EUVIMEDEUVIMED entdecken
Crossref

U‐Net With Channel‐Gating AFE and Local–Global Fusion Attention for Colon Polyp Segmentation

Jiajun Liu, Weiwei Gao, Dabin Ren, Yu Fang

International Journal of Imaging Systems and Technology · 2026-11 · 36 (6)

DOI: 10.1002/ima.70470 ↗

Kurz gefasst

Fachartikel in der Zeitschrift International Journal of Imaging Systems and Technology, online erschienen am 10.11.2026. Verfasst von Jiajun Liu und 3 weiteren Autor:innen. Die Arbeit nennt 35 Literaturangaben. Gefördert von: National Natural Science Foundation of China.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Colonoscopy is recognized as the gold standard for colorectal cancer screening. However, polyps in colonoscopy are characterized by complex morphologies, blurred edges, and high similarity to the surrounding mucosa, leading to a relatively high polyp miss rate. Based on this, a U‐Net colon polyp image segmentation model with local–global fusion attention (Window Axial Fusion Attention UNet, WAFA‐Unet) is proposed. This model takes U‐Net as its basic framework. To address the problems of blurred polyp boundaries and large variations in size, WAFA attention is embedded in the encoder to improve feature extraction capability. Subsequently, an adaptive feature enhancement module with channel gating weighted fusion is incorporated into the skip connections to optimize feature propagation. Furthermore, a hybrid loss function combining cross‐entropy, Dice, and Boundary DoU losses is designed to further improve the segmentation accuracy of polyp boundaries. WAFA‐Unet achieves favorable segmentation performance on the CVC‐ClinicDB dataset, with mIoU and mDSC reaching 89.45% and 94.27%, respectively. Compared with U‐Net, the two metrics are improved by 10.51% and 8.95%. Intra‐dataset evaluations are conducted on four datasets including CVC‐ColonDB, Kvasir‐SEG, CVC‐300, and ETIS. The average mIoU and average mDSC across these datasets are 88.64% and 93.68%. Finally, the segmentation capability of WAFA‐Unet is further validated on the self‐built dataset. Experimental results show that WAFA‐Unet accurately segments polyps by enhancing tiny polyp features and achieves stable performance on multiple test datasets.

Förderung

  • National Natural Science Foundation of China

Zitieren

APA

Liu, J., Gao, W., Ren, D., & Fang, Y. (2026). U‐Net With Channel‐Gating AFE and Local–Global Fusion Attention for Colon Polyp Segmentation. International Journal of Imaging Systems and Technology, 36(6). https://doi.org/10.1002/ima.70470

Vancouver

Liu J, Gao W, Ren D, Fang Y. U‐Net With Channel‐Gating AFE and Local–Global Fusion Attention for Colon Polyp Segmentation. International Journal of Imaging Systems and Technology. 2026;36(6). doi:10.1002/ima.70470

BibTeX
@article{Liu2026,
  author = {Liu, Jiajun and Gao, Weiwei and Ren, Dabin and Fang, Yu},
  title = {U‐Net With Channel‐Gating AFE and Local–Global Fusion Attention for Colon Polyp Segmentation},
  journal = {International Journal of Imaging Systems and Technology},
  year = {2026},
  volume = {36},
  number = {6},
  doi = {10.1002/ima.70470}
}
Kostenlos starten →