SPIRiT-Diffusion: Self-Consistency Driven Diffusion Model for Accelerated MRI

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computational Science

Background:

• Diffusion models excel in image generation and magnetic resonance imaging (MRI) reconstruction.
• Current diffusion model-based MRI reconstruction methods operate in the image domain, limiting quality due to coil sensitivity map inaccuracies.
• k-space interpolation offers a solution but is incompatible with conventional diffusion models.

Purpose of the Study:

• To develop a novel diffusion model for k-space interpolation in MRI reconstruction.
• To address the limitations of image-domain diffusion models by incorporating k-space physics.
• To improve MRI reconstruction quality and robustness against coil sensitivity map errors.

Main Methods:

• Introduced SPIRiT-Diffusion, a diffusion model for k-space interpolation inspired by the SPIRiT method.
• Formulated a novel stochastic differential equation (SDE) using the SPIRiT iterative solver's k-space physical prior.
• Executed the diffusion process in k-space for data interpolation, termed model-driven diffusion.

Main Results:

• SPIRiT-Diffusion demonstrated superior performance compared to image-domain reconstruction methods.
• Achieved high-quality reconstructions at a significant acceleration factor of 10.
• Validated on a 3D joint intracranial and carotid vessel wall imaging dataset.

Conclusions:

• SPIRiT-Diffusion effectively performs k-space interpolation for MRI reconstruction.
• The model-driven diffusion approach aligns diffusion processes with physical priors, enhancing reconstruction.
• This method offers a promising direction for accelerated and robust MRI acquisition.