Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

GENRE-CMR: Generalizable Deep Learning for Diverse Multi-Domain Cardiac MRI Reconstruction

Kian Anvari Hamedani, Narges Razizadeh, Shahabedin Nabavi, Mohsen Ebrahimi Moghaddam

TL;DR

The paper tackles the challenge of accelerated Cardiovascular Magnetic Resonance (CMR) reconstruction under cross-domain domain shifts from diverse acquisitions. It introduces GENRE-CMR, a GAN-based residual deep unrolled reconstruction framework augmented with Edge-Aware Reconstruction (EAR) and Statistical Distribution Alignment (SDA) losses to improve edge fidelity and cross-domain generalization, respectively. The approach reports state-of-the-art performance on both in-distribution and out-of-distribution data, with metrics such as $ ext{SSIM}$, $ ext{PSNR}$, and $ ext{NMSE}$ achieving notable gains (e.g., unseen $ ext{SSIM}=0.9552$, $ ext{PSNR}=38.90$ dB). By leveraging a large, heterogeneous dataset (CMRxRecon 2025) and a curriculum training strategy, the method demonstrates robust generalization across contrasts, trajectories, and vendors, indicating strong potential for clinically adaptable deployment, albeit with high computational requirements.

Abstract

Accelerated Cardiovascular Magnetic Resonance (CMR) image reconstruction remains a critical challenge due to the trade-off between scan time and image quality, particularly when generalizing across diverse acquisition settings. We propose GENRE-CMR, a generative adversarial network (GAN)-based architecture employing a residual deep unrolled reconstruction framework to enhance reconstruction fidelity and generalization. The architecture unrolls iterative optimization into a cascade of convolutional subnetworks, enriched with residual connections to enable progressive feature propagation from shallow to deeper stages. To further improve performance, we integrate two loss functions: (1) an Edge-Aware Region (EAR) loss, which guides the network to focus on structurally informative regions and helps prevent common reconstruction blurriness; and (2) a Statistical Distribution Alignment (SDA) loss, which regularizes the feature space across diverse data distributions via a symmetric KL divergence formulation. Extensive experiments confirm that GENRE-CMR surpasses state-of-the-art methods on training and unseen data, achieving 0.9552 SSIM and 38.90 dB PSNR on unseen distributions across various acceleration factors and sampling trajectories. Ablation studies confirm the contribution of each proposed component to reconstruction quality and generalization. Our framework presents a unified and robust solution for high-quality CMR reconstruction, paving the way for clinically adaptable deployment across heterogeneous acquisition protocols.

GENRE-CMR: Generalizable Deep Learning for Diverse Multi-Domain Cardiac MRI Reconstruction

TL;DR

The paper tackles the challenge of accelerated Cardiovascular Magnetic Resonance (CMR) reconstruction under cross-domain domain shifts from diverse acquisitions. It introduces GENRE-CMR, a GAN-based residual deep unrolled reconstruction framework augmented with Edge-Aware Reconstruction (EAR) and Statistical Distribution Alignment (SDA) losses to improve edge fidelity and cross-domain generalization, respectively. The approach reports state-of-the-art performance on both in-distribution and out-of-distribution data, with metrics such as , , and achieving notable gains (e.g., unseen , dB). By leveraging a large, heterogeneous dataset (CMRxRecon 2025) and a curriculum training strategy, the method demonstrates robust generalization across contrasts, trajectories, and vendors, indicating strong potential for clinically adaptable deployment, albeit with high computational requirements.

Abstract

Accelerated Cardiovascular Magnetic Resonance (CMR) image reconstruction remains a critical challenge due to the trade-off between scan time and image quality, particularly when generalizing across diverse acquisition settings. We propose GENRE-CMR, a generative adversarial network (GAN)-based architecture employing a residual deep unrolled reconstruction framework to enhance reconstruction fidelity and generalization. The architecture unrolls iterative optimization into a cascade of convolutional subnetworks, enriched with residual connections to enable progressive feature propagation from shallow to deeper stages. To further improve performance, we integrate two loss functions: (1) an Edge-Aware Region (EAR) loss, which guides the network to focus on structurally informative regions and helps prevent common reconstruction blurriness; and (2) a Statistical Distribution Alignment (SDA) loss, which regularizes the feature space across diverse data distributions via a symmetric KL divergence formulation. Extensive experiments confirm that GENRE-CMR surpasses state-of-the-art methods on training and unseen data, achieving 0.9552 SSIM and 38.90 dB PSNR on unseen distributions across various acceleration factors and sampling trajectories. Ablation studies confirm the contribution of each proposed component to reconstruction quality and generalization. Our framework presents a unified and robust solution for high-quality CMR reconstruction, paving the way for clinically adaptable deployment across heterogeneous acquisition protocols.

Paper Structure

This paper contains 7 sections, 6 equations, 4 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Materials and Methods
  3. Dataset
  4. A Residual Deep Unrolled Reconstruction Framework
  5. Implementation Details
  6. Experimental Results and Discussion
  7. Conclusion

Figures (4)

  • Figure 1: Overview of the GENRE-CMR method.
  • Figure 2: Illustration of the EAR Loss computation. Top-left: original/reconstructed image. Top-right: binary edge mask extracted from the image. Bottom-left: edge mask fused on the image. Bottom-right: image edges extracted using the mask.
  • Figure 3: Dynamic weighting of loss components during training. The plot illustrates how each loss weight changes over time, enabling adaptive emphasis on different loss terms throughout the optimization process.
  • Figure 4: Qualitative comparison of reconstruction results under three different sampling trajectories at an acceleration factor of 16.