Abstract:Deep learning has shown the capability to substantially accelerate MRI reconstruction while acquiring fewer measurements. Recently, diffusion models have gained burgeoning interests as a novel group of deep learning-based generative methods. These methods seek to sample data points that belong to a target distribution from a Gaussian distribution, which has been successfully extended to MRI reconstruction. In this work, we proposed a Cold Diffusion-based MRI reconstruction method called CDiffMR. Different from conventional diffusion models, the degradation operation of our CDiffMR is based on \textit{k}-space undersampling instead of adding Gaussian noise, and the restoration network is trained to harness a de-aliaseing function. We also design starting point and data consistency conditioning strategies to guide and accelerate the reverse process. More intriguingly, the pre-trained CDiffMR model can be reused for reconstruction tasks with different undersampling rates. We demonstrated, through extensive numerical and visual experiments, that the proposed CDiffMR can achieve comparable or even superior reconstruction results than state-of-the-art models. Compared to the diffusion model-based counterpart, CDiffMR reaches readily competing results using only $1.6 \sim 3.4\%$ for inference time. The code is publicly available at <a class="link-external link-https" href="https://github.com/ayanglab/CDiffMR" rel="external noopener nofollow">this https URL</a>.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is whether K - space undersampling can be used to replace Gaussian noise in Fast MRI to accelerate the MRI reconstruction process. Specifically, the paper proposes an MRI reconstruction method based on the Cold Diffusion Model, named CDiffMR. Different from traditional diffusion models, CDiffMR uses K - space undersampling instead of adding Gaussian noise in the degradation operation and trains a de - aliasing function as a restoration network. In addition, the paper also designs the Starting Point Conditioning (SPC) and Data Consistency Conditioning (DCC) strategies to guide and accelerate the reverse process. Through extensive numerical and visual experiments, the paper shows that CDiffMR can achieve performance comparable to or even better than the existing state - of - the - art models in reconstruction results while significantly reducing the inference time. ### Main Contributions 1. **Innovative Cold - Diffusion MRI Reconstruction Method**: CDiffMR is the first diffusion model that utilizes K - space undersampling degradation for MRI reconstruction. 2. **Two New K - Space Conditioning Strategies**: SPC and DCC, which are used to guide and accelerate the reverse process. 3. **Generality of Pretrained Models**: Once CDiffMR is trained, it can be used for reconstruction tasks with different undersampling rates without retraining. ### Methods 1. **Model Components and Training** - **Degradation Operation**: Use the K - space undersampling degradation module (KSUD), that is $ D(x, t)=F^{-1}M_tFx $, where $ M_t $ is a mask of time step $ t $ that controls the undersampling rate. - **Restoration Operation**: Use an improved U - Net structure with a time - embedding module as a restoration network $ R_\theta(x, t) $. 2. **K - Space - Conditioned Reverse Process** - **Starting Point Conditioning (SPC)**: Select the starting point $ T' $ of the reverse process according to the undersampling rate, so that the reverse process starts from an intermediate step instead of the last step. - **Data Consistency Conditioning (DCC)**: Apply data consistency in each step of the reverse process to further guide the reverse process. ### Experimental Results - **Quantitative Evaluation**: Use metrics such as PSNR, SSIM, and LPIPS to evaluate the reconstruction quality. The results show that CDiffMR achieves performance comparable to or better than the existing state - of - the - art methods when the acceleration factors are 8 and 16. - **Qualitative Evaluation**: Through visual comparison, the advantages of CDiffMR in detail preservation and overall image quality are shown. - **Inference Time**: The inference time of CDiffMR is only 1.6% to 3.4% of that of the existing diffusion model methods, which significantly improves the efficiency. ### Discussion and Conclusion The paper successfully solves the key problems in fast MRI reconstruction by introducing K - space undersampling as a degradation operation in the diffusion model. CDiffMR not only performs excellently in reconstruction quality but also greatly shortens the inference time. This provides new ideas for solving other inverse problems and can be further extended to other medical imaging fields in the future.

CDiffMR: Can We Replace the Gaussian Noise with K-Space Undersampling for Fast MRI?

K-space Cold Diffusion: Learning to Reconstruct Accelerated MRI without Noise

Learning to reconstruct accelerated MRI through K-space cold diffusion without noise

Deep Learning for Under-Sampled Non-Cartesian ASL MRI Reconstruction

High-Frequency Space Diffusion Models for Accelerated MRI

High-Frequency Space Diffusion Model for Accelerated MRI

DiffCMR: Fast Cardiac MRI Reconstruction with Diffusion Probabilistic Models

Autoregressive Image Diffusion: Generation of Image Sequence and Application in MRI

Towards performant and reliable undersampled MR reconstruction via diffusion model sampling

Sub-DM:Subspace Diffusion Model with Orthogonal Decomposition for MRI Reconstruction

Self-Supervised Adversarial Diffusion Models for Fast MRI Reconstruction

Model-Based Deep Learning for Reconstruction of Joint k-q Under-sampled High Resolution Diffusion MRI

Sampling-Pattern-Agnostic MRI Reconstruction through Adaptive Consistency Enforcement with Diffusion Model

Fast Controllable Diffusion Models for Undersampled MRI Reconstruction

Zero-shot Dynamic MRI Reconstruction with Global-to-local Diffusion Model

Self-Supervised MRI Reconstruction with Unrolled Diffusion Models

Highly Accelerated MRI via Implicit Neural Representation Guided Posterior Sampling of Diffusion Models

Deep learning for undersampled MRI reconstruction

Using Diffusion Model to Generate High-Resolution MRI

Physics-Informed DeepMRI: k-Space Interpolation Meets Heat Diffusion

Correlated and Multi-frequency Diffusion Modeling for Highly Under-sampled MRI Reconstruction