Fast unconditional diffusion model for accelerated MRI reconstruction.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computational Science

Background:

• Accelerated magnetic resonance imaging (MRI) reconstruction from undersampled k-space data is a critical challenge in medical imaging.
• Diffusion models show promise for MRI reconstruction but suffer from computationally expensive inference.
• Existing diffusion models require thousands of steps, leading to reconstruction times of tens of minutes.

Purpose of the Study:

• To develop a novel fast diffusion model for MRI reconstruction (FDMR) to accelerate inference and enhance reconstruction quality.
• To overcome the computational limitations of current diffusion models for MRI applications.
• To enable rapid and high-fidelity MRI image reconstruction.

Main Methods:

• Proposed a novel fast diffusion model for MRI reconstruction (FDMR).
• Employed adversarial training of a denoising diffusion Generative Adversarial Network (GAN) to learn diffusion priors.
• Introduced a three-stage inference framework: fast diffusion generation, early stopped deep generative prior adaptation, and diffusion refinement.

Main Results:

• FDMR achieves superior reconstruction accuracy compared to state-of-the-art diffusion methods.
• The proposed method operates 4-10 times faster than existing approaches.
• FDMR enables MRI reconstruction in as little as 8 seconds.

Conclusions:

• FDMR offers a significant advancement in accelerated MRI reconstruction by drastically reducing inference time.
• The model maintains high reconstruction quality, making it a viable solution for clinical applications.
• This work paves the way for faster and more efficient MRI scans.