Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

UPCMR: A Universal Prompt-guided Model for Random Sampling Cardiac MRI Reconstruction

Donghang Lyu, Chinmay Rao, Marius Staring, Matthias J. P. van Osch, Mariya Doneva, Hildo J. Lamb, Nicola Pezzotti

TL;DR

The paper tackles the challenge of rapid, high-quality cardiac MRI by proposing UPCMR, a universal unrolled reconstruction model that uses two learnable prompts (undersampling-specific and spatial-specific) embedded in a two-level UNet per cascade block. It jointly learns to reconstruct image sequences and predict k-space trajectory and acceleration factor, guided by a curriculum-learning training strategy to handle diverse random undersampling scenarios. Key contributions include the prompt-guided UNet architecture, the dual-prompt mechanism with block-wise aggregation, and an evaluation showing superior performance over traditional methods across multiple sampling configurations, along with ablations confirming the importance of each component. The work advances practical, adaptable CMR reconstruction with potential to reduce scan times while maintaining diagnostic image quality, and suggests future improvements in temporal modeling and contrast-aware prompting.

Abstract

Cardiac magnetic resonance imaging (CMR) is vital for diagnosing heart diseases, but long scan time remains a major drawback. To address this, accelerated imaging techniques have been introduced by undersampling k-space, which reduces the quality of the resulting images. Recent deep learning advancements aim to speed up scanning while preserving quality, but adapting to various sampling modes and undersampling factors remains challenging. Therefore, building a universal model is a promising direction. In this work, we introduce UPCMR, a universal unrolled model designed for CMR reconstruction. This model incorporates two kinds of learnable prompts, undersampling-specific prompt and spatial-specific prompt, and integrates them with a UNet structure in each block. Overall, by using the CMRxRecon2024 challenge dataset for training and validation, the UPCMR model highly enhances reconstructed image quality across all random sampling scenarios through an effective training strategy compared to some traditional methods, demonstrating strong adaptability potential for this task.

UPCMR: A Universal Prompt-guided Model for Random Sampling Cardiac MRI Reconstruction

TL;DR

The paper tackles the challenge of rapid, high-quality cardiac MRI by proposing UPCMR, a universal unrolled reconstruction model that uses two learnable prompts (undersampling-specific and spatial-specific) embedded in a two-level UNet per cascade block. It jointly learns to reconstruct image sequences and predict k-space trajectory and acceleration factor, guided by a curriculum-learning training strategy to handle diverse random undersampling scenarios. Key contributions include the prompt-guided UNet architecture, the dual-prompt mechanism with block-wise aggregation, and an evaluation showing superior performance over traditional methods across multiple sampling configurations, along with ablations confirming the importance of each component. The work advances practical, adaptable CMR reconstruction with potential to reduce scan times while maintaining diagnostic image quality, and suggests future improvements in temporal modeling and contrast-aware prompting.

Abstract

Cardiac magnetic resonance imaging (CMR) is vital for diagnosing heart diseases, but long scan time remains a major drawback. To address this, accelerated imaging techniques have been introduced by undersampling k-space, which reduces the quality of the resulting images. Recent deep learning advancements aim to speed up scanning while preserving quality, but adapting to various sampling modes and undersampling factors remains challenging. Therefore, building a universal model is a promising direction. In this work, we introduce UPCMR, a universal unrolled model designed for CMR reconstruction. This model incorporates two kinds of learnable prompts, undersampling-specific prompt and spatial-specific prompt, and integrates them with a UNet structure in each block. Overall, by using the CMRxRecon2024 challenge dataset for training and validation, the UPCMR model highly enhances reconstructed image quality across all random sampling scenarios through an effective training strategy compared to some traditional methods, demonstrating strong adaptability potential for this task.

Paper Structure

This paper contains 12 sections, 7 equations, 2 figures, 1 table.

Table of Contents

  1. Introduction
  2. Preliminaries and Dataset
  3. Preliminaries
  4. CMRxRecon2024 Dataset
  5. Methodology
  6. Model Structure
  7. Loss Function
  8. Training Strategy
  9. Experiments
  10. Implementation Details
  11. Results
  12. Discussion and Conclusion

Figures (2)

  • Figure 1: Overview of the UPCMR model. In each cascade block, green lines represent feature maps transitioning from the previous one to the next block. Blue and red lines highlight the direction of undersampling-specific prompts and spatial-specific prompts, respectively. Black dashed lines refer to the skip connections between the corresponding encoder and decoder components.
  • Figure 2: Visualization of UPCMR reconstruction results for four contrasts under two k-space trajectories with an acceleration factor of 20. 'GND' indicates images from fully-sampled k-space, 'ZF' denotes images from zero-filled undersampled k-space, 'G' refers to Gaussian k-space, and 'R' stands for pseudo radial k-space. Images are cropped to focus on the central cardiac region. Rows from top to bottom show Aorta (sagittal), Cine (LAX), Mapping (T1), and Tagging.