A novel Brownian bridge diffusion model in motion space is developed 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.
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 acquisitions with limited availability. 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 https://github.com/Rishov-MIA/Brownian-Bridge-strain-analysis.
This work extends the previously proposed PINN framework with spatiotemporal implicit neural representations (INRs) to represent the MR signal as a continuous spatiotemporal function and to improve the accuracy, smoothness, and physical consistency of the PINN model.
Christos Tsepas, Yan Chang, M. Fuetterer et al.· 0 citations
Cardiac strains provide significant information to evaluate cardiac performance. They can be evaluated using full field voxel-wise displacements measured, for example, using displacement encoding with stimulated echoes (DENSE) magnetic resonance imaging (MRI) or from features and textures that are measured using Cine MRI. While DENSE MRI provides a more complete description of myocardial motion, it remains challenging to acquire and its adoption is not widespread. In contrast, Cine MRI is routinely acquired, and Cine-based cardiac strains could be more easily adopted in a clinical setting. To evaluate the feasibility of automatically estimating cardiac strains from Cine MRI, we compare slice-wise circumferential, longitudinal, and radial strains obtained with three mid-ventricular models: a two-slice DENSE model, a one-slice DENSE model, and a one-slice Cine model. Strains are computed after automatic image segmentation, and the models are evaluated using a computational deforming phantom and data acquired in forty (N=40\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$N=40$$\end{document}) healthy volunteers. When evaluated against the computational phantom, all models perform well in terms of circumferential strains, while differences are present in longitudinal and radial strains depending on model and location. When applied to volunteer data, the one-slice Cine model agrees reasonably well with the DENSE-based models in terms of endocardial circumferential strains while, overall, it leads to higher (in magnitude) radial and longitudinal strains, and lower (in magnitude) epicardial circumferential strains. The DENSE models lead to nearly identical estimates for circumferential and radial strains, while longitudinal strains are not correctly estimated by the one-slice DENSE model paired with automatic segmentation. The strain differences are discussed both group-wise and at the individual subject level.
Mohammad Naqizadeh Jahromi, Rodrigo Menna Costa, Augusto Delavald Marques et al.· Biomechanics and Modeling in...· 0 citations
Purpose: To develop and evaluate a diffusion-based reconstruction framework for highly accelerated 2D real-time (RT) cine cardiovascular magnetic resonance imaging (CMR). Methods: We trained an unconditional patch-based diffusion model and incorporated it into a reconstruction framework, termed CineDiff, using diffusion posterior sampling for data consistency. CineDiff was evaluated in four settings: (i) 30 retrospectively undersampled breath-held cine at 1.5T and 3T from healthy participants across multiple acceleration rates, (ii) 15 prospectively undersampled free-breathing RT cine at 1.5T and 3T from patients indicated for clinical CMR, and (iii) 10 prospectively undersampled mid-field (0.55T) free-breathing scans, including five from healthy subjects and five from porcine models. For retrospective undersampling, reconstruction quality was assessed using peak signal-to-noise ratio (PSNR), structural similarity index measure (SSIM), learned perceptual image patch similarity (LPIPS), and deep image structure and texture similarity (DISTS). For prospective undersampling, image quality was evaluated by blinded expert scoring on a 5-point Likert scale. Results: In retrospectively undersampled breath-held cine data, CineDiff achieved higher PSNR and SSIM and lower LPIPS and DISTS than the comparison methods across all evaluated acceleration rates. In prospectively undersampled free-breathing RT cine data, CineDiff received higher expert image-quality scores. Qualitatively, CineDiff reduced block-like artifacts and preserved finer anatomical detail compared with traditional compressed sensing and a variational network method, termed CineVN. Conclusion: CineDiff enabled high-quality reconstruction of highly accelerated 2D RT cine CMR. The method also demonstrated robustness to out-of-distribution data, including mid-field and porcine acquisitions.
Xuan Lei, Philip Schniter, Juliet Varghese et al.· 0 citations
Cardiac magnetic resonance (CMR) imaging provides complementary information on cardiac anatomy, function, and tissue characterization across multiple sequences and views. In this work, we investigate foundation model pretraining for 2D CMR and introduce CMRVision, a CMR-specific foundation model trained using DINOv3-style self-supervised learning on a multi-center, multi-sequence cohort of 36 million CMR images. We systematically evaluate architectural and training design choices for domain-specific pretraining. CMRVision is evaluated on two downstream tasks: multi-task segmentation across cine, late gadolinium enhancement (LGE), and mapping sequences, and cine view classification. Our experiments show that CMR-specific pretraining, smaller patch sizes, and patch-level objectives consistently improve downstream performance. Across a multi-task segmentation benchmark, CMRVision achieved the strongest overall performance, outperforming prior natural-image (NI), medical-image, supervised, and CMR foundation model baselines. Improvements were modest but consistent across structures and sequences, with Dice scores ranging from 0.940-0.967 for LV and 0.855-0.905 for myocardium, and reaching 0.929 for RV, 0.920 for LA, and 0.931 for RA. The largest gains were observed for myocardium segmentation in LGE and mapping images. In a zero-shot segmentation task on unseen LGE long-axis views, the model achieved an average Dice score of 0.692, demonstrating cross-view generalization. For cine view classification, CMRVision achieved the highest average accuracy (0.906), compared to prior methods reported in the literature. These results highlight the potential of CMRVision to support robust and generalizable cardiac MRI analysis across multiple sequences and views.
Athira J. Jacob, Puneet Sharma, D. Rueckert· 0 citations
It is a common desire to break traditional limits of sampling, circumventing Nyquist requirements using advanced statistical algorithms. In cardiac ultrasound imaging, reducing the number of transmissions per frame, thereby breaking spatial sampling restrictions, enables capturing faster moving structures or larger fields of view. However, the spatial and temporal image properties that may enable such acceleration have been underexplored in the ultrasound literature. This work provides a fundamental study of the structure of ultrasound images using simulations, phantoms, and in vivo cardiac data to quantify tensor rank and spatial/temporal roughness and explores the implications for tensor completion algorithms. Although low-rank reconstruction is a powerful approach that has been previously applied in this context, these data do not appear to be sufficiently low-rank for high-quality reconstructions. Two methods relying on local information-inverse distance-weighted (IDW) interpolation and the fast multiway delay-embedding transform-demonstrate significantly more accurate reconstruction (e.g., structured similarity index measure 0.76 vs 0.67 at 25% sampling and 0.63 vs 0.38 at 10% sampling for IDW versus low-rank reconstruction). Roughness in both space and time is shown to inversely correlate with tensor completion success. Motion compensation is shown to reduce both temporal roughness and rank, improving tensor completion.
Joshua Fry, Nazli Javadi Eshkalak, S. Becker et al.· Journal of the Acoustical So...· 0 citations
Purpose Quantitative assessment of myocardial deformation is increasingly important in clinical cardiology, yet conventional two-dimensional (2D) echocardiography and standard three-dimensional (3D) approaches remain limited by out-of-plane motion and incomplete characterization of transmural mechanics. To address these limitations, we introduce a physics-informed framework for 3D echocardiography that reconstructs the full finite strain tensor across the entire myocardial wall. As an initial methodological study, we demonstrate the framework and validate it against cardiac magnetic resonance in a small cohort. Methods Endocardial and epicardial surfaces were segmented from 3D echocardiographic datasets and tracked throughout the cardiac cycle using speckle-tracking techniques. An optimization framework with a soft volumetric penalty was implemented, permitting volume change at finite cost while maintaining tracking fidelity and geometric smoothness. The resulting deformation field enabled reconstruction of the complete 3D strain tensor. Global strain measurements derived from the method were validated against cardiac magnetic resonance (CMR) measurements obtained in two subjects. Results Global longitudinal and circumferential strain values obtained from the proposed framework showed strong agreement with CMR measurements. The optimization procedure also demonstrated robustness to segmentation variability and reduced errors associated with epicardial tracking. Beyond conventional strain indices, the method enabled reconstruction of spatially resolved principal strain fields throughout the ventricular wall, revealing physiologically consistent transmural gradients and contraction patterns. Conclusion Physics-informed integration of speckle tracking with biomechanical constraints enables robust reconstruction of 3D myocardial deformation from echocardiography. This framework provides a comprehensive and physically consistent characterization of myocardial mechanics from widely available 3D echocardiographic data. These initial results support the feasibility of the proposed framework and motivate future evaluation in larger, more diverse patient cohorts to establish its clinical reliability.
Satyaki Pradhan, Arash Yavari, Issac D. Lindley et al.· Annals of Biomedical Enginee...· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
