Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

TC-DiffRecon: Texture coordination MRI reconstruction method based on diffusion model and modified MF-UNet method

Chenyan Zhang, Yifei Chen, Zhenxiong Fan, Yiyu Huang, Wenchao Weng, Ruiquan Ge, Dong Zeng, Changmiao Wang

TL;DR

TC-DiffRecon tackles diffusion-model–based MRI reconstruction by addressing texture and coherence issues that arise from unconditional generation. It replaces the U-Net backbone with MF-UNet, introducing backbone and skip feature scaling factors and Fourier-domain modulation to preserve texture without excessive smoothing. The TCKG module integrates K-space data consistency in a texture-coordinated fashion, and a Coarse-to-Fine (C2F) sampling scheme accelerates inference while stabilizing texture. Experiments on FastMRI knee data demonstrate improved generalization across varying acceleration factors and superior image quality compared with state-of-the-art methods, with the approach available in open-source code.

Abstract

Recently, diffusion models have gained significant attention as a novel set of deep learning-based generative methods. These models attempt to sample data from a Gaussian distribution that adheres to a target distribution, and have been successfully adapted to the reconstruction of MRI data. However, as an unconditional generative model, the diffusion model typically disrupts image coordination because of the consistent projection of data introduced by conditional bootstrap. This often results in image fragmentation and incoherence. Furthermore, the inherent limitations of the diffusion model often lead to excessive smoothing of the generated images. In the same vein, some deep learning-based models often suffer from poor generalization performance, meaning their effectiveness is greatly affected by different acceleration factors. To address these challenges, we propose a novel diffusion model-based MRI reconstruction method, named TC-DiffRecon, which does not rely on a specific acceleration factor for training. We also suggest the incorporation of the MF-UNet module, designed to enhance the quality of MRI images generated by the model while mitigating the over-smoothing issue to a certain extent. During the image generation sampling process, we employ a novel TCKG module and a Coarse-to-Fine sampling scheme. These additions aim to harmonize image texture, expedite the sampling process, while achieving data consistency. Our source code is available at https://github.com/JustlfC03/TC-DiffRecon.

TC-DiffRecon: Texture coordination MRI reconstruction method based on diffusion model and modified MF-UNet method

TL;DR

TC-DiffRecon tackles diffusion-model–based MRI reconstruction by addressing texture and coherence issues that arise from unconditional generation. It replaces the U-Net backbone with MF-UNet, introducing backbone and skip feature scaling factors and Fourier-domain modulation to preserve texture without excessive smoothing. The TCKG module integrates K-space data consistency in a texture-coordinated fashion, and a Coarse-to-Fine (C2F) sampling scheme accelerates inference while stabilizing texture. Experiments on FastMRI knee data demonstrate improved generalization across varying acceleration factors and superior image quality compared with state-of-the-art methods, with the approach available in open-source code.

Abstract

Recently, diffusion models have gained significant attention as a novel set of deep learning-based generative methods. These models attempt to sample data from a Gaussian distribution that adheres to a target distribution, and have been successfully adapted to the reconstruction of MRI data. However, as an unconditional generative model, the diffusion model typically disrupts image coordination because of the consistent projection of data introduced by conditional bootstrap. This often results in image fragmentation and incoherence. Furthermore, the inherent limitations of the diffusion model often lead to excessive smoothing of the generated images. In the same vein, some deep learning-based models often suffer from poor generalization performance, meaning their effectiveness is greatly affected by different acceleration factors. To address these challenges, we propose a novel diffusion model-based MRI reconstruction method, named TC-DiffRecon, which does not rely on a specific acceleration factor for training. We also suggest the incorporation of the MF-UNet module, designed to enhance the quality of MRI images generated by the model while mitigating the over-smoothing issue to a certain extent. During the image generation sampling process, we employ a novel TCKG module and a Coarse-to-Fine sampling scheme. These additions aim to harmonize image texture, expedite the sampling process, while achieving data consistency. Our source code is available at https://github.com/JustlfC03/TC-DiffRecon.
Paper Structure (15 sections, 4 equations, 2 figures, 2 tables)

This paper contains 15 sections, 4 equations, 2 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Method
  3. Model overview
  4. MF-UNet module
  5. TCKG module
  6. Data consistency strategy
  7. Texture coordination strategy
  8. C2F sampling method
  9. Experiment Result
  10. Dataset and implementation details
  11. Comparisons
  12. Ablation studies
  13. Conclusion
  14. Compliance with Ethical Standards
  15. Acknowledgment

Figures (2)

  • Figure 1: Overall structure of the TC-DiffRecon. It substitutes U-Net in the diffusion model with MF-UNet, which can dynamically adjust the weights of jump connections and backbone connections by two feature coordination factors. Within the TCKG module, the model progressively integrates K-space information into the denoising process in a texture-coordinated manner.
  • Figure 2: C2F Sampling Process. C2F sampling methods produce multiple samples following the T/k steps of denoising. These samples are subsequently averaged over $y_0^{avg}$, followed by iterative refinement through $T_{refine}$ steps.