EUVIMEDEUVIMED entdecken
Crossref

Stability-guaranteed learning for partially supervised cardiac image segmentation via decomposition and conditional propagation

Ke Zhang, Hangqi Zhou, Bomin Wang, Xiahai Zhuang

Medical Image Analysis · 2027-01 · 115 · 104342

DOI: 10.1016/j.media.2026.104342 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Medical Image Analysis, 2027-01. Verfasst von Ke Zhang und 3 weiteren Autor:innen. Beteiligte Einrichtung: School of Data Science, Fudan University, Shanghai, 200433, China; Department of Electrical and Computer Engineering, Johns Hopkins University, Balti…. Hauptthemen laut MEDLINE: Supervised Machine Learning, Image Interpretation, Computer-Assisted, Image Processing, Computer-Assisted. Die Arbeit nennt 45 Literaturangaben. Gefördert von: Shanghai Municipal Education Commission, National Natural Science Foundation of China.

AbstractSchlagwörterBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Partially-supervised learning can be challenging for image segmentation due to the lack of supervision for unlabeled structures. From the perspective of optimization, directly applying conventional loss functions in fully-supervised learning could lead to the unstable training problem in partially-supervised segmentation, since the ground truth is not in the solution set of the optimization problem. To address this challenge, we propose a stability-guaranteed learning (SGL) framework, which trains a single multi-label segmentation network using images with only partial annotations. We first give the definition of stability and propose a stable loss formulation for partially-supervised segmentation. Specifically, we decompose the stable loss into two basic components, i.e., the positive loss and the negative loss. The former targets pixels with the positive ground truth, and the latter is for pixels with the negative ground truth. Then, we propose a prior loss based on the negative loss, which captures inter-class information from inclusiveness and exclusiveness relationships among images via label propagation. Additionally, we extend the proposed framework via a dual learning strategy, to provide substantial supervision for unlabeled structures. We show that this framework is applicable to different loss functions used in fully-supervised and partially-supervised segmentation. Results on three cardiac segmentation datasets (ACDC, MSCMRseg, and MMWHS) demonstrate that our SGL outperformed state-of-the-art partially-supervised segmentation methods, and could achieve performance matching fully-supervised models.

Schlagwörter

MeSH-Schlagwörter (MEDLINE)

Supervised Machine LearningImage Interpretation, Computer-AssistedImage Processing, Computer-AssistedAlgorithmsHumans

Stichwörter der Autor:innen

Medical image analysisMedical image segmentationPartially-supervised learningStable loss

Autor:innen und Institutionen

  • School of Data Science, Fudan University, Shanghai, 200433, China; Department of Electrical and Computer Engineering, Johns Hopkins University, Baltimore, MD 21211, USA.
  • School of Data Science, Fudan University, Shanghai, 200433, China.
  • School of Data Science, Fudan University, Shanghai, 200433, China. Electronic address: zxh@fudan.edu.cn.

Förderung

  • Shanghai Municipal Education Commission
  • National Natural Science Foundation of China

Zitieren

APA

Zhang, K., Zhou, H., Wang, B., & Zhuang, X. (2027). Stability-guaranteed learning for partially supervised cardiac image segmentation via decomposition and conditional propagation. Medical Image Analysis, 115, 104342. https://doi.org/10.1016/j.media.2026.104342

Vancouver

Zhang K, Zhou H, Wang B, Zhuang X. Stability-guaranteed learning for partially supervised cardiac image segmentation via decomposition and conditional propagation. Medical Image Analysis. 2027;115:104342. doi:10.1016/j.media.2026.104342

BibTeX
@article{Zhang2027,
  author = {Zhang, Ke and Zhou, Hangqi and Wang, Bomin and Zhuang, Xiahai},
  title = {Stability-guaranteed learning for partially supervised cardiac image segmentation via decomposition and conditional propagation},
  journal = {Medical Image Analysis},
  year = {2027},
  volume = {115},
  pages = {104342},
  doi = {10.1016/j.media.2026.104342}
}
Kostenlos starten →