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

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Signal Processing

Background:

• Deep learning models have advanced MRI reconstruction.
• Diffusion models excel in various image generation tasks.
• Cold diffusion models generalize diffusion processes for diverse image transformations.

Purpose of the Study:

• To propose a novel k-space cold diffusion model for accelerated MRI reconstruction.
• To perform image degradation and restoration directly in k-space.
• To evaluate the model's performance against existing deep learning methods.

Main Methods:

• Developed a k-space cold diffusion model for MRI reconstruction.
• Implemented image degradation and restoration in the k-space domain.
• Compared the proposed model with multiple deep learning-based MRI reconstruction techniques.
• Validated the model on a large, open-source MRI dataset.

Main Results:

• The k-space cold diffusion model generates high-quality MRI reconstruction images.
• The novel k-space degradation approach is effective for accelerated MRI.
• The proposed method shows competitive or superior performance compared to other deep learning models.

Conclusions:

• The proposed k-space cold diffusion model offers a promising new direction for accelerated MRI.
• Performing diffusion in k-space without Gaussian noise is a viable strategy for high-fidelity reconstruction.
• This approach has the potential to improve the efficiency and quality of MRI scans.