EUVIMEDEUVIMED entdecken
Open Access · CC BY-NC-ND 4.0

S2UX‐Net: Enhancing 3D Intracranial Aneurysm Segmentation via Advanced Multi‐Scale Context and Adaptive Feature Modulation

Tao Wu, Mingquan Ye, Shuwen Yang, Yunfeng Zhou

IET Computer Vision · 2026-01 · 20 (1)

DOI: 10.1049/cvi2.70076 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IET Computer Vision, online erschienen am 24.07.2026. Verfasst von Tao Wu und 3 weiteren Autor:innen. Die Arbeit nennt 30 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Accurate segmentation of Intracranial Aneurysms (IAs) from 3D Computed Tomography Angiography (CTA) is critical for diagnosis and treatment planning. Although existing convolutional networks and their hybrid variants with Transformers have shown promise, they often struggle with the complex, multi‐scale spatial characteristics and high morphological variability of IAs. This paper introduces S2UX‐Net, a novel 3D segmentation network that augments the 3D UX‐Net backbone by integrating two strategically enhanced core modules. First, we propose an enhanced Spatial Pyramid Squeeze Attention module (SPSA3D), which incorporates triple‐scale convolutional kernels (1 1 1, 3 3 3, 5 5 5) and a sequential dual‐attention mechanism (channel followed by spatial) for superior feature calibration and context aggregation. Second, we design a dynamic Spatial Adaptive Feature Modulation module (SAFM3D), which utilises multi‐branch dilated convolutions with configurable rates and a learnable adaptive fusion mechanism to capture rich contextual information and perform spatially variant feature refinement. Furthermore, we explore two paradigms for synergistically integrating these modules: a sequential connection (SPSA3D SAFM3D) and a parallel connection with learnable fusion. Extensive experiments on a clinical dataset of 500 IA patients demonstrate that S2UX‐Net achieves competitive performance, with a Dice score of 84.3%, an IoU of 74.4%, a Precision of 80.2% and a Recall of 91.0%, outperforming several strong baseline methods. Our work highlights that substantial improvements in 3D intracranial aneurysm segmentation can be achieved through principled enhancement and synergistic integration of advanced feature modelling components.

Zitieren

APA

Wu, T., Ye, M., Yang, S., & Zhou, Y. (2026). S2UX‐Net: Enhancing 3D Intracranial Aneurysm Segmentation via Advanced Multi‐Scale Context and Adaptive Feature Modulation. IET Computer Vision, 20(1). https://doi.org/10.1049/cvi2.70076

Vancouver

Wu T, Ye M, Yang S, Zhou Y. S2UX‐Net: Enhancing 3D Intracranial Aneurysm Segmentation via Advanced Multi‐Scale Context and Adaptive Feature Modulation. IET Computer Vision. 2026;20(1). doi:10.1049/cvi2.70076

BibTeX
@article{Wu2026,
  author = {Wu, Tao and Ye, Mingquan and Yang, Shuwen and Zhou, Yunfeng},
  title = {S2UX‐Net: Enhancing 3D Intracranial Aneurysm Segmentation via Advanced Multi‐Scale Context and Adaptive Feature Modulation},
  journal = {IET Computer Vision},
  year = {2026},
  volume = {20},
  number = {1},
  doi = {10.1049/cvi2.70076}
}
Kostenlos starten →