Scholarlink

<p>In recent years, two-branch networks combining CNNs and transformers have advanced medical image segmentation by leveraging CNNs for local feature extraction and transformers for global context modeling. However, challenges remain in feature fusion and computational efficiency. In this paper, we propose a dual-branch network with enhanced cross-fusion and spatial-channel attention. The dual-stream cross-fusion (DCF) module facilitates interaction between CNN and transformer features, while the cross-scale feature fusion (CS) module in the decoder enhances contextual understanding. To further optimize performance, an attention enhancement (SCAM) module is introduced between the DCF output and the CS input to enable deeper fusion of local and global information. The DCF module is also augmented with multi-layer factorial attention for multi-level feature interaction and fusion. Additionally, a dynamic loss mechanism with adaptive weight adjustment improves training stability and generalization in the presence of class imbalance. Experiments demonstrate a significant improvement in segmentation accuracy, underscoring the effectiveness of multi-level fusion and attention enhancement in medical image analysis.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Dcsca-Net: a dual-branch network with enhanced cross-fusion and spatial-channel attention for precise medical image segmentation

  • 期刊论文

  • 发表日期：

    2025年4月12日

查看全文
  • Nianhao Wang,
  • Han Wang

摘要

In recent years, two-branch networks combining CNNs and transformers have advanced medical image segmentation by leveraging CNNs for local feature extraction and transformers for global context modeling. However, challenges remain in feature fusion and computational efficiency. In this paper, we propose a dual-branch network with enhanced cross-fusion and spatial-channel attention. The dual-stream cross-fusion (DCF) module facilitates interaction between CNN and transformer features, while the cross-scale feature fusion (CS) module in the decoder enhances contextual understanding. To further optimize performance, an attention enhancement (SCAM) module is introduced between the DCF output and the CS input to enable deeper fusion of local and global information. The DCF module is also augmented with multi-layer factorial attention for multi-level feature interaction and fusion. Additionally, a dynamic loss mechanism with adaptive weight adjustment improves training stability and generalization in the presence of class imbalance. Experiments demonstrate a significant improvement in segmentation accuracy, underscoring the effectiveness of multi-level fusion and attention enhancement in medical image analysis.