Scholarlink

<p>Breast cancer classification using mammography images is a challenging task due to intricate spatial relationships and multi-scale pixel structures. This study proposes AMFFR-Net, a new deep learning framework that integrates Adaptive Multi-Scale Feature Fusion (AMFF) for multi-scale feature extraction, Graph Attention Networks (GATs) for spatial relationship modeling, and Residual Learning to enhance feature propagation. To progress robustness, the model undergoes extensive preprocessing, including data augmentation, contrast enhancement, and normalization. On the CBIS-DDSM dataset, the proposed AMFFR-Net outperforms other models, including ResNet50, ViT, MIL, and U-Net. Ablation studies establish the important contributions of AMFF, GAT, and Residual Learning in developing classification accuracy and reliability. Based on experimental results, AMFFR-Net displays strong potential as a computer-aided diagnosis technique for breast cancer, representing high effectiveness in differentiating among benign and malignant tumors.</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

AMFFR-Net: Adaptive Multi-scale Feature Fusion and Graph Attention Networks for Breast Cancer Classification in Mammography Images

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年2月11日

查看全文
  • Kudithikunta Anusha,
  • K. Reddy Madhavi

摘要

Breast cancer classification using mammography images is a challenging task due to intricate spatial relationships and multi-scale pixel structures. This study proposes AMFFR-Net, a new deep learning framework that integrates Adaptive Multi-Scale Feature Fusion (AMFF) for multi-scale feature extraction, Graph Attention Networks (GATs) for spatial relationship modeling, and Residual Learning to enhance feature propagation. To progress robustness, the model undergoes extensive preprocessing, including data augmentation, contrast enhancement, and normalization. On the CBIS-DDSM dataset, the proposed AMFFR-Net outperforms other models, including ResNet50, ViT, MIL, and U-Net. Ablation studies establish the important contributions of AMFF, GAT, and Residual Learning in developing classification accuracy and reliability. Based on experimental results, AMFFR-Net displays strong potential as a computer-aided diagnosis technique for breast cancer, representing high effectiveness in differentiating among benign and malignant tumors.