Attention-Based Two-Branch Hybrid Fusion Network for Medical Image Segmentation

• DOI: 10.3390/app14104073
• Publication Date: 2024-05-14T07:20:29Z
• Authors: Jie Liu; Songren Mao; Liangrui Pan
• Keywords: Merge (version control)

Accurate segmentation of medical images is vital for disease detection and treatment. Convolutional Neural Networks (CNN) Transformer models are widely used in image due to their exceptional capabilities recognition segmentation. However, CNNs often lack an understanding the global context may lose spatial details target, while Transformers struggle with local information processing, leading reduced geometric detail target. To address these issues, this research presents a Global-Local Fusion network model (GLFUnet) based on U-Net framework attention mechanisms. The employs dual-branch that utilizes ConvNeXt Swin simultaneously extract multi-level features from pathological images. It enhances ConvNeXt’s feature extraction up-sampling modules, improving Transformer’s dependency channel attention. Attention Feature module skip connections efficiently merge detailed coarse CNN branches at various scales. fused then progressively restored original resolution pixel-level prediction. Comprehensive experiments datasets stomach liver cancer demonstrate GLFUnet’s superior performance adaptability segmentation, holding promise clinical analysis diagnosis.