Analyzing and adapting diffusion segmentation behavior for medical images
Biomedical Signal Processing and Control · 2026-02 · 112 · 108619
DOI: 10.1016/j.bspc.2025.108619 ↗Kurz gefasst
Fachartikel in der Zeitschrift Biomedical Signal Processing and Control, 2026-02. Verfasst von Mathias Öttl und 11 weiteren Autor:innen. Die Arbeit wurde bisher 1-mal zitiert (Crossref) und nennt 31 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Deutsche Forschungsgemeinschaft, Interdisziplinäres Zentrum für Klinische Forschung, Universitätsklinikum Würzburg.
Abstract (Originalsprache)
Denoising diffusion probabilistic models have become increasingly popular due to their ability for probabilistic modeling and the generation of diverse outputs. This versatility inspired their adaptation for image segmentation, where multiple predictions of the model can produce segmentation results that not only achieve high quality but also capture the uncertainty inherent in the model, which makes them especially interesting for medical data. Powerful architectures were proposed to improve the performance of diffusion segmentation. However, detailed analyses and discussions of the differences between diffusion segmentation and image generation are missing. Furthermore, thorough evaluations are lacking that distinguish the improvements of the utilized architectures for segmentation in general from their benefit for diffusion segmentation specifically. In this work, we critically analyze and discuss how diffusion segmentation differs from diffusion image generation, with a particular focus on training behavior. We assess how the proposed diffusion segmentation architectures perform when trained directly for segmentation and explore how different segmentation tasks influence the behavior of diffusion segmentation. Drawing from our analysis, we suggest a method for modifying the diffusion segmentation procedure for image segmentation across various medical datasets. We showcase the improvements achieved by implementing our suggested modifications across three different diffusion segmentation methods applied to three separate datasets. Our goal is to offer a comprehensive understanding of the dynamics of diffusion segmentation, which will enable improved designs and assessments of diffusion segmentation techniques moving forward.
Förderung
- Deutsche Forschungsgemeinschaft
- Interdisziplinäres Zentrum für Klinische Forschung, Universitätsklinikum Würzburg
Zitieren
APA
Öttl, M., Mei, S., Wilm, F., Qiu, J., Steenpass, J., Rübner, M., Hartmann, A., Hübner, H., Fasching, P., Maier, A., Erber, R., & Breininger, K. (2026). Analyzing and adapting diffusion segmentation behavior for medical images. Biomedical Signal Processing and Control, 112, 108619. https://doi.org/10.1016/j.bspc.2025.108619
Vancouver
Öttl M, Mei S, Wilm F, Qiu J, Steenpass J, Rübner M, et al. Analyzing and adapting diffusion segmentation behavior for medical images. Biomedical Signal Processing and Control. 2026;112:108619. doi:10.1016/j.bspc.2025.108619
BibTeX
@article{ttl2026,
author = {Öttl, Mathias and Mei, Siyuan and Wilm, Frauke and Qiu, Jingna and Steenpass, Jana and Rübner, Matthias and Hartmann, Arndt and Hübner, Hanna and Fasching, Peter and Maier, Andreas and Erber, Ramona and Breininger, Katharina},
title = {Analyzing and adapting diffusion segmentation behavior for medical images},
journal = {Biomedical Signal Processing and Control},
year = {2026},
volume = {112},
pages = {108619},
doi = {10.1016/j.bspc.2025.108619}
}
EUVIMED