Scholarlink

<p>Skin cancer is among the most widespread types of cancer globally, and detecting it can be difficult, even for experienced dermatologists. Early detection is key to successful treatment, and deep learning methods, especially deep convolutional neural networks (DCNNs), have demonstrated significant potential in this area. However, achieving high accuracy with these models requires large datasets, which individual medical institutions often lack. Sharing medical data directly are also challenging due to privacy and legal concerns. To address this problem, we propose a federated learning approach to develop a privacy-preserving and accurate system for classifying skin cancer, helping dermatologists make better decisions. In this manuscript, we used five Deep CNN models (i.e., Densenet169, VGG16, InceptionV3, Xception, and InceptionResNetV2) to identify key characteristics from skin lesion images. Then a fine-tuning layer is used to refine the entire model and reduce complexity. A fully connected layer combined with a softmax activation function is applied to perform the classification task. Finally, we apply the federated learning approach with the Deep CNN model for the privacy-preserving of the patient information. We evaluate our method using the ISBI2016 dataset containing 1279 skin lesion images. The FL-VGG16 model performed the best among all models, achieving 92.08% accuracy, 76.92% F1-measure, 90.91% precision, and 98.36% specificity. Additionally, we also used the Local Interpretable Model-Agnostic Explanations (LIME) technique to make the predictions easier to understand. We believe our model can help healthcare professionals make accurate predictions about skin cancer and support better treatment decisions.</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

SkinFLNet: A Federated Learning Approach for Skin Cancer Detection Utilizing Skin Dermoscopy Images

  • 期刊论文

  • 发表日期：

    2025年7月10日

查看全文
  • Sajeeb Saha,
  • Shuvo Biswas,
  • Sneha Sarkar,
  • Md. Abdul Hadi,
  • Nilanjana Basak,
  • Mst. Zakia Sultana

摘要

Skin cancer is among the most widespread types of cancer globally, and detecting it can be difficult, even for experienced dermatologists. Early detection is key to successful treatment, and deep learning methods, especially deep convolutional neural networks (DCNNs), have demonstrated significant potential in this area. However, achieving high accuracy with these models requires large datasets, which individual medical institutions often lack. Sharing medical data directly are also challenging due to privacy and legal concerns. To address this problem, we propose a federated learning approach to develop a privacy-preserving and accurate system for classifying skin cancer, helping dermatologists make better decisions. In this manuscript, we used five Deep CNN models (i.e., Densenet169, VGG16, InceptionV3, Xception, and InceptionResNetV2) to identify key characteristics from skin lesion images. Then a fine-tuning layer is used to refine the entire model and reduce complexity. A fully connected layer combined with a softmax activation function is applied to perform the classification task. Finally, we apply the federated learning approach with the Deep CNN model for the privacy-preserving of the patient information. We evaluate our method using the ISBI2016 dataset containing 1279 skin lesion images. The FL-VGG16 model performed the best among all models, achieving 92.08% accuracy, 76.92% F1-measure, 90.91% precision, and 98.36% specificity. Additionally, we also used the Local Interpretable Model-Agnostic Explanations (LIME) technique to make the predictions easier to understand. We believe our model can help healthcare professionals make accurate predictions about skin cancer and support better treatment decisions.