet evaluates on cine; no domain adaptation or fine-tuning on cine is described.
  • domain assumption Myocardial contours used as conditioning c are available at inference time and are consistent across modalities
    The conditioning signal c is described as segmented myocardium contours, but the paper does not state whether contours at inference come from manual annotation or an automated segmenter, nor how they are aligned with the motion fields.

pith-pipeline@v1.3.0-daily-deepseek · 7775 in / 17674 out tokens · 174298 ms · 2026-08-04T23:03:37.913745+00:00 · methodology

0 comments
Cite this review

Pith. "Pith review of Generative Brownian Bridge Diffusion In Motion Space For Enhanced Myocardial Strain Analysis." pith.science (2026). https://pith.science/paper/EAPNLDPJ

@misc{pith2026260801677,
  author       = {Pith},
  title        = {Pith review of: Generative Brownian Bridge Diffusion In Motion Space For Enhanced Myocardial Strain Analysis},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/EAPNLDPJ}},
  note         = {Machine review of arXiv:2608.01677}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Myocardial strain analysis of cardiac magnetic resonance (CMR) images provides an important tool for evaluating cardiac function. However, current techniques require either human-adjusted post-processing with suboptimal regional accuracy, or specialized and costly imaging acquisitions. In this paper, we propose to leverage the power of generative models to synthesize high-quality motion-derived strain values from routinely acquired CMR sequences. Specifically, we develop a novel Brownian bridge diffusion model in motion space to learn the probabilistic mapping between standard CMR motion estimated from widely adopted registration methods and highly accurate motion provided by advanced strain imaging techniques. To promote the fidelity of anatomical structure in the generation process, our model is conditioned on the corresponding CMR images. We validate our method on large-scale multi-center CMR datasets including subjects of paired standard cine CMR and advanced strain imaging acquisitions. Experimental results demonstrate that our framework significantly improves the accuracy of motion prediction and strain analysis from standard CMRs compared to existing learning-based approaches. Our research represents a new paradigm for potentially developing cost-effective, clinically deployable AI tools for cardiac function assessment with enhanced strain accuracy in busy clinical workflows. Our code is publicly available at anonymous.4open.science/r/Brownian-Bridge-strain-analysis-1140.

Figures

Figures reproduced from arXiv: 2608.01677 by Frederick H. Epstein, Miaomiao Zhang, Rishov Paul.

Figure 1
Figure 1. Figure 1: An overview of our Brownian bridge diffusion model in motion space. [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Left to right: visual examples of strain curves VS. ES displacement frames [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗

discussion (0)

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

Reference graph

Works this paper leans on

36 extracted references · 22 canonical work pages

  1. [1]

    European Heart Journal-Cardiovascular Imaging20(6), 605–619 (2019)

    Amzulescu, M.S., De Craene, M., Langet, et al.: Myocardial strain imaging: re- view of general principles, validation, and sources of discrepancies. European Heart Journal-Cardiovascular Imaging20(6), 605–619 (2019)

    work page 2019

  2. [2]

    In: Annales de l’institut Fourier

    Arnold, V.: Sur la géométrie différentielle des groupes de lie de dimension infinie et ses applications à l’hydrodynamique des fluides parfaits. In: Annales de l’institut Fourier. vol. 16, pp. 319–361 (1966)

    work page 1966

  3. [3]

    In: International Conference on Medical Image Computing and Computer-Assisted Intervention

    Arsigny, V., Commowick, O., Pennec, X., Ayache, N.: A log-euclidean framework for statistics on diffeomorphisms. In: International Conference on Medical Image Computing and Computer-Assisted Intervention. pp. 924–931. Springer (2006)

    2006

  4. [4]

    IEEE transactions on medical imaging38(8), 1788–1800 (2019)

    Balakrishnan, G., Zhao, A., Sabuncu, M.R., Guttag, J., Dalca, A.V.: Voxelmorph: a learning framework for deformable medical image registration. IEEE transactions on medical imaging38(8), 1788–1800 (2019)

    2019

  5. [5]

    In: Proceedings of the IEEE conference on computer vision and pattern recognition

    Balakrishnan, G., Zhao, A., Sabuncu, M.R., et al.: An unsupervised learning model for deformable medical image registration. In: Proceedings of the IEEE conference on computer vision and pattern recognition. pp. 9252–9260 (2018)

    work page 2018

  6. [6]

    Computer Methods in Applied Mechanics and Engineering423, 116845 (2024)

    Barnafi, N., Regazzoni, F., et al.: Reconstructing relaxed configurations in elastic bodies: Mathematical formulations and numerical methods for cardiac modeling. Computer Methods in Applied Mechanics and Engineering423, 116845 (2024)

    work page 2024

  7. [7]

    International journal of computer vision61, 139–157 (2005)

    Beg, M.F., Miller, M.I., Trouvé, A., Younes, L.: Computing large deformation metric mappings via geodesic flows of diffeomorphisms. International journal of computer vision61, 139–157 (2005)

    2005

  8. [8]

    In: Biennial International Conference on Information Processing in Medical Imaging

    Chandrashekara, R., Rao, A., Sanchez-Ortiz, G.I., Mohiaddin, R.H., Rueckert, D.: Construction of a statistical model for cardiac motion analysis using nonrigid image registration. In: Biennial International Conference on Information Processing in Medical Imaging. pp. 599–610. Springer (2003)

    work page 2003

  9. [9]

    arXiv preprint arXiv:2102.04306 (2021)

    Chen, J., Lu, Y., Yu, Q., et al.: Transunet: Transformers make strong encoders for medical image segmentation. arXiv preprint arXiv:2102.04306 (2021)

    Pith/arXiv arXiv 2021

  10. [10]

    Medical image analysis82, 102615 (2022)

    Chen, J., Frey, E.C., He, Y., et al.: Transmorph: Transformer for unsupervised medical image registration. Medical image analysis82, 102615 (2022)

    work page 2022

  11. [11]

    Medical image analysis57, 226–236 (2019)

    Dalca, A.V., Balakrishnan, G., Guttag, J., Sabuncu, M.R.: Unsupervised learning of probabilistic diffeomorphic registration for images and surfaces. Medical image analysis57, 226–236 (2019)

    work page 2019

  12. [12]

    Frontiers in Cardiovascular Medicine10, 1095159 (2023)

    Ghadimi, S., Abdi, M., Epstein, F.H.: Improved computation of lagrangian tissue displacement and strain for cine dense mri using a regularized spatiotemporal least squares method. Frontiers in Cardiovascular Medicine10, 1095159 (2023)

    work page 2023

  13. [13]

    In: Proceedings of the IEEE/CVF winter conference on applications of computer vision

    Hatamizadeh, A., Tang, Y., Nath, V., et al.: Unetr: Transformers for 3d medi- cal image segmentation. In: Proceedings of the IEEE/CVF winter conference on applications of computer vision. pp. 574–584 (2022)

    work page 2022

  14. [14]

    He,K.,Zhang,X.,Ren,S.,Sun,J.:Deepresiduallearningforimagerecognition.In: Proceedings of the IEEE conference on computer vision and pattern recognition. pp. 770–778 (2016)

    2016

  15. [15]

    CoRR abs/1412.6980(2014), https://api.semanticscholar.org/CorpusID:6628106

    Kingma, D.P., Ba, J.: Adam: A method for stochastic optimization. CoRR abs/1412.6980(2014), https://api.semanticscholar.org/CorpusID:6628106

    Pith/arXiv arXiv 2014

  16. [16]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern Recognition

    Li, B., Xue, K., Liu, B., Lai, Y.K.: Bbdm: Image-to-image translation with brow- nian bridge diffusion models. In: Proceedings of the IEEE/CVF conference on computer vision and pattern Recognition. pp. 1952–1961 (2023)

    1952

  17. [17]

    In: The Eleventh International Conference on Learning Representations (2023), https://openreview.net/forum?id=PqvMRDCJT9t 10 R

    Lipman, Y., Chen, R.T.Q., Ben-Hamu, H., Nickel, M., Le, M.: Flow matching for generative modeling. In: The Eleventh International Conference on Learning Representations (2023), https://openreview.net/forum?id=PqvMRDCJT9t 10 R. Paul et al

    work page 2023

  18. [18]

    JAMA cardiology4(3), 287–294 (2019)

    Marwick, T.H., Shah, S.J., Thomas, J.D.: Myocardial strain in the assessment of patients with heart failure: a review. JAMA cardiology4(3), 287–294 (2019)

    work page 2019

  19. [19]

    Frontiers in Cardiovascular Medicine8, 730316 (2021)

    Morales,M.A.,VandenBoomen,M.,Nguyen,C.,etal.:Deepstrain:adeeplearning workflow for the automated characterization of cardiac mechanics. Frontiers in Cardiovascular Medicine8, 730316 (2021)

    work page 2021

  20. [20]

    In: International Workshop on Machine Learning for Medical Image Reconstruction

    Pan, J., Rueckert, D., Küstner, T., Hammernik, K.: Efficient image registration network for non-rigid cardiac motion estimation. In: International Workshop on Machine Learning for Medical Image Reconstruction. pp. 14–24. Springer (2021)

    work page 2021

  21. [21]

    Medical Physics47(9), 4189–4198 (2020)

    Qiao, M., Wang, Y., Guo, Y., Huang, L., Xia, L., Tao, Q.: Temporally coherent cardiacmotiontrackingfromcinemri:Traditionalregistrationmethodandmodern cnn method. Medical Physics47(9), 4189–4198 (2020)

    work page 2020

  22. [22]

    Medical Image Analysis83, 102682 (2023)

    Qin, C., Wang, S., Chen, C., Bai, W., Rueckert, D.: Generative myocardial motion tracking via latent space exploration with biomechanics-informed prior. Medical Image Analysis83, 102682 (2023)

    2023

  23. [23]

    In: Medical Image Computing and Computer-Assisted Intervention–MICCAI 2015: 18th International Conference, Munich, Germany, Oc- tober 5-9, 2015, Proceedings, Part III 18

    Ronneberger, O., Fischer, P., Brox, T.: U-net: Convolutional networks for biomed- ical image segmentation. In: Medical Image Computing and Computer-Assisted Intervention–MICCAI 2015: 18th International Conference, Munich, Germany, Oc- tober 5-9, 2015, Proceedings, Part III 18. pp. 234–241. Springer (2015)

    2015

  24. [24]

    In: International Conference on Medical Image Computing and Computer-Assisted Intervention

    Rueckert,D.,Aljabar,P.,Heckemann,R.A.,Hajnal,J.V.,Hammers,A.:Diffeomor- phic registration using b-splines. In: International Conference on Medical Image Computing and Computer-Assisted Intervention. pp. 702–709. Springer (2006)

    work page 2006

  25. [25]

    In: 4th MICCAI Workshop on Mathematical Foundations of Computational Anatomy

    Seo, D., Ho, J., Traverse, J.H., Forder, J., Vemuri, B.: Computing diffeomorphic paths with applications to cardiac motion analysis. In: 4th MICCAI Workshop on Mathematical Foundations of Computational Anatomy. pp. 83–94. Citeseer (2013)

    work page 2013

  26. [26]

    Smiseth, O.A., Torp, H., Opdahl, A., et al.: Myocardial strain imaging: how useful is it in clinical decision making? European heart journal37(15), 1196–1207 (2016)

    work page 2016

  27. [27]

    IEEE transactions on medical imaging32(7), 1153–1190 (2013)

    Sotiras, A., Davatzikos, C., Paragios, N.: Deformable medical image registration: A survey. IEEE transactions on medical imaging32(7), 1153–1190 (2013)

    2013

  28. [28]

    arXiv preprint arXiv:2302.00482 (2023)

    Tong, A., Fatras, K., et al.: Improving and generalizing flow-based generative mod- els with minibatch optimal transport. arXiv preprint arXiv:2302.00482 (2023)

    Pith/arXiv arXiv 2023

  29. [29]

    Advances in neural information processing systems30(2017)

    Vaswani, A., Shazeer, N., Parmar, N., et al.: Attention is all you need. Advances in neural information processing systems30(2017)

    work page 2017

  30. [30]

    IEEE transactions on medical imaging29(1), 1–18 (2010)

    Vercauteren, T., Pennec, X., Perchant, A., Ayache, N.: Symmetric log-domain dif- feomorphic registration: A demons-based approach. IEEE transactions on medical imaging29(1), 1–18 (2010)

    work page 2010

  31. [31]

    In: Advances in Neural Information Processing Systems (NeurIPS) (2018)

    Wang, T.C., Liu, M.Y., Zhu, J.Y., et al.: Video-to-video synthesis. In: Advances in Neural Information Processing Systems (NeurIPS) (2018)

    work page 2018

  32. [32]

    Radiology: Cardiothoracic Imaging 5(3), e220196 (2023)

    Wang, Y., Sun, C., Ghadimi, S., et al.: Strainnet: Improved myocardial strain anal- ysis of cine mri by deep learning from dense. Radiology: Cardiothoracic Imaging 5(3), e220196 (2023)

    work page 2023

  33. [33]

    In: International Conference on Medical Image Computing and Computer-Assisted Intervention

    Wu, N., Xing, J., Zhang, M.: Tlrn: Temporal latent residual networks for large deformation image registration. In: International Conference on Medical Image Computing and Computer-Assisted Intervention. pp. 728–738. Springer (2024)

    2024

  34. [34]

    In: International Workshop on Shape in Medical Imaging

    Xing, J., Jayakumar, N., Wu, N., Wang, Y., Epstein, F.H., Zhang, M.: Lamod: Latent motion diffusion model for myocardial strain generation. In: International Workshop on Shape in Medical Imaging. pp. 164–177. Springer (2024)

    2024

  35. [35]

    Zhang, L., Rao, A., Agrawala, M.: Adding conditional control to text-to-image diffusion models (2023)

    2023

  36. [36]

    Magnetic resonance in medicine64(4), 1089–1097 (2010)

    Zhong, X., Spottiswoode, B.S., Meyer, C.H., Kramer, C.M., Epstein, F.H.: Imaging three-dimensional myocardial mechanics using navigator-gated volumetric spiral cine dense mri. Magnetic resonance in medicine64(4), 1089–1097 (2010)

    work page 2010