Energy-guided polyp segmentation with active contour priors and attention mechanism
IET Conference Proceedings · 01.07.2026 · 2026 (7) · 200-207
DOI: 10.1049/icp.2026.2762 ↗Kurz gefasst
Fachartikel in der Zeitschrift IET Conference Proceedings, erschienen am 01.07.2026. Verfasst von Yang Tang.
Abstract (Originalsprache)
Polyp segmentation is critical for early colorectal cancer screening, yet it remains challenging due to large appearance variations, fuzzy boundaries, and complex noise in colonoscopy images. To address these issues, we propose an energy-guided polyp segmentation framework termed Energy-Guided Active Contour Network (EGAC-Net). It integrates active contour priors with attention-based feature learning in an end-to-end manner. Specifically, a cross-layer feature fusion module (CLFF) is designed to aggregate shallow spatial details and deep semantic representations for enhanced multi-scale interaction. An energy-guided dual-path decoder (EGDPD) is then introduced to progressively refine segmentation by leveraging LPF-derived spatial and channel masks, enabling boundary-aware enhancement and adaptive semantic injection. Furthermore, an ACM-guided active contour loss (ACMLoss) is incorporated to explicitly regularize spatial consistency and boundary smoothness during training. Extensive experiments on five public polyp segmentation benchmarks demonstrate that the proposed method achieves competitive and stable performance on both seen and unseen datasets. It also shows strong robustness under cross-dataset variations and complex clinical conditions.
Zitieren
APA
Tang, Y. (2026). Energy-guided polyp segmentation with active contour priors and attention mechanism. IET Conference Proceedings, 2026(7), 200-207. https://doi.org/10.1049/icp.2026.2762
Vancouver
Tang Y. Energy-guided polyp segmentation with active contour priors and attention mechanism. IET Conference Proceedings. 2026;2026(7):200-207. doi:10.1049/icp.2026.2762
BibTeX
@article{Tang2026,
author = {Tang, Yang},
title = {Energy-guided polyp segmentation with active contour priors and attention mechanism},
journal = {IET Conference Proceedings},
year = {2026},
volume = {2026},
number = {7},
pages = {200--207},
doi = {10.1049/icp.2026.2762}
}
EUVIMED