A hybrid CNN-transformer network with global-local context feature aggregation for lung medical image segmentation
Year: 2026
Type: article
Abstract: Accurate lung segmentation is a key prerequisite for computer-aided diagnosis of pulmonary diseases on chest CT and X-ray images. Lung segmentation, however, remains highly challenging due to the complex anatomy of the lung, ambiguous lesion boundaries, low tissue contrast, and the marked appearance gap across imaging modalities. In this work, we propose HGCANet, a hybrid CNN-Transformer network for lung segmentation based on global-local context feature aggregation. The encoder of HGCANet combines CNN and Transformer layer to extract local and global features and incorporates three proposed key modules: (1) A Cross-branch Global-Local Feature Aggregation block consists of three parallel transformation branches to effectively fuse global and local features; (2) A Deep Multi-Scale Attention block, which captures multi-scale contextual information through multi-branch dilated convolutions and jointly applies spatial and channel attention to enhance the response of salient regions; (3) A Global Context Collaborative Attention block, which integrates multi-scale perception with global channel-spatial attention to realize adaptive deep fusion between encoder and decoder features. Comparative experiments conducted on two public and two in-house lung-image datasets demonstrate that HGCANet achieves accurate and efficient lung-region segmentation, showing potential for clinical lung-segmentation applications.
Source: Scientific Reports
Authors: Zhaoyun Sun, Mingying Dai, Hanqun Liu, Junyan Chen
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $2,690
Funder: Qingdao University
OpenAlex ID: W7219750653
FWCI:
Cites:
Cited by:
Related to: