EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Reconstructing and analyzing the invariances of low‐dose CT image denoising networks

Elias Eulig, Fabian Jäger, Joscha Maier, Björn Ommer, Marc Kachelrieß

Medical Physics · 2025

DOI: 10.1002/mp.17413 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Medical Physics, 2025. Verfasst von Elias Eulig und 4 weiteren Autor:innen. Beteiligte Einrichtung: Division of X-Ray Imaging and Computed Tomography, German Cancer Research Center (DKFZ), Heidelberg, Germany. Hauptthemen laut MEDLINE: Tomography, X-Ray Computed, Image Processing, Computer-Assisted, Radiation Dosage, Deep Learning. Die Arbeit wurde bisher 6-mal zitiert (Crossref) und nennt 52 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Helmholtz International Graduate School for Cancer Research.

AbstractSchlagwörterBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Background

Deep learning‐based methods led to significant advancements in many areas of medical imaging, most of which are concerned with the reduction of artifacts caused by motion, scatter, or noise. However, with most neural networks being black boxes, they remain notoriously difficult to interpret, hindering their clinical implementation. In particular, it has been shown that networks exhibit invariances w.r.t. input features, that is, they learn to ignore certain information in the input data.

Purpose

To improve the interpretability of deep learning‐based low‐dose CT image denoising networks.

Methods

We learn a complete data representation of low‐dose input images using a conditional variational autoencoder (cVAE). In this representation, invariances of any given denoising network are then disentangled from the information it is not invariant to using a conditional invertible neural network (cINN). At test time, image‐space invariances are generated by applying the inverse of the cINN and subsequent decoding using the cVAE. We propose two methods to analyze sampled invariances and to find those that correspond to alterations of anatomical structures.

Results

The proposed method is applied to four popular deep learning‐based low‐dose CT image denoising networks. We find that the networks are not only invariant to noise amplitude and realizations, but also to anatomical structures.

Conclusions

The proposed method is capable of reconstructing and analyzing invariances of deep learning‐based low‐dose CT image denoising networks. This is an important step toward interpreting deep learning‐based methods for medical imaging, which is essential for their clinical implementation.

Schlagwörter

MeSH-Schlagwörter (MEDLINE)

Tomography, X-Ray ComputedImage Processing, Computer-AssistedRadiation DosageDeep LearningSignal-To-Noise RatioHumansNeural Networks, Computer

Stichwörter der Autor:innen

computed tomographydeep learningexplainabilityinvarianceslow‐doserobustness

Autor:innen und Institutionen

  • Division of X-Ray Imaging and Computed Tomography, German Cancer Research Center (DKFZ), Heidelberg, Germany.
  • Faculty of Physics and Astronomy, Heidelberg University, Heidelberg, Germany.
  • CompVis @ LMU Munich and MCML, Munich, Germany.
  • Medical Faculty Heidelberg, Heidelberg University, Heidelberg, Germany.

Förderung

  • Helmholtz International Graduate School for Cancer Research

Zitieren

APA

Eulig, E., Jäger, F., Maier, J., Ommer, B., & Kachelrieß, M. (2025). Reconstructing and analyzing the invariances of low‐dose CT image denoising networks. Medical Physics. https://doi.org/10.1002/mp.17413

Vancouver

Eulig E, Jäger F, Maier J, Ommer B, Kachelrieß M. Reconstructing and analyzing the invariances of low‐dose CT image denoising networks. Medical Physics. 2025. doi:10.1002/mp.17413

BibTeX
@article{Eulig2025,
  author = {Eulig, Elias and Jäger, Fabian and Maier, Joscha and Ommer, Björn and Kachelrieß, Marc},
  title = {Reconstructing and analyzing the invariances of low‐dose CT image denoising networks},
  journal = {Medical Physics},
  year = {2025},
  doi = {10.1002/mp.17413}
}
Kostenlos starten →