EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Fast and Robust Diffusion Posterior Sampling for MR Image Reconstruction Using the Preconditioned Unadjusted Langevin Algorithm

Moritz Blumenthal iD, Tina Holliber iD, Jonathan I Tamir, Martin Uecker iD

Magnetic Resonance in Medicine · 2026-09 · 96 (3) · 1323-1332

DOI: 10.1002/mrm.70416 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Magnetic Resonance in Medicine, online erschienen am 10.05.2026. Verfasst von Moritz Blumenthal und 3 weiteren Autor:innen. Beteiligte Einrichtung: Institute of Biomedical Imaging, Graz University of Technology, Graz, Austria. Hauptthemen laut MEDLINE: Algorithms, Brain, Image Processing, Computer-Assisted, Diffusion Magnetic Resonance Imaging. Die Arbeit nennt 35 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: National Science Foundation, Austrian Science Fund.

AbstractSchlagwörterBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Purpose

The Unadjusted Langevin Algorithm ( ULA ) in combination with diffusion models can generate high quality MRI reconstructions with uncertainty estimation from highly undersampled k‐space data. However, sampling methods such as diffusion posterior sampling ( DPS ) or likelihood annealing suffer from long reconstruction times and the need for parameter tuning. The purpose of this work is to develop a robust sampling algorithm with fast convergence. Theory and Methods In the reverse diffusion process used for sampling the posterior, the exact likelihood is multiplied with the diffused prior at all noise scales. To overcome the issue of slow convergence, preconditioning is used. The method is trained on fastMRI data and tested on retrospectively undersampled brain data of a healthy volunteer.

Results

For posterior sampling in Cartesian and non‐Cartesian accelerated MRI the new approach outperforms annealed sampling and DPS in terms of reconstruction speed and sample quality.

Conclusion

The proposed exact likelihood with preconditioning enables rapid and reliable posterior sampling across various MRI reconstruction tasks without the need for parameter tuning.

Schlagwörter

MeSH-Schlagwörter (MEDLINE)

AlgorithmsBrainImage Processing, Computer-AssistedDiffusion Magnetic Resonance ImagingMagnetic Resonance ImagingHumansBayes TheoremReproducibility of ResultsImage Enhancement

Stichwörter der Autor:innen

Bayesian reconstructionMRIdiffusion posterior samplingimage reconstructionparallel imaging

Autor:innen und Institutionen

  • Institute of Biomedical Imaging, Graz University of Technology, Graz, Austria.
  • Department of Radiology, Boston Children's Hospital, Harvard Medical School, Boston, USA.
  • Chandra Family Department of Electrical Engineering, University of Texas at Austin, Austin, USA.
  • Department of Diagnostic Medicine, Dell Medical School, University of Texas at Austin, Austin, USA.
  • BioTechMed-Graz, Graz, Austria.

ORCID-Profile: Moritz Blumenthal, Tina Holliber, Martin Uecker

Förderung

  • National Science Foundation
  • Austrian Science Fund

Zitieren

APA

Blumenthal, M., Holliber, T., Tamir, J. I., & Uecker, M. (2026). Fast and Robust Diffusion Posterior Sampling for MR Image Reconstruction Using the Preconditioned Unadjusted Langevin Algorithm. Magnetic Resonance in Medicine, 96(3), 1323-1332. https://doi.org/10.1002/mrm.70416

Vancouver

Blumenthal M, Holliber T, Tamir JI, Uecker M. Fast and Robust Diffusion Posterior Sampling for MR Image Reconstruction Using the Preconditioned Unadjusted Langevin Algorithm. Magnetic Resonance in Medicine. 2026;96(3):1323-1332. doi:10.1002/mrm.70416

BibTeX
@article{Blumenthal2026,
  author = {Blumenthal, Moritz and Holliber, Tina and Tamir, Jonathan I and Uecker, Martin},
  title = {Fast and Robust Diffusion Posterior Sampling for MR Image Reconstruction Using the Preconditioned Unadjusted Langevin Algorithm},
  journal = {Magnetic Resonance in Medicine},
  year = {2026},
  volume = {96},
  number = {3},
  pages = {1323--1332},
  doi = {10.1002/mrm.70416}
}
Kostenlos starten →