SECT-Net: hybrid dual-decoder network with SE-convolution transformer for liver tumor segmentation
Frontiers in Physiology · 11.03.2026 · 17
DOI: 10.3389/fphys.2026.1753152 ↗Kurz gefasst
Fachartikel in der Zeitschrift Frontiers in Physiology, online erschienen am 11.03.2026. Verfasst von Qiquan Zeng und 3 weiteren Autor:innen. Beteiligte Einrichtung: College of Mechanical Engineering, Quzhou University, Quzhou, China. Die Arbeit nennt 54 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Introduction
Early accurate diagnosis of liver tumors plays a pivotal role in improving patient prognosis and guiding effective treatment planning. However, the automated segmentation of liver tumors remains a highly challenging task due to several intrinsic factors, including heterogeneous intensity distribution, blurred or indistinct boundaries, irregular tumor shapes, and wide variations in size and appearance across patients. To overcome these limitations, we propose a hybrid dual-decoder network that integrates squeeze-and-excitation convolution (SE-convolution) and Transformer-based attention mechanism for liver tumor segmentation.
Methods
Specifically, SECT-Net adopts the classical encoder-decoder architecture as its foundation and introduces a dual-decoder mask mechanism to enhance feature discrimination during segmentation. To enhance the encoder’s capability in capturing both global contextual dependencies and fine-grained local features, the SE-convolution Transformer module (SECTM) is integrated into the second, third, and fourth layers of the encoder. Furthermore, a deep feature capture module (DFCM) is embedded at the bottleneck layer to enhance the network’s ability to extract and preserve high-level semantic representations. After that, the extracted deep features are seamlessly integrated through skip connections with the decoder. Results and Discussion To comprehensively assess the effectiveness and generalization capability of SECT-Net, extensive experiments were conducted on the liver tumor datasets collected from Quzhou People’s Hospital. On the arterial phase dataset, SECT-Net demonstrated excellent segmentation performance, achieving Dice of 0.8452, Mcc of 0.8411, and Jaccard of 0.7339. Similarly, on the portal venous phase dataset, SECT-Net maintained robust generalization, with Dice of 0.8425, Mcc of 0.8396, and Jaccard of 0.7339. Furthermore, on the public 3DIRCADb dataset, SECT-Net also achieved competitive performance, with Dice, Mcc, and Jaccard scores of 0.8845, 0.8855 and 0.7969. These consistent results across both private and public datasets further demonstrate the strong reliability, robustness, and generalization capability of SECT-Net in segmenting liver tumors with diverse intensity distributions and morphological characteristics.
Schlagwörter
Stichwörter der Autor:innen
Autor:innen und Institutionen
- College of Mechanical Engineering, Quzhou University, Quzhou, China.
- College of Electrical and Information Engineering, Quzhou University, Quzhou, China.
- Department of Infectious Diseases, The Quzhou Affiliated Hospital of Wenzhou Medical University, Quzhou People's Hospital, Quzhou, China.
Zitieren
APA
Zeng, Q., Ye, D., Chen, M., & Jiang, X. (2026). SECT-Net: hybrid dual-decoder network with SE-convolution transformer for liver tumor segmentation. Frontiers in Physiology, 17. https://doi.org/10.3389/fphys.2026.1753152
Vancouver
Zeng Q, Ye D, Chen M, Jiang X. SECT-Net: hybrid dual-decoder network with SE-convolution transformer for liver tumor segmentation. Frontiers in Physiology. 2026;17. doi:10.3389/fphys.2026.1753152
BibTeX
@article{Zeng2026,
author = {Zeng, Qiquan and Ye, Dongfen and Chen, Meiqin and Jiang, Xiaoliang},
title = {SECT-Net: hybrid dual-decoder network with SE-convolution transformer for liver tumor segmentation},
journal = {Frontiers in Physiology},
year = {2026},
volume = {17},
doi = {10.3389/fphys.2026.1753152}
}
EUVIMED