Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

$k$-$t$ CLAIR: Self-Consistency Guided Multi-Prior Learning for Dynamic Parallel MR Image Reconstruction

Liping Zhang, Weitian Chen

TL;DR

This work tackles the slow acquisition problem in dynamic cardiac MRI by introducing k-t CLAIR, a self-consistency guided multi-prior learning framework that exploits spatiotemporal redundancy across the $x$-$t$, $x$-$f$, and $k$-$t$ domains within an unrolled architecture. It jointly trains priors in each domain via specialized CNNs ($xt$-CNN, $xf$-CNN, $kt$-CNN) and a calibration-aware $calib$-CNN, coordinated by a frequency fusion layer and enhanced with coil-sensitivity learning, to enforce data fidelity and high-frequency restoration. The model optimizes a composite loss with $L_1$ and SSIM terms and leverages ACS data for calibration in the $k$-$t$ domain, yielding improved reconstructions for both accelerated cine and T1/T2 mapping tasks at accelerations up to $10\times$. Experiments on the CMRxRecon dataset demonstrate superior quantitative (SSIM, NMSE, PSNR) and qualitative performance against baselines including Zero-filled, FastMRI U-Net, and E2EVarNet, indicating strong potential for real-time dynamic CMR. Overall, kt CLAIR advances dynamic parallel MRI by effectively integrating multi-domain priors with self-consistency and calibration to achieve faithful, artifact-suppressed reconstructions.

Abstract

Cardiac magnetic resonance imaging (CMR) has been widely used in clinical practice for the medical diagnosis of cardiac diseases. However, the long acquisition time hinders its development in real-time applications. Here, we propose a novel self-consistency guided multi-prior learning framework named $k$-$t$ CLAIR to exploit spatiotemporal correlations from highly undersampled data for accelerated dynamic parallel MRI reconstruction. The $k$-$t$ CLAIR progressively reconstructs faithful images by leveraging multiple complementary priors learned in the $x$-$t$, $x$-$f$, and $k$-$t$ domains in an iterative fashion, as dynamic MRI exhibits high spatiotemporal redundancy. Additionally, $k$-$t$ CLAIR incorporates calibration information for prior learning, resulting in a more consistent reconstruction. Experimental results on cardiac cine and T1W/T2W images demonstrate that $k$-$t$ CLAIR achieves high-quality dynamic MR reconstruction in terms of both quantitative and qualitative performance.

$k$-$t$ CLAIR: Self-Consistency Guided Multi-Prior Learning for Dynamic Parallel MR Image Reconstruction

TL;DR

This work tackles the slow acquisition problem in dynamic cardiac MRI by introducing k-t CLAIR, a self-consistency guided multi-prior learning framework that exploits spatiotemporal redundancy across the -, -, and - domains within an unrolled architecture. It jointly trains priors in each domain via specialized CNNs (-CNN, -CNN, -CNN) and a calibration-aware -CNN, coordinated by a frequency fusion layer and enhanced with coil-sensitivity learning, to enforce data fidelity and high-frequency restoration. The model optimizes a composite loss with and SSIM terms and leverages ACS data for calibration in the - domain, yielding improved reconstructions for both accelerated cine and T1/T2 mapping tasks at accelerations up to . Experiments on the CMRxRecon dataset demonstrate superior quantitative (SSIM, NMSE, PSNR) and qualitative performance against baselines including Zero-filled, FastMRI U-Net, and E2EVarNet, indicating strong potential for real-time dynamic CMR. Overall, kt CLAIR advances dynamic parallel MRI by effectively integrating multi-domain priors with self-consistency and calibration to achieve faithful, artifact-suppressed reconstructions.

Abstract

Cardiac magnetic resonance imaging (CMR) has been widely used in clinical practice for the medical diagnosis of cardiac diseases. However, the long acquisition time hinders its development in real-time applications. Here, we propose a novel self-consistency guided multi-prior learning framework named - CLAIR to exploit spatiotemporal correlations from highly undersampled data for accelerated dynamic parallel MRI reconstruction. The - CLAIR progressively reconstructs faithful images by leveraging multiple complementary priors learned in the -, -, and - domains in an iterative fashion, as dynamic MRI exhibits high spatiotemporal redundancy. Additionally, - CLAIR incorporates calibration information for prior learning, resulting in a more consistent reconstruction. Experimental results on cardiac cine and T1W/T2W images demonstrate that - CLAIR achieves high-quality dynamic MR reconstruction in terms of both quantitative and qualitative performance.
Paper Structure (17 sections, 7 equations, 5 figures, 2 tables)

This paper contains 17 sections, 7 equations, 5 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Methods
  3. Dynamic Parallel MRI Problem Formulation
  4. $k$-$t$ CLAIR for Dynamic Parallel MRI Reconstruction
  5. Image enhancement prior in the $x$-$t$ domain
  6. Dynamic temporal prior in the $x$-$f$ domain
  7. Self-Consistency guided prior in the $k$-$t$ domain
  8. Frequency fusion block
  9. Objective Function and Evaluation Metric
  10. Experiment and Results
  11. Data and Implementation Details
  12. Implementation Details
  13. Experiments and Results
  14. Reconstruction of Accelerated T1/T2 Acquisitions for T1/T2 mapping
  15. Accelerated Cine Images Reconstruction
  16. ...and 2 more sections

Figures (5)

  • Figure 1: Illustration of the overall architecture of the proposed $k$-$t$ CLAIR.
  • Figure 2: Reconstruction of 10$\times$ accelerated T1W/T2W images and masked error maps.
  • Figure 3: Performance for T1W/T2W reconstruction on the unseen training data.
  • Figure 4: Reconstruction of 10$\times$ accelerated CINE and masked error maps.
  • Figure 5: Performance for CINE reconstruction on the unseen training data.