Scholarlink

<p>Melanoma is a highly aggressive form of skin cancer with a significant mortality rate. Over the past two decades, its incidence has been on the rise and is projected to increase further in the future. Early treatment has a higher survival rate. Currently, dermatologists rely on visual inspection followed by biopsy to diagnose melanoma, but there is considerable variability in diagnostic accuracy among different dermatologists. Therefore, developing a convenient, universal, and reliable early diagnostic technology for melanoma is crucial for early diagnosis and treatment.&#xa0;This study aims to develop a deep learning model for identifying melanoma from images captured by mobile phones. To enhance the generalization and robustness of the model, we implemented data augmentation and multiscale and multibranch learning strategies and optimized the model architecture. Finally, we designed a new network architecture: DEFA-Net, and we trained a melanoma recognition model based on this architecture. Experiments show that our proposed model achieves a validation accuracy of 92.5% and a test accuracy of 91.0% on a dataset of melanoma images captured by mobile phones, which is comparable to the diagnostic performance of experienced dermatologists. Based on the results of this study, a smartphone platform for melanoma prediagnosis was developed to assist in the early diagnosis of melanoma. We anticipate that this technology will serve as an effective tool for global skin cancer screening and prevention efforts.</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

Differential Attention Feature Aggregator (DAFE) for Advanced Melanoma Detection

  • 期刊论文

  • 发表日期：

    2026年5月11日

查看全文
  • YuJie Chen,
  • ChunLin Wang,
  • Jianzhong Peng,
  • XianBin Cao,
  • Rong Huang,
  • WenJia Yang,
  • BaoFeng Guo,
  • WuJia Yu

摘要

Melanoma is a highly aggressive form of skin cancer with a significant mortality rate. Over the past two decades, its incidence has been on the rise and is projected to increase further in the future. Early treatment has a higher survival rate. Currently, dermatologists rely on visual inspection followed by biopsy to diagnose melanoma, but there is considerable variability in diagnostic accuracy among different dermatologists. Therefore, developing a convenient, universal, and reliable early diagnostic technology for melanoma is crucial for early diagnosis and treatment. This study aims to develop a deep learning model for identifying melanoma from images captured by mobile phones. To enhance the generalization and robustness of the model, we implemented data augmentation and multiscale and multibranch learning strategies and optimized the model architecture. Finally, we designed a new network architecture: DEFA-Net, and we trained a melanoma recognition model based on this architecture. Experiments show that our proposed model achieves a validation accuracy of 92.5% and a test accuracy of 91.0% on a dataset of melanoma images captured by mobile phones, which is comparable to the diagnostic performance of experienced dermatologists. Based on the results of this study, a smartphone platform for melanoma prediagnosis was developed to assist in the early diagnosis of melanoma. We anticipate that this technology will serve as an effective tool for global skin cancer screening and prevention efforts.