RSFR: A Coarse-to-Fine Reconstruction Framework for Diffusion Tensor Cardiac MRI with Semantic-Aware Refinement

TL;DR

RSFR addresses the critical challenges of cardiac diffusion tensor imaging reconstruction under high undersampling by integrating a coarse-to-fine pipeline with zero-shot SAM priors. The framework combines a Vision Mamba-based Reconstruction Backbone with a Fusion & Refinement module that leverages semantic priors for improved myocardial fidelity, achieving state-of-the-art reconstruction quality and DT parameter accuracy. Extensive experiments and ablations validate the benefits of semantic integration and backbone design, highlighting robustness, scalability, and potential for clinical translation in quantitative cardiac DTI. The modular architecture supports end-to-end training with frozen SAM and adaptable backbones, offering a practical pathway toward reliable, artifact-resilient cDTI in routine clinical settings.

Abstract

Cardiac diffusion tensor imaging (DTI) offers unique insights into cardiomyocyte arrangements, bridging the gap between microscopic and macroscopic cardiac function. However, its clinical utility is limited by technical challenges, including a low signal-to-noise ratio, aliasing artefacts, and the need for accurate quantitative fidelity. To address these limitations, we introduce RSFR (Reconstruction, Segmentation, Fusion & Refinement), a novel framework for cardiac diffusion-weighted image reconstruction. RSFR employs a coarse-to-fine strategy, leveraging zero-shot semantic priors via the Segment Anything Model and a robust Vision Mamba-based reconstruction backbone. Our framework integrates semantic features effectively to mitigate artefacts and enhance fidelity, achieving state-of-the-art reconstruction quality and accurate DT parameter estimation under high undersampling rates. Extensive experiments and ablation studies demonstrate the superior performance of RSFR compared to existing methods, highlighting its robustness, scalability, and potential for clinical translation in quantitative cardiac DTI.

Figures (6)

• Figure 1: The contribution of our RSFR framework: (Q1) Conventional cDWI reconstruction suffers from low SNR, severe aliasing artefacts, and ROI-specific intensity sensitivity, resulting in limited reconstruction performance. (S1) Semantic-aware reconstruction integrates semantic priors into the reconstruction process, enhancing image quality and fidelity. (Q2) Conventional semantic-aware reconstruction extracts semantic priors directly from undersampled DWIs, which is unreliable due to poor segmentation quality from degraded images. (S2) Coarse-to-fine semantic-aware reconstruction improves segmentation reliability by generating semantic priors from coarse reconstructions, leading to better reconstruction performance. (Q3) Conventional segmentation models rely on labour-intensive manual annotation or specialised models, which suffer from distribution shifts and require additional training datasets. (S3) Zero-shot segmentation foundation model eliminates the need for manual annotation and task-specific model training, improving generalizability.
• Figure 2: The architecture of (A) the Reconstruction Model, (B) the Fusion & Refinement Model, and (C) the Semantic Feature Integration module.
• Figure 3: Visualised reconstructed DWI samples and corresponding error maps at AF$\times 2$, $\times 4$ and $\times 8$.
• Figure 4: Mean absolute error (MAE) of global mean diffusivity (MD), fractional anisotropy (FA), and helix angle (HA) gradient.
• Figure :