Scholarlink

<p>This study introduces HAMF-Net (Hybrid Attention Multi-scale Fusion Network), a novel deep learning framework for accurate and robust skin cancer detection. HAMF-Net combines multi-scale feature fusion from pre-trained convolutional networks with adaptive attention mechanisms to enhance lesion representation across varied image conditions. The architecture captures both local texture and global contextual patterns while maintaining computational efficiency suitable for teledermatology and mobile screening applications. Evaluations on the pigmented skin lesion dataset from the Federal University of Espírito Santo or PAD-UFES 20 dataset under augmented and non-augmented conditions demonstrate that HAMF-Net achieves 97.85% accuracy, 97.42% precision, 97.38% recall, and 97.41% F1-score, outperforming state-of-the-art models by up to 4%. These results confirm the framework’s effectiveness as a reliable and scalable diagnostic support tool for early skin cancer detection.</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

HAMF-Net: hybrid attention multi-stage fusion network for skin cancer detection

  • 期刊论文

  • 发表日期：

    2025年12月20日

查看全文
  • Ahmad Sanmorino

摘要

This study introduces HAMF-Net (Hybrid Attention Multi-scale Fusion Network), a novel deep learning framework for accurate and robust skin cancer detection. HAMF-Net combines multi-scale feature fusion from pre-trained convolutional networks with adaptive attention mechanisms to enhance lesion representation across varied image conditions. The architecture captures both local texture and global contextual patterns while maintaining computational efficiency suitable for teledermatology and mobile screening applications. Evaluations on the pigmented skin lesion dataset from the Federal University of Espírito Santo or PAD-UFES 20 dataset under augmented and non-augmented conditions demonstrate that HAMF-Net achieves 97.85% accuracy, 97.42% precision, 97.38% recall, and 97.41% F1-score, outperforming state-of-the-art models by up to 4%. These results confirm the framework’s effectiveness as a reliable and scalable diagnostic support tool for early skin cancer detection.