EUVIMEDEUVIMED entdecken
Crossref

A unified approach for maintaining MRI reconstruction quality and quantifying both aleatoric and epistemic uncertainty

Satoshi Kuroki, Naoto Fujita, Yasuhiko Terada

Magnetic Resonance Imaging · 2026-09 · 131 · 110680

DOI: 10.1016/j.mri.2026.110680 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Magnetic Resonance Imaging, 2026-09. Verfasst von Satoshi Kuroki und 2 weiteren Autor:innen. Beteiligte Einrichtung: Institute of Pure and Applied Sciences, University of Tsukuba, Tsukuba, Japan. Hauptthemen laut MEDLINE: Magnetic Resonance Imaging, Image Processing, Computer-Assisted. Die Arbeit nennt 31 Literaturangaben. Gefördert von: Japan Society for the Promotion of Science, Japan Science and Technology Agency, Japan Science and Technology Corporation.

AbstractSchlagwörterBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

This study investigates practical design choices for Bayesian uncertainty quantification (UQ) in model-based deep learning (MoDL) for accelerated MRI reconstruction. Specifically, we examine where to introduce stochastic layers to capture epistemic uncertainty while maintaining reconstruction quality, and how to integrate heteroscedastic learning to capture aleatoric uncertainty. We compare Monte Carlo dropout (MCD) and Bayes-by-Backprop (BBB) across multiple layer placement patterns and propose a two-stage training strategy in which the reconstruction network is frozen before training the aleatoric uncertainty network. Experiments demonstrate that the placement of stochastic layers significantly impacts performance for MCD, whereas BBB exhibits greater robustness. Furthermore, the proposed two-stage strategy yields statistically significant improvements in peak signal-to-noise ratio compared with conventional joint training. Under representative domain shifts (anatomical and acceleration-factor shifts), the estimated uncertainties capture expected in-domain versus out-of-distribution trends, validating the proposed framework. These results provide implementation-level guidance for realizing effective Bayesian UQ in MoDL-based MRI reconstruction.

Schlagwörter

MeSH-Schlagwörter (MEDLINE)

Magnetic Resonance ImagingImage Processing, Computer-AssistedBayes TheoremSignal-To-Noise RatioUncertaintyDeep LearningAlgorithmsMonte Carlo MethodHumansBrainComputer Simulation

Stichwörter der Autor:innen

Bayesian neural networksDeep learningMRI reconstructionModel-based reconstructionUncertainty quantification

Autor:innen und Institutionen

  • Institute of Pure and Applied Sciences, University of Tsukuba, Tsukuba, Japan.
  • Institute of Pure and Applied Sciences, University of Tsukuba, Tsukuba, Japan. Electronic address: terada.yasuhiko.fu@u.tsukuba.ac.jp.

Förderung

  • Japan Society for the Promotion of Science
  • Japan Science and Technology Agency
  • Japan Science and Technology Corporation
  • Fujifilm Corporation

Zitieren

APA

Kuroki, S., Fujita, N., & Terada, Y. (2026). A unified approach for maintaining MRI reconstruction quality and quantifying both aleatoric and epistemic uncertainty. Magnetic Resonance Imaging, 131, 110680. https://doi.org/10.1016/j.mri.2026.110680

Vancouver

Kuroki S, Fujita N, Terada Y. A unified approach for maintaining MRI reconstruction quality and quantifying both aleatoric and epistemic uncertainty. Magnetic Resonance Imaging. 2026;131:110680. doi:10.1016/j.mri.2026.110680

BibTeX
@article{Kuroki2026,
  author = {Kuroki, Satoshi and Fujita, Naoto and Terada, Yasuhiko},
  title = {A unified approach for maintaining MRI reconstruction quality and quantifying both aleatoric and epistemic uncertainty},
  journal = {Magnetic Resonance Imaging},
  year = {2026},
  volume = {131},
  pages = {110680},
  doi = {10.1016/j.mri.2026.110680}
}
Kostenlos starten →