EUVIMEDEUVIMED entdecken
Crossref

Boundary-Ambiguity-Aware Lesion Evidence Learning for Breast Ultrasound Segmentation and Multimodal Classification

Jinlin Ye, Deming Hu, Yuhan Liu, Shangjie Ren, Changjun Wang, Yidong Zhou, Wei Zhang

IEEE Transactions on Medical Imaging · 2026 · 1-1

DOI: 10.1109/tmi.2026.3752041 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Transactions on Medical Imaging, 2026. Verfasst von Jinlin Ye und 6 weiteren Autor:innen. Gefördert von: National Natural Science Foundation of China, Hebei Provincial Department of Education.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Multimodal analysis of breast ultrasound images and structured clinical data can support lesion characterization across diverse clinical classification tasks. However, existing image and clinicopathological data fusion methods usually fuse imaging and clinical information as feature vectors, without explicitly organizing ultrasound-derived evidence according to anatomical roles such as the lesion core, ambiguous boundary and peritumoral tissue. Therefore, this paper proposes a boundary-ambiguity-aware lesion evidence learning framework (BAA-LEL) for joint breast ultrasound lesion segmentation and multimodal classification. BAA-LEL first extracts a lesion prior with a MedSAM-assisted segmentation branch and refines ambiguous margins through an anchor-constrained boundary graph. The refined lesion prior is organized into core, boundary, and peritumoral regions, while boundary geometry and ambiguity are retained as complementary evidence. A morphology-clinical heterogeneous graph then models interactions among regional morphology, boundary evidence, and structured clinical variables for task-specific classification. Experiments on HER2USC, LNMUSC, BrEaST, BUSI, and ISIC IMA++ demonstrate strong segmentation performance and competitive classification performance across breast ultrasound and cross-domain settings. Controlled ablations, statistical comparisons, calibration analysis, and interpretability experiments further support the contributions of boundary-aware refinement, zonal morphology encoding, and morphology-clinical heterogeneous reasoning. The source code is available at: https://github.com/JinlinYY/BAA-LEL.

Förderung

  • National Natural Science Foundation of China
  • Hebei Provincial Department of Education

Zitieren

APA

Ye, J., Hu, D., Liu, Y., Ren, S., Wang, C., Zhou, Y., & Zhang, W. (2026). Boundary-Ambiguity-Aware Lesion Evidence Learning for Breast Ultrasound Segmentation and Multimodal Classification. IEEE Transactions on Medical Imaging, 1-1. https://doi.org/10.1109/tmi.2026.3752041

Vancouver

Ye J, Hu D, Liu Y, Ren S, Wang C, Zhou Y, et al. Boundary-Ambiguity-Aware Lesion Evidence Learning for Breast Ultrasound Segmentation and Multimodal Classification. IEEE Transactions on Medical Imaging. 2026:1-1. doi:10.1109/tmi.2026.3752041

BibTeX
@article{Ye2026,
  author = {Ye, Jinlin and Hu, Deming and Liu, Yuhan and Ren, Shangjie and Wang, Changjun and Zhou, Yidong and Zhang, Wei},
  title = {Boundary-Ambiguity-Aware Lesion Evidence Learning for Breast Ultrasound Segmentation and Multimodal Classification},
  journal = {IEEE Transactions on Medical Imaging},
  year = {2026},
  pages = {1--1},
  doi = {10.1109/tmi.2026.3752041}
}
Kostenlos starten →