English

Product of Gaussian Mixture Diffusion Model for non-linear MRI Inversion

Image and Video Processing 2025-01-16 v1 Computer Vision and Pattern Recognition Machine Learning

Abstract

Diffusion models have recently shown remarkable results in magnetic resonance imaging reconstruction. However, the employed networks typically are black-box estimators of the (smoothed) prior score with tens of millions of parameters, restricting interpretability and increasing reconstruction time. Furthermore, parallel imaging reconstruction algorithms either rely on off-line coil sensitivity estimation, which is prone to misalignment and restricting sampling trajectories, or perform per-coil reconstruction, making the computational cost proportional to the number of coils. To overcome this, we jointly reconstruct the image and the coil sensitivities using the lightweight, parameter-efficient, and interpretable product of Gaussian mixture diffusion model as an image prior and a classical smoothness priors on the coil sensitivities. The proposed method delivers promising results while allowing for fast inference and demonstrating robustness to contrast out-of-distribution data and sampling trajectories, comparable to classical variational penalties such as total variation. Finally, the probabilistic formulation allows the calculation of the posterior expectation and pixel-wise variance.

Keywords

Cite

@article{arxiv.2501.08662,
  title  = {Product of Gaussian Mixture Diffusion Model for non-linear MRI Inversion},
  author = {Laurenz Nagler and Martin Zach and Thomas Pock},
  journal= {arXiv preprint arXiv:2501.08662},
  year   = {2025}
}
R2 v1 2026-06-28T21:06:55.993Z